INFORMATION TO USERS
This manuscript has been reproduced from the microfilm master. UMI films the text directly from the original or copy submitted. Thus, some thesis and dissertation copies are in typewriter face, while others may be from any type of computer printer.
The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleedthrough, substandard margins, and improper alignment can adversely affect reproduction.
In the unlikely event that the author did not send UMI a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion.
Oversize materials (e.g., maps, drawings, charts) are reproduced by sectioning the original, beginning at the upper left-hand comer and continuing from left to right in equal sections with small overlaps.
Photographs included in the original manuscript have been reproduced xerographically in this copy. Higher quality 6” x 9” black and white photographic prints are available for any photographs or illustrations appearing in this copy for an additional charge. Contact UMI directly to order.
ProQuest information and Learning 300 North Zeeb Road, Ann Aibor, Ml 48106-1346 USA 800-521-0600 UMI
EVALUATING THE EFFECTS OF CONTEXT AND SCALE ON INDIVDIUAL ACCESSIBILITY: A MULTILEVEL APPROACH
DISSERTATION
Presented in Partial Fulfillment of the Requirements for
the Degree Doctor of Philosophy in the Graduate
School of The Ohio State University
By
Joseph Weber, B.A., M.A.
The Ohio State University
2001
Dissertation Committee: Approved by Professor Mei-Po Kwan, Adviser
Professor Morton O’Kelly Adviser Professor W. Randy Smith Department of Geography UMI Number: 3022600
UMI
UMI Microform 3022600 Copyright 2001 by Bell & Howell Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code.
Bell & Howell Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 ABSTRACT
The issue o f the intraurban accessibiUty o f individuals has long been an important topic, but the relationships between accessibihty and urban form have not been fully
addressed. Conventional proximity-based accessibihty measures and the prevailing monocentric and polycentric models of urban form treat accessibility as a function of distance and do not allow any socioeconomic or behavioral variations in the population to influence accessibility. The use of space-time measures of accessibility overcomes many of these limitations by allowing individual’s daily travel and activity patterns, as well as characteristics such as gender, race, and age, to define their intraurban accessibility. But there is also strong reason to beheve that place-specific characteristics are important to individual accessibility by mediating people’s access to transportation, employment, shopping, and overall knowledge of the city. The importance and role of these influences can be expected to vary by time of day as mobiUty is reduced due to congestion, as well as limited business hours at night. Because of the difficulty of distinguishing contextual effects firom socioeconomic variations with conventional methods, the mediating influence of location among areas and at different scales has been evaluated in Portland,
Oregon using both single level multivariate regression as well as multilevel modeling techniques. This latter method allows the isolation of accessibility variations resulting
11 from variability in the population (compositional effects) from those resulting from differences between areas (contextual effects).
The results of the analysis show that while distance to some urban centers is of importance in explaining variations in individual accessibility, the Portland CBD is only one such center. Household time constraints related to the number of hours worked per week and household size are also important, and provide more consistent explanations for accessibihty than do distance. Time is also important to accessibihty because the effects of reductions in accessibihty due to congestion and limited business hours are not distributed evenly throughout the metropohtan area. These relationships were examined across a range of spatial scales within Portland, but no significant scale variations in accessibihty relationships were found. While the characteristics o f neighborhoods provide some explanation for observed variations, individual and household characteristics again provide more consistent explanations for accessibihty within
Portland. These results are in contrast with common expectations about accessibihty, urban form, and the future land use and transportation plan for Portland, Oregon, leading to the conclusions that many statements about cities and human behavior are of limited usefulness in describing or explaining accessibihty patterns.
Ill DEDICATED TO MY PARENTS
IV ACKNOWLEDGMENTS
I wish to thank my adviser, Mei-Po Kwan, for her encouragement, guidance, and feedback throughout this research. I also wish to thank Tom Kloster and Kyung-Hwa
Kim at the Portland Metropolitan Service District (Metro) for discussing transportation and land use planning issues in Portland, as well as Joan Amfield in the Department of
Sociology at Ohio State University for making available that department’s computer facilities. I am also very grateful to Irene Casas for her generous assistance with
Arc View and other computer problems, as well as her fnendship and support during the long dissertation writing process. Finally, I am grateful for having had the unfailing support, inspiration, and encouragement of many friends, especially Donna Rogers,
Margaret Popovich, Yong-Sook Lee, Bae-Gyoon Park, Geeta Chandra, and Janek
Mandel. VITA
1992 ...... B.A. Geography, University o f Arizona
1994 ...... M.A. Geography, University of Arizona
1995- present ...... Graduate Teaching and Research Associate, The Ohio State University
FIELDS OF STUDY Major Field: Geography
VI TABLE OF CONTENTS
Page
Dedication ...... iv
Acknowledgments ...... v
Vita...... vi
List of Tables...... x
List of Figures...... xi
Chapters:
1. Introduction ...... 1
1.1 Introduction ...... 1 1.2 Research Objectives...... 4 1.3 Organization of this Dissertation ...... 6
2. Accessibility and the Changing American City ...... 12
2.1 Introduction ...... 12 2.2 The Changing American City ...... 14 2.2.1 Changing urban form ...... 17 2.2.2 Human behavior and distance ...... 22 2.2.3 Postmodern urbanism ...... 25 2.3 Measuring Accessibility in North American Cities ...... 28 2.3.1 Space-time accessibility measures ...... 33 2.3.2 Geographical context and space-time accessibility measures 38 2.4 Conclusions ...... 41
3. Data and Methodology ...... 44
3.1 Introduction ...... 44 3.2 Study Area and Data ...... 45 3.2.1 Individual accessibility data ...... 50 vii 3.3 Street Network and Estimation of Travel Times ...... 54 3.4 Space-Time Accessibility Measures ...... 59 3.5 Specifying Accessibility Measures ...... 64 3.6 Conclusions ...... 69
Accessibility in Portland ...... 71
4.1 Introduction ...... 71 4.2 Individual Accessibility in Portland ...... 73 4.3 Accessibility and Distance ...... 78 4.3.1 Accessibility patterns and the monocentric model ...... 80 4.3.2 Accessibility patterns and the polycentric model ...... 88 4.4 Explaining Individual Accessibihty Patterns in Portland ...... 94 4.4.1 Socioeconomic characteristics and accessibility ...... 101 4.5 Discussions and Conclusions ...... 104
Accessibihty and Scale in Portland ...... I ll
5.1 hitroduction ...... I ll 5.2 Individual Accessibihty and Scale ...... 113 5.3 Incorporating Scale and Contextual Influence on Individual Accessibihty ...... 118 5.3.1 Identifying spatial scales ...... 120 5.3.2 Representing neighborhoods ...... 122 5.4 Multilevel Modeling of Individual Accessibihty ...... 125 5.5 Conclusions ...... 131
Accessibihty and Geographical Context in Portland ...... 135
6.1 Introduction ...... 135 6.2 Incorporating Geographical Contexts ...... 137 6.2.1 Accessibihty and geographical context ...... 142 6.3 Contextual characteristics and multilevel modeling ...... 147 6.4 Discussion and Conclusions ...... 153
Conclusions ...... 159
7.1 hitroduction ...... 159 7.2 Individual Accessibihty in the Late Twentieth Century City ...... 161 7.3 Time, Distance, and Accessibihty ...... 165 viii 7.4 Significance of this Research and Future Directions ...... 171
Bibliography...... 176
IX LIST OF TABLES
Table Page
4.1 Accessibility characteristics of sample individuals ...... 74
4.2 Average accessibility under monocentric expectation ...... 82
4.3 Average accessibility under polycentric expectations ...... 92
4.4 Regression results for individual accessibility ...... 97
5.1 Results of multilevel models with individual characteristics ...... 127-128
6.1 Regression results with individual and contextual characteristics ...... 143
6.2 Results of multilevel models with contextual characteristics ...... 148-149
6.3 Regression results with local areas ...... 153
6.4 Multilevel results using local areas ...... 154 LIST OF FIGURES
Figure Page
3.1 Portland, Oregon, study area ...... 46
3.2 Residential location of individuals in sample ...... 52
3.3 Free flow speeds through Portland street network ...... 57
3.4 Peak period speed reductions in Portland street network ...... 61
3.5 Example of network Potential Path Area (PPA) ...... 63
3.6 Weighted opportunity density surface of Portland study area ...... 68
4.1 Weighted opportunity individual accessibility surface for Portland study area.. .78
4.2 Average individual accessibility by distance from the Portland Central Business District (CBD)...... 81
4.3 Average percent reduction in individual accessibility by distance from the Portland Central Business District (CBD) ...... 84
4.4 Major polycentric centers in Portland study area...... 90
4.5 Average individual accessibility by distance from twelve regional centers in the Portland metro area ...... 91
4.6 Average percent reduction in individual accessibility by distance from twelve regional centers in the Portland metro area ...... 93
5.1 Nesting of 21 NBOl neighborhood zones within 12 NB02 zones ...... 124
XI CHAPTER 1
INTRODUCTION
1.1 Introduction
Accessibility has traditionally been conceptualized in geography as the proximity
of one location (whether zone or point) to other specified locations. Since the concept
first became estabhshed in geography, it has been extended in various directions by
including a range of additional information. Mediating factors, such as the effects of
unequal access to transportation systems, division of labor, and household responsibilities have been inserted between people and places, breaking down the autonomy of individuals and their ability to move about as they wish. The measurement of distance has also been expanded by greater reference to actual street networks and travel times.
These expanded concepts and the mediation between people and place now provides the core set of ideas in accessibility.
This concept of individual intraurban accessibility requires certain conditions to be meaningful. There must be a spatial separation of activities (such as home, work, school, or shopping) that must be overcome by individuals in the conduct of their daily fives (by commuting to work, taking children to school, or going on a shopping trip).
1 Individuals must have some means to overcome this spatial separation (by walking, driving, taking the train, etc) but there must also be some cost associated with this movement (such as the time needed for a trip, the cost of riding the bus or paying for parking, and the aggravation of dealing with trafBc congestion). Without such cost, mobility would be unlimited and accessibility would lose all meaning.
But these conditions have not always been found and are far from universal even today. Prior to the 19th century there was little division between home and work within cities, as well as little differentiation of land uses within cities (Vance, 1966; Fishman,
1987). Many work activities were carried out within homes and there was little notion of distinguishing between work activities and domestic activities. Within cities different social groups and classes intermingled in the same areas and worked together. It is only with social and spatial segregation of people and land uses that increased with the development of suburbs that the concept o f intraurban accessibility has its current meaning. Under such conditions people will live apart from the place of their employment and where people live will be closely related to what social group they are a member of, with the wealthy living farther out on the margins of the city. This 19th century patterning of land uses and population is the characteristic notion of the monocentric city that remains in common use (Lloyd, 1981), and though it has increasingly been modified by the presence of substantial employment outside of traditional downtowns, the separation between home and work and consequent need for mobility remains. More recently it has been suggested that the need for activities outside the home is being eroded by new communications technologies, such as the hitemet, so that many activities could increasingly be carried out with greater freedom with respect to
2 their location and timing, and so reduce the spatial separation of home and work of the
modem-era city (as well as the importance of fixed work hours). The specialization of
land uses and the rhythms of daily life is therefore contingent on certain technologies and
urban forms. These conditions roughly limit the concept of intraurban accessibility to the
period following the interior differentiation of cities in the 19th century and before some
(still hypothetical) fixture time when communications technology eliminates the need for
movement or allows for 'virtual' travel with few impediments.
Any discussion of individual accessibihty must therefore take into account the
ways in which the development of cities has structured urban space and mobility patterns.
This is especially so because of the way in which our understanding of cities has been
stmctured by models of urban form, each with their own logic and implications for
individual accessibihty. The reliance on a particular urban model assumes much about
what accessibihty is and how (or if) it can be measured, even if the user does not make
this exphcit. This issue of the conceptuahzation o f urban form is therefore fundamental
to any meaningful discussion of individual accessibihty.
For much of the 20th century the monocentric model has been the prevailing model of urban form, with a centrally located employment location and land use densities declining away from this center. However, in recent decades our notions of cities have been changing, including the allowance for suburban employment centers which compete with the traditional downtown and create a polycentric city. More recently, the emphasis placed on distance as an influence on urban form and human behavior has been questioned, with many suggesting instead that the characteristics of areas may be a considerably more important influence. According to this view, local community ties,
3 the quality of schools and housing, and other local land use attributes can be expected to provide an explanation for residential decision making or commuting that have little to do with distance. Land values and the existence of large office complexes in the suburbs may have more to do with the prestige of the area, its proximity to valued amenities, or to the construction of large projects by single developers using standardized guidelines than they do to distance. These issues strongly suggest that the standard monocentric and polycentric conceptualizations of cities are no longer fully adequate for explaining contemporary cities. Because these urban conceptualizations are open to question, our concepts of individual accessibility can no longer be accepted without further investigation as to their properties and how they relate to urban form and human behavior. It has already been shown that conventional accessibility measures ignore considerable accessibility differences between men and women (Kwan, 1998, 1999a), a finding that the assumptions of the monocentric model would not support. However, the relative importance of distance and geographic context within the city to individual accessibility has not yet been fully investigated.
1.2 Research Objectives
The objective of this research is to identify the extent that existing conceptualizations of accessibility, based on prevailing notions of urban structure and human behavior, remain appropriate in light of the massive changes that cities have undergone in recent decades. Underlying this question is the larger issue of whether access to employment or services is correctly based on proximity, or whether it is more appropriate to view accessibility as based on the interaction of people and the places in
4 which they live, work, shop, and carry out other activities over the course of the day.
This in turns leads to the questions of identifying the best spatial scales and areas within which to measure distance or define local geographic contexts, and how accessibility might vary among these. Further, because prevailing urban models are timeless, the
question of the importance of time to accessibility is also important. Mobility within cities will clearly vary according to time of day, and the ability of people to carry out
activities is dependent on the businesses or facilities being open when individuals have time available.
It can be hypothesized that local geographic contexts (such as land uses o f the surrounding residential area and socioeconomic characteristics of an individual's neighborhood or community) can explain accessibility variations to a greater extent than measuring distance to certain features within cities. This is because the uneven development of cities may influence an individual's behavior or constraints in ways that common urban models cannot reduce to distance firom one or more central locations. An individual's activity patterns (and therefore their accessibility) will likely reflect the local availability of employment or shopping opportunities as well as the local transportation system. While these influences will not likely be random, emerging notions about contemporary urbanism neither require nor can be expected to generate a single explanation for accessibility variations. Instead, it can be expected that there will be a range of relationships between location, contexts, and individuals that will shape accessibility. These relationships will likely vary across spatial scales, from local neighborhoods through political subdivisions to the scale of the entire metropolitan area.
Further, these relationships between individual accessibihty and geographical contexts
5 will likely be modified by the significance of time to mobility and access to services, as well as the need for people to balance conflicting demands of work, household responsibilities, and the need to engage in travel to carry out these demands.
Geographical context, scale, and time are therefore all likely to be necessary to adequately explain accessibility.
1.3 Organization of this Dissertation
The ways in which notions of urban form and human behavior within cities have been conceptualized over time, and how these conceptualizations are related to distinct approaches to evaluating individual accessibility, will be discussed in Chapter Two. The monocentric and related polycentric models of urban form will be reviewed, together with their implications for travel behavior and accessibility. Although these models have been the standard urban models for much of the twentieth century, there is considerable doubt about their continued relevance or explanatory ability. These doubts are particularly strong regarding these models’ poor ability to explain individual travel behavior and household residential mobility within cities. As a result, new conceptualizations of American cities, often labeled as ‘postmodem’, have come into use to better explain the realities of late 20th century and early 21st century urban patterns.
While these statements have been strongly contested, they do offer perspectives that appear to better account for observed travel patterns, and have implications for individual mobility that are strongly at odds with the conclusions reached by the monocentric and polycentric models. Measures o f accessibility are also reviewed in Chapter Two, along with the several ways in which they conceptuahze individual access to employment or services.
Several distinct perspectives exist as to how accessibihty should be evaluated, differing primarily in whether access is based on the proximity of a point (often representing the household residence) to employment or service locations, or by using the daily movements o f individuals in space and time to evaluate the feasibiUty of reaching potential employment or activity locations and carrying out tasks at these locations. The choice of which set of models is most appropriate is not neutral, as each has characteristics that can be associated with the various urban models. Proximity based measures use distance to evaluate access, and require that accessibihty be the same for all individuals living at the same location (or even within the same zone), which is similar to the use of distance in monocentric and polycentric models. In contrast, space-time measures are not based on distance but on the mobility and time constraints of individuals. This allows for access levels to differ greatly for people living at adjacent locations, and accessibihty may be only shghtly, if at all, related to distance from the center of the city or the workplace. Space-time accessibihty measures are therefore potentially quite suited to emerging perspectives on urbanism, while the analytical power they offer is unnecessary within the notions of monocentric or polycentric cities.
The Portland, Oregon, study area and the travel/activity diary data set used in this research will be introduced in Chapter Three. The Portland area offers an ideal opportunity for evaluating the importance of places to accessibility because of several characteristics. The urban area’s growth has been limited by an urban growth boundary,
Portland has an unusually strong CBD area, and future growth plans for the area
7 explicitly stress a polycentric pattern. As a result, the monocentric and polycentric models could be expected to apply to this city to a greater extent than elsewhere. Finding results that do not support distance, and favor the importance of geographic contexts, could therefore strongly undermine the expectations of these models.
Chapter Three will also discuss the formulation and computation of five space time measures of accessibility, each o f which captures a different aspect of individual accessibility. Mobility, the number of potential activity opportunities (representing employment, retail, and other service property locations), and the size of these opportunities are all represented within the five measures of accessibihty. Because space-time measures require highly detailed spatial data sets used to allow the accurate representation of potential activity locations and the modeling of driving times under fi-ee flow and congested conditions within Portland, these will be discussed in detail. In addition, the incorporation of time (in terms of both congestion and the limited availability of potential activities because of business hours) will be discussed.
The resulting accessibihty values for the sample of individuals are analyzed in
Chapter Four with the goal of identifying to what extent spatial variations in accessibilities can be explained by the use of distance in the monocentric and polycentric model. This is carried out by the use of surfaces and multivariate regression modeling.
While the representation of the individual accessibihty data as a surface shows a highly variable pattern with no clear decline by distance fi-om central locations, the use of a subsample shows that people located in suburban areas tend to have higher accessibihty.
This is in contrast with the monocentric model. The effects of congestion on accessibihty are also not distributed evenly, again contradicting the expectations o f this model. The
8 importance of time of day to accessibility is also highlighted, as evening congestion and limited business hours have considerable impact on accessibihty patterns. These impacts are not distributed evenly throughout the city, as would be expected using the monocentric and polycentric models. Instead, it can be expected that there are considerable variations among areas within the city and among different individuals.
This unevermess provides an argument in favor of postmodern urbanism.
However, regression modeling does show that distance to some urban centers is of some importance in explaining variations in individual accessibility. Despite this, the
Portland CBD is only one such center, with two other major suburban employment and retail centers showing up as more important with several accessibility measures. While these results do support the expectations of the polycentric model, as accessibihty does dechne with distance from several centers, the evidence also casts doubt on the importance of distance. Additionally, household time constraints related to the number of hours worked per week and household size are important, and these time constraints provide more consistent explanations for accessibihty than do distance.
Because the results of Chapter Four suggest that distance is only of limited utility in explaining accessibility at the metropohtan scale, additional testing of these results is carried out in Chapter Five using multilevel modeling. This methodology extends multivariate regression by grouping individual level data by zones. Using this approach accessibility variations were examined at a range of spatial scales within Portland, making use of neighborhoods, school district boundaries, city boundaries, and commutersheds around polycentric centers. The results show that the single level regression analysis carried out in Chapter Four is adequate for modeling accessibihty in
9 Portland, as there are no significant scale variations in accessibility relationships.
Instead, space-time measures of accessibility appear to be ‘firameless’ in that results do
not vary across spatial scales and are not likely to coincide easily with common spatial
firameworks such as political boundaries or neighborhoods.
The characteristics of neighborhoods were incorporated into the multilevel
modeling process in Chapter Six in order to identify whether the contexts in which
individuals live, work, shop, and carry out other activities can provide a better level of
explanation than does distance from the CBD (or other centers) or even individual level
attributes. The results do not support these conclusions, though they do show local
contexts do provide some explanation for accessibility variations. Instead, as in Chapter
Four and Five, individual and household characteristics provide more consistent
explanations for variations in accessibility within Portland. These explanations, and the
limited role of geographic context, are often at variance from common expectations about
accessibility as well as the future land use and transportation plan for Portland, Oregon.
Finally, Chapter Seven will siunmarize the results of this research and relate them
to competing conceptualizations of urban form. While there is a lack o f support for
standard urban models, there is also little evidence to directly support competing
conceptualizations of cities. The geography of individual accessibility within cities
cannot therefore be read off from the spatial form of the city or even from the way
mobility has been structured by the construction of transportation systems. The potential
impact of new spatial technologies, such as the Internet, hitelligent Transportation
Systems, and telecommuting, on the research results are also discussed. While most discussion of these technologies focuses on their potential geographic impact, the results
10 from this research instead suggest that their impacts on accessibility will be greatest to the extent that they alter individual or household time constraints, and thereby give people both more time to engage in flexible activities as well as greater flexibility in the time of day they do so. Finally, the significance of this dissertation and important directions for fixture research will be discussed.
11 CHAPTER 2
ACCESSIBILITY AND THE CHANGING AMERICAN CITY
2.1 Introduction
This chapter will discuss recent changes in American cities and their implications for the measurement of individual accessibility. The first part of the chapter will examine the prevailing monocentric and polycentric models of urban form and human behavior within cities and how they conceptualize individual accessibility. Next, a variety of evidence will be discussed that strongly suggests that the importance given to distance in these models is no longer appropriate. Not only are these models of limited use for adequately explaining a variety of processes that shape cities, but they also misrepresent individuals and households and their travel and activity behavior. This evidence strongly suggests that many aspects of land use and travel behavior are more likely to be influenced by their location within the city, or characteristics o f these areas, than by distance from one or more central points. The ability o f these models to help understand individual accessibility is therefore questionable.
12 As a result, competing perspectives on contemporary cities have been developed.
While the originahty and theoretical vaUdity of these perspectives have been hotly contested, they focus attention on the social processes and relations that are associated with the diversity present within real cities and therefore break away from the static equilibrium concepts of the monocentric and polycentric models. The importance of political jurisdictions or desirability of areas to land uses (and resulting inequalities in the location of employment or services) is emphasized, and rather than ignoring interpersonal differences, these variations are highlighted. Individual experience of the city is central to understanding accessibility, and must take into account the geographic contexts in which individuals live and move around in. For these reasons they appear to form a more useful foundation for conceptualizing individual accessibility within cities.
The second part of the chapter will discuss accessibility measures and their conceptual relationships to urban models. Conventional aggregate proximity-based measures are conceptually similar to the monocentric and polycentric models in their treatment of accessibility and inability to incorporate time or interpersonal differences.
For this reason they appear to have limited use for understanding individual accessibility within contemporary cities. In contrast, space-time measures derived from time geography are ideally suited to represent individual accessibility by making use of people’s daily activities and movements through space and time. While these measures incorporate the importance of geography they do not directly allow its’ relative importance to be assessed. For this reason, multilevel modeling, a statistical analysis
13 procedure that incorporates both areal and individual data, will be discussed as a tool with which to better understand the relationship between individual behavior and the urban environment within individual accessibility.
2.2 The Changing American City
For much of the 20* century most conceptualizations and formal models of urban structure have been based on the idea of monocentric cities. The formal statement of monocentric models is usually credited to work conducted by the Chicago School of urban ecology (Burgess, 1925) and similar research (Hoyt, 1939; Harris and Ullman,
1945) as well as by urban economists (Alonso, 1964 and Muth, 1969), but the monocentric model is often treated as a single composite model. Both models recognize the existence of a single high-density central business district (CBD) that exists at or near the center of the city, and is the location of virtually all employment and retail activity within the city. Residential areas surround the CBD and are differentiated by distance from the center of the city, with housing density decreasing with increasing distance from the CBD. Within the urban ecology model this differentiation is the result of invasion and succession by immigrant groups who originally settled in high density, low income residential areas adjacent to the CBD. As these groups rose in status and income they would be expected to seek more desirable (and lower density) housing, requiring a move outward. Significantly, this process of invasion and succession resulted in cities being seen as dynamic entities, with households continuously shifting in location and socio economic status. The later economic notions of monocentric form also explained decreasing residential density by reference to preferences for low density housing, but in
14 this case assumed an equilibrium condition in which all households (and land uses) sorted themselves out within the city according to their willingness to pay the cost of land.
Because of the importance of centrality within the city and the high cost of transportation, land prices decreased away from the CBD. While housing costs decrease away from the center, commuting costs increase, and individual households must make a decision trading one for the other.
In both versions the outcome is essentially the same. The CBD contains the majority o f the city’s employment and services and is the most accessible and desirable location in the city as well as the center of cultural activity and political power.
Commercial, industrial land and poor neighborhoods (occupied by minorities and recent immigrants) surround the CBD with rings of decreasing density residential land
(occupied by progressively higher income residents) found farther out. Commuting is reduced to movement from the peripheral home location to the CBD, where all employment is located (Hanson and Pratt, 1988; Holzer, 1991). Most importantly, within the economic version, distance from the center o f the city is given considerable causal power to determine land uses, household incomes and housing preferences. Should transport costs to the center become higher, residents and land uses will respond by moving closer to the center. If transport costs are reduced by new technology the tradeoff between housing costs and commuting costs will be altered and households will move farther outwards. Because all households and individuals living at a particular location are treated as homogenous, all will make the same trade-off decisions.
15 Accessibility in a monocentric city is simply centrality within the city. The closer one lives to the CBD, the higher one's accessibility, and vice versa. Accessibility is therefore an attribute o f places (defined merely as distance firom the CBD), not people.
Significantly, because income increases with distance from the center the poor are supposed to possess the greatest accessibility (McLafferty, 1982). This is an important point because it implies that monocentric cities are socially equitable (everyone is able to select a residential location that provides them with an identical level of utility regardless of their income). Although variations in the monocentric model allow for directional growth variations or the possibility o f multiple nuclei, these require no significant alterations to the logic of the monocentric model (Hoyt, 1939; Harris and Ullman, 1945;
Adams, 1970; Muth, 1985; Krakover and Casetti, 1988; Giuliano, 1989). The monocentric model continues to influence the description o f cities and urban patterns
(Davis, 1998; Marshall, 2000), though it is often expressed as a city/suburb dichotomy, in which the entire city is reduced to a central city containing all employment and retail activity and low income (and often non-white) population, while the suburbs consist of homogenous and prosperous residential areas (Lloyd, 1981; as can be seen in Gober and
Behr, 1982; Bookout, 1992; OhUllachain and Reid, 1992; and Newsome, Walcott and
Smith, 1998). Monocentric notions are also present in the idea of Neo-Traditional
Development, which attempts to design residential areas using grid street patterns with shopping opportunities centrally located within the neighborhood (Bookout, 1992;
Handy, 1992; Boamet and Crane, 2001). The idea that homogenous households will act to minimize distance traveled to carry out activities is deeply rooted in this concept, as is a nostalgia for the centralized business districts found in older cities.
16 2.2.1 Changing urban form
Despite the widespread use of the monocentric model there has been a growing dissatisfaction with using it as an explanation of contemporary American cities since the late 1960s. This is largely the result of the belated observation of new urban forms that cannot be easily made to fit within the monocentric model. Foremost among these have been large and highly visible employment and retail centers in suburban areas, which clearly violate monocentric assumptions (Kersten and Ross, 1968; Baerwald, 1978, 1989;
Erickson and Gentry, 1985; Hartshorn and Muller, 1989; Garreau, 1991; Knox, 1991).
The assumptions of the monocentric model have also been strongly challenged by the decline of traditional monocentric commuting patterns, as the majority of commuting trips now take place between suburban locations (Plane, 1986; Pisarski, 1996). This further suggests that the CBD is declining in importance as an employment center and focus of urban life.
A popular method for handling these issues has been to extend the logic of the monocentric model to allow for the existence of multiple centers. Rather than ignoring the suburban 'edge cities', these centers are instead directly incorporated into the model and treated as similar to the original CBD, producing the polycentric model. Within this model these suburban centers are assumed to be surrounded by their own commutershed o f workers who commute into this center firom outlying residential areas. Each center and its surrounding area therefore function as separate monocentric cities or urban realms
(Vance, 1964,1990, 1991; Hartshom and Muller, 1989, 1992; Muller, 1989, 1995). The
17 development of polycentric cities is seen as simply being the latest stage in the evolution of urban form, but one in which distance continues to be the dominant organizing principle.
Because polycentric cities are an extension of the assumptions imderlying the monocentric model the same general principles can be applied to explain the location of firms and residential choice (Greene, 1980; Baerwald, 1982; Erickson, 1986; Giuliano,
1989), while declining densities and land values can also be identified around polycentric centers, usually producing better results than using the CBD (Gordon, Richardson, and
Wong, 1986; Heikkila, et al, 1989; McDonald and Prather, 1994; Small and Song, 1994).
As the new urban realms will most likely exist in areas classified as suburban, commuting inwards to the center of these realms will be classified as within-suburb commuting, easily explaining the growth of this trend in recent decades. The polycentric conceptualization continues to make use of the same notions of accessibility, though now it reflects distance to the center of one's urban realm, and not necessarily the traditional
CBD. Those living nearer to a center are assumed to have higher accessibilities, but because polycentric cities allow individuals greater opportunity to find a residence close to an employment center they can actually be considered to be more efficient or equitable than monocentric cities (Gordon and Wong, 1985; Gordon, Richardson, and Wong, 1986;
Gordon, Kumar, and Richardson, 1989a, 1989b).
Even with the increasing recognition that the monocentric model is obsolete (as discussed in detail in a special issue of Geographical Analysis edited by Berry and Kim,
1993), there is growing evidence that the addition of multiple employment centers to the model is itself inadequate because the model continues to make use of monocentric logic.
18 The use of this underlying framework has been challenged by a variety of evidence. The most visible concern is the difficulty of fitting actual cities into the polycentric firamework, as suburban employment centers are neither fimctionally equivalent nor independent of one another (Giuliano and Small, 1991). Despite the considerable amount of attention devoted to identifying polycentric centers there is still no consistent or universal methodology for doing so, and considerable disagreement about exactly what such a center is (Gordon, Richardson, and Wong, 1986; McDonald, 1987; Heikkila, et al,
1989; McDonald and McMillen, 1990; Pivo, 1990, 1993; Giuliano and Small, 1991;
Waddell and Shukla, 1993; McDonald and Prather, 1994; Small and Song, 1994; Forstall and Greene, 1997; Bogart and Ferry, 1999). This issue, along with the growing dispersion of employment, has led several researchers to question the existence of discrete polycentric centers, suggesting instead that cities increasingly contain more dispersed employment than is generally recognized (Waddell and Shukla, 1993; Gordon and Richardson, 1996). While edge cities may be highly visible urban features and attract considerable numbers of commuters or shoppers, they are not necessarily the dominant features around which the urban environment is structured.
More problematic for the monocentric model is the apparent weakening in the power of distance to organize land uses. Among studies that related distance to housing value, apartment rents, or the location and size of office employment, the distance to polycentric centers generally provided better explanations of spatial variations than the distance to the CBD (Hoch and Waddell, 1993; Waddell, Berry, and Hoch, 1993).
However, many have found that distance to other (non-workplace) features (such as universities, airports, shopping malls, major streets, freeway interchanges, or the ocean)
19 often work as well or better than distance to polycentric employment centers (Erickson and Gentry, 1985; Heikkila, et al, 1989; Hoch and Waddell, 1993; Waddell, Berry, and
Hoch, 1993). While declines in land values away from suburban freeway interchanges and the ocean suggest that distance remains of some importance, they do not directly support polycentric expectations, and in one case distance to any of these features was completely unable to predict the size and location of office employment concentrations
(Archer and Smith, 1993).
Further, even when distance is important, the significance of proximity to streets, shopping centers, freeways, airports, and employment centers to rents or housing values is often not linear (Hoch and Waddell, 1993; Waddell, Berry, and Hoch, 1993). Instead, an 'inverse U' relationship may be seen, with rents/values peaking at several miles distance from the feature and being lower at both nearer and farther distances. This may be explained by the presence of disamenities such as noise and congestion at closer distances and lack of conveniences farther out, suggesting that it is not distance that is important but relative location. The quality of housing and local land use characteristics provide additional (and potentially similar levels of) explanation, indicating that the ways in which distance interacts with housing and neighborhood characteristics is an important issue for study (Giuliano, 1989; Hoch and Waddell, 1993; Waddell, Berry, and Hoch,
1993).
This decline in the organization of cities by distance from the center has been attributed to a variety of forces. The construction of freeway networks has increased mobility and so eliminated any locational advantages that the cost of movement traditionally provided to central locations (Giuliano, 1989). The influence of distance has
20 therefore been reduced by the creation of a more uniform and efficient transport system.
However, this view needs to be modified by the existence of strong evidence that the
development of urban transportation (especially that of public transit) has been strongly
influenced by place-based competition between competing interests, particularly between
those favoring the traditional CBD and those favoring suburban locations (Adler, 1986,
1987, 1988, 1991; Hodge, 1990, 1995; Ruthheiser, 1996). Because "transport facilities
inevitably create location advantages for some places and relatively disadvantage others"
(Adler, 1986, 321) these facihties will be fought over by different place-based interests
interested in maintaining investment in their area, and "since place advantage is the key
objective, facility location and related choices- rather than conflicts between modes of
transport- are the fundamental dimensions of metropolitan transport poUtics. The
primary activists in transport questions, therefore, are those rooted in competing places"
(Adler, 1986, 322). Transportation systems cannot be expected to have even distributions
or even quality throughout urban areas, and these will obviously affect accessibility. So
even while transport improvements may reduce the importance of distance, they will still provide advantages to certain locations within the city. The reduction of the influence of
distance does not therefore mean that the city has become more homogenous, in fact it
likely indicates that the development within the city has become more heterogeneous and polarized between various locations.
21 2.2.2 Human behavior and distance
There is also a range of evidence that the monocentric model, and the emphasis given to distance, is of limited use for describing human behavior within cities. In large part this results from the fact that the monocentric model reduces households and individuals to their home location, and represents homes and workplaces as isolated points in urban space. The relationship between these points is then limited to the notion that the workplace will determine the location of the home as a result of a distance minimization behavior by households. Should the workplace change (or commuting times increase), households will rationally relocate to preserve their original commuting times (and their original distance/housing costs tradeoff).
However, this is not supported by actual travel behavior (Giuliano, 1989). Actual commuting distances strongly suggest that very few people are acting to minimize their journey to work by relocating either their home or workplace (Hamilton, 1982, Small and
Song, 1992), though it has also been argued that these observations are actually evidence in support of polycentric cities (Gordon, Richardson, and Jun, 1991, Song, 1992;
Levinson and Kumar, 1994; Levinson, 1998). According to this argument, the fact that commuting travel times have remained constant within American cities, despite increases in highway congestion, indicates that households must be adjusting their home and/or job locations to compensate for the increasing travel times they would have otherwise incurred. This supports the continued existence of the distance/housing cost tradeoff (and adjustments to this optimum solution) within non-monocentric cities, though it is not known to what extent commuting has actually been transformed into polycentric commutersheds. However, this research has made use of highly aggregate data, and has
22 not been supported by other work that suggests that households are faced with a wide range o f factors when contemplating relocation decisions (Clark and Burt, 1980;
Giuliano, 1989,1991, 1995; Giuliano and Small, 1993; Wachs, et al, 1993; Lowe, 1998).
Distance is only one o f many such factors, and there is very little direct evidence that households will move with the goal of reducing commuting distance.
The minor influence of distance on relocation decisions can be largely explained by the fact that homes and workplaces are not simply points in space, but instead are tightly connected within a web of economic and social interrelationships (Hanson and
Pratt, 1988; 1992; Giuhano, 1989; England, 1991; Waddell, 1993; Hanson, Kominiak and
Carlin, 1997). As a result, residential (and workplace) mobihty is far from easy or automatic, and since these relationships work in both directions, the monocentric model's emphasis on workplace location is incomplete. Relocating a home or job will likely incur significant social and economic costs, such as the difficulty in selling a home and/or obtaining new housing, losing contact with friends and neighbors, or switching schools, potentially resulting in a high degree of inertia or 'spatial fixity (England, 1993). Home locations for many individuals may therefore essentially be fixed, regardless of the presence of suitable nearby employment (especially for women) or changes in workplace location (Hanson and Pratt, 1988, 1992). As a result, "at any given time, a large number of (rational) household and employment locations may in fact be 'suboptimal' with respect to transport cost" (Giuhano, 1989, 152), violating the assumptions of the monocentric and polycentric models. This spatial fixity and a lack of mobihty can be
23 expected to be important to accessibility, not just because of local variations in mobility or opportunities, but because of the way they influence individuals' knowledge of cities, employment prospects, and attitudes.
The representation of households and individuals by the monocentric model is also problematic because it hides important differences that exist between individuals within households. Rather than being homogenous, individuals and households will differ considerably in their needs, preferences, activities, perceptions of the urban environment, and ability to process spatial information. Perhaps the most well known example of this are gender differences between men and women, which are hidden by the monocentric model in its assumption of one worker per household. This assumption has been criticized for its impUed sexism that "the urban population is made up of essentially one household type, consisting of one (male) worker and one (female) full-time homemaker" (Hanson and Pratt, 1988, 302), and can no longer be maintained due to the entry of increasingly large number of women into the workforce in the past several decades.
The invisibihty of gender masks important differences among the urban population on such activities as employment, access to transport, commuting, knowledge of employment possibihties, labor market segmentation, spatial mismatch, household activities, and accessibility (Hanson and Pratt, 1988, 1992; McLafferty and Preston,
1992; England, 1993, 1995; Blumen, 1994; Hanson, Kominiak, and Carlin, 1997; Wyly,
1998; Kwan, 1999b). This 'invisible' characteristic is fundamental to many aspects of spatial behavior, and cannot be ignored in any discussion of intraurban accessibiUty.
Other individual characteristics, such as age, physical disability, income, race, or sexual
24 orientation, are also likely to be of considerable importance to spatial behavior (for example, McLafferty and Preston, 1992; Tacken, 1998). The distribution of populations possessing particular characteristics will vary within the city and is likely to persist over long periods of time despite the economic restructuring of cities (Wyly, 1999). These effects are unhkely to be additive, but bound up with each other and gender in complex ways, and cannot easily be isolated (McLafferty and Preston, 1992). The inability to incorporate these characteristics and interrelationships is clearly a serious weakness of the monocentric and polycentric models.
2.2.3 Postmodern urbanism
The increasing recognition of the limitations of the monocentric and polycentric models in describing contemporary urban land use patterns, household structure, and human behavior, has led to recent interest in developing alternatives to these conceptualizations (Waddell and Shukla, 1993). Rather than revising these models some feel that the urban changes observed in recent decades represents an entirely new pattern of urbanism that require new perspectives. Further, universal explanations offered by earlier urban perspectives are not in line with postmodernist theories or thinking (Dear and Flusty, 1997,1998, 1999; Hall, 1998; Dear, 2000). While the extent to which contemporary cities (or the theories used to understand them) are truly postmodern has been hotly contested (Lake, 1999; Beauregard, 1999; Jackson, 1999; Sui, 1999; Wyly,
1999), recent theoretical perspectives offer a stronger basis for conceptualizing and evaluating individual accessibility.
25 In place of the overriding dominance of distance, as in the monocentric model, this view holds that contemporary cities should be conceptualized as having instead been shaped by the various locational logics of industries, firms, and individuals (Fishman,
1990, Castells, 1989). Land uses in a contemporary city are therefore a “noncontigous collage of parcelized, consumption-oriented landscapes devoid of conventional centers"
(Dear and Flusty, 1998, 66). This pattern includes edge cities, downtown redevelopment projects, shopping centers, and large residential subdivisions (Dear and Flusty, 1998), which others had included in spatial models of cities over 50 years earlier (Harris and
Ullman, 1945). But while these forms may approximate the characteristics of monocentric cities, they need not retain their characteristics, and there would be little expectation of finding distinct commutersheds around edge cities or high residential mobility in a city fragmented among highly stratified residential subdivisions and competing political jurisdictions offering the best services. Location within particular jurisdictions or areas is therefore likely to be very important in influencing the proximity and kind of activities as well as the characteristics and behavior of individuals and households. The removal of distance as a determinant of land use and human behavior opens up new possibilities for the study of accessibility. Rather than the equilibrium and equality of the monocentric model, postmodern urbanism draws attention to inequality and divergences, including the polarization (both social and spatial) between those with and without access to adequate jobs, services, and information.
Another important implication of this perspective is the need to incorporate new scales of analysis. In one direction growing economic ties to a larger economic system and new communication systems have broken down the idea of independent, self
26 contained cities (Knox, 1997; Muller, 1997; Dear and Flusty, 1998). An individual's
access to the larger world as well as their ability to substitute communications for actual travel must be taken into account. In the other direction postmodern urbanism also questions whether it is even appropriate to discuss the idea o f a single city shared by all inhabitants, as the monocentric and polycentric cities would assume. The monocentric model includes the assumption that there exists a 'normal' commute inward to the CBD from homogenous suburbs, which guarantees that every (male) worker will have essentially the same experience of the city. The variety of individuals and households that actually exist within cities, as well as the presence o f a wide variety of daily activity patterns, instead suggest that each individual may have a very different experience than other individuals, even those living nearby. These experiences of the city will not be random, as they will be structured by the presence of localized labor markets or community networks (Hanson and Pratt, 1992; England, 1995). Neither can these individual experiences of the city simply be reduced to polycentric urban realms, as there is no evidence that work or shopping trips are actually limited to an individual realm
(Fujii and Hartshom, 1995; Pickus and Gober, 1988). It can perhaps best be said that
"famihes create their own 'cities' out of the destinations they can reach (usually travelling by car) in a reasonable length of time....The pattern formed by these destinations represents 'the city" for that particular family or individual. The more varied one's destinations, the richer and more diverse is one's personal 'city"' (Fishman, 1990, 38).
Instead of a single shared city, or even a set of shared urban realms, there is instead an overlapping collection of individual 'personal cities'. This has been supported by empirical evidence that shows that experience of the city gained through commuting may
27 differ according to an individual's age, race, income, residential and workplace locations, the size of their household, and the time of day they engage in certain activities
(Newsome, Walcott, and Smith, 1998; Kwan, 1999b). Location within the urban environment, not only of the home and workplace but also the activity space of individuals, is an important component of this individual experience of the city
(Newsome, Walcott, and Smith, 1998). This is clearly a different perspective than that of the monocentric and polycentric models, and presents a considerable challenge to the ability of urban models to account for accessibility patterns. The ability of accessibility measures to adapt and incorporate these dynamic individual personal experiences of the city will be discussed in the next section.
2.3 Measuring accessibility in North American cities
Although a variety of classifications of accessibility measures exist (Morris et al. 1979; Pirie 1979; Handy and Niemeier 1997), these measures can be grouped into conventional aggregate or zonal measures and space-time accessibility measures (derived firom time geographic concepts), due to the fimdamentally different ways these measures represent accessibility (Kwan 1998, 1999a). Further, these conventional aggregate measures produce very similar pattems, allowing them to be grouped together (Kwan,
1998,1999a). These measures generally evaluate the proximity a point or zone (or a set of points or zones) to a set of opportunities (likely represented by other points or zones), using some measure of distance to limit or reduce the importance of opportunities
28 depending on their distance from the origin locations. As all of these measures treat accessibility as equivalent to proximity, they can be distinguished on the basis of how they measure distance.
The simplest approach treats accessibility as the sum o f the straight-line (or
Euclidean) distances from one origin location to all possible destination locations, with a shorter total distance indicating greater accessibility (Ingram, 1971). Measuring these distances through a transport network produces topological accessibility measures that have been (and remain) quite popular at the interurban level for the analysis of transportation network accessibility (Garrison, 1960; Gauthier, 1968; Marchand, 1973,
Murayama, 1994; and Spence & Liimeker, 1994). However, they have only rarely been applied at the intraurban level (Muraco, 1972; Lee and Lee, 1998), perhaps because of the difficulty of translating a dense street network into graph form before the widespread availability of digital networks within GIS. More sophisticated cumulative-opportunity measures sum the number of opportunities (such as employment locations) within certain distances or travel times from an origin, so that the more opportunities that can be reached in less time the greater the accessibility of a location will be (Wachs and
Kumagai, 1973). The most sophisticated of these conventional measures, commonly known as gravity or potential measures, were actually the earliest to be developed, with a lineage traceable back to the 1940s (Stewart, 1942; Carrothers, 1956). These evaluate the accessibility of a location using the distance from that origin location to a set of destinations along with the importance of potential opportunities at the destination
29 locations. Though originally used at the national level these measures have been commonly used in studies of intraurban accessibility (as represented by Knox, 1978,
1982; Guy, 1983; Geertman and Van Eck, 1995).
Though videly implemented, the potential limitations of these conventional accessibility measures have been increasingly recognized. One fundamental weakness is the issue of specifying the influence of distance in these measures. A number of choices are possible, such as whether to use linear distance, travel times, or travel costs as a measure of distance and whether to measure this distance as a straight line or through a transport network. While the representation of distance within Euclidean distance and topological measures is straightforward it is also not very accurate for most intraurban applications in which movement is confined to street networks. Further, the choice of time or distance increments for cumulative opportunity measures requires considerable judgement based on the particular application. The representation of distance within gravity models has received the most attention, with the choice of negative exponential or inverse power functions and the issue of how to handle self-potential (the accessibility of a zone to itself) receiving considerable attention. Each of these has received a range of treatments, meaning that there has been little standardization in the use of this measure
(Pooler, 1987; Geertman and Van Eck, 1995), and it has been demonstrated that gravity and cumulative opportunity measures tend to give similar results regardless of how distance is specified (Kwan, 1998).
The limitations of these measures are compounded by the way they treat those origin locations for which accessibility is being measured. Conventional accessibility measures all require the definition of point or zonal locations (typically the home or
30 workplace) from which proximity to opportunities is to be assessed. With most of these measures accessibility is therefore a property of zones or points, not households or individuals, (though with topological measures accessibility would actually be a property of transport network nodes). Households and individuals possess accessibility only by virtue of living at a particular location, which creates several problems for the use of these measures when discussing individual accessibility. First, measuring access to employment or services from these locations assumes that they are the center of an individual's daily activities and the origin of each individuals' daily travel. This is not a valid assumption as it denies the existence of considerable amounts of multistop trips over the course of the day, and of course a person may spend considerable time away from home during the course of the day (Kwan, 1998,1999a). By ignoring the ways that individuals combine various activities and destinations into a single trip these conventional accessibility measures may be underestimating an individual's accessibility.
The proximity of opportunities to one another, or to predetermined trips such as the daily commute, may be more useful than accessibility to a fixed home location.
Conventional accessibility measures also ignore differences among households and individuals by requiring that everyone living in the same zone or at the same point must necessarily have the same accessibility, and are affected in the same way by changes in accessibility (Kwan, 1998). Even when accessibility inequahties are found within a city using conventional accessibility measures (as by Wachs and Kumagai,
1973; Black and Conroy, 1977; Knox, 1978, 1982, Talen, 1997), it is not clear for whom these inequahties may be a problem or how they are distributed among the population.
Not all individuals represented by a point or zone will share the same characteristics,
31 activities, constraints, or preferences, raising the question of for whom such access is important or relevant. The role of behavior, age, income, access to different modes of transportation, gender, or other differences among individual is therefore rendered invisible by the use of conventional accessibility measures. This issue has been shown clearly with the issue of gender. While gender has been shown to be important to accessibility because of the mobility constraints and extra daily responsibility faced by women (Blumen, 1994; England, 1993; Hanson and Pratt, 1988), this characteristic cannot be handled by conventional accessibility measures. It is not just location but the daily activities of women that in part determines their accessibility, and traditional measures cannot handle this information. This has been shown clearly by the incorporation of gender within accessibility measures that are more sensitive to individual activity behavior (Kwan, 1998, 1999a).
Conventional accessibility measures also ignore time. This refers not only to the amount o f time available to individuals to carry out travel and activities but also how these activities are scheduled throughout the day. Given that men and women tend to have different time constraints on their activities, as well as a different temporal scheduling of activities throughout the course of the day, the absence of time from conventional measures ignores an important source of accessibility variation (Kwan,
1998, 1999a, 1999b). These measures also ignore the importance of time to mobility due to traffic congestion or changes in transit schedules at different hours of the day. The varying availability of opportunities through the day, as with business hours, are also not represented, potentially overestimating mobility while ignoring that many opportunities will not exist during the night. Conventional accessibility measures therefore take a
32 static, timeless view of mobility and accessibility, which denies the ways in which behavior, activity pattems, and even population composition varies by time of day
(Goodchild and Janelle, 1984). While these measures could be modified to include measures of distance and opportunity sets that are appropriate for specific times of the day, these changes would still encounter the limitation of conventional measures in treating all individuals as identical. Updated travel speeds or opportunity sets will be apphed to all individuals equally, regardless of their circumstances.
As can be seen, despite the challenges to the monocentric model, its assumptions about cities and people remain imphcit within conventional proximity-based accessibility measures. Distance is the fundamental concept of these measures, as it is in the monocentric and polycentric models, accessibility is a property of places, and differences among people are ignored. In fact, with a fairly even distribution of points these measures are actually surrogates for centrality within a city. What is needed is an accessibility measure that makes use of individuals, for which many are increasingly turning to space-time measures.
2.3.1 Space-time accessibility measures
Space-time measures are originally derived from Hagerstrand's (1970) time geography concepts, which provide a framework for analyzing individual movement through space and time within the limits of constraints operating on him or her. This time geographic framework has found wide applications in geography, including the study of accessibility (Bums, 1979; Villoria, 1989; Kwan, 1998; 1999a, Kwan and Hong,
1998; Huisman and Forer, 1998b; Forer and Huisman, 1998; Recker, Chen, and McNally,
33 2001), travel behavior (Newsome, Walcott, and Smith, 1998), epidemiology
(Schaerstrom, 1996; Loytonen, 1998), and studies of the homeless (Wolch, Rahimian, and Koegel, 1993; RoHinson, 1998).
The initial foundation for time geography is that "in time-space the individual describes a path, starting at the point of birth and ending at the point of death.... the concept of a life path (or parts of it such as the day path, week path, etc.) can easily be shown graphically if we agree to collapse three-dimensional space into a two- dimensional plain or even a one-dimensional island, and use perpendicular direction to represent time" (Hagerstrand, 1970, 10). Movement through space-time will not be random or unconstrained, as Hagerstrand identified three types of constraints on individual mobility: those of capability, coupling, and authority. Capability constraints are based on the physical limits of mobility and biology, as because people must spend a certain amount of time every day eating and sleeping this time is not available for movement. The technology available for movement is also a constraint, as an individual cannot move at a faster speed or for a longer distance than the means of travel will allow.
Even with a high degree of mobility an individual will still be subject to coupling constraints. These "define where, when, and for how long, the individual has to join other individuals, tools, and materials in order to produce, consume, and transact"
(Hagerstrand, 1970, 14). More specifically, these show the requirement that individuals must be in certain places for certain lengths of time for work or must arrive at or depart businesses by certain times. These times and places constitute 'fixed' activities that cannot be ignored or rescheduled and so provide the framework of an individual's daily activities. Activities that allow for more flexible scheduling, duration, or location must
34 be fit into the time available between successive fixed activities, so that even where
capability constraints allow great potential freedom of movement coupling constraints
can be a severe restriction on an individual's activity schedule. These constraints exist not just for those with jobs but even for the unemployed or homeless (Wolch, Rahimian,
and Koegel, 1993).
Further constraints on movement exist. Individuals are also subject to authority constraints, which influence the ways that certain areas may be effectively off limits to certain individuals at certain times. Such constraints could include the fact that most activities (such as businesses or government offices) are not available twenty four hours a day, as they are closed at night. While an individual may have the mobility and time to reach such places, they will not have the right to enter the premises or carry out their activities. Some have also suggested a fourth category, that of social relational constraints, which would encompass constraints on behavior resulting from cultural norms, such as limits on women's mobility (Wolch, Rahimian, and Koegel, 1993).
The mobility allowed by these constraints is shown by an individual's time-space prism, though because of the difficulty of working with three dimensional prims these have commonly been simplified into a two dimensional projection, known as a Potential
Path Area, or PPA (Leimtorp, 1976, 1978). These potential path areas do not include time as a third dimension but display the geographic area that can be reached within the time constraints established by an individual's fixed activities and their mobility. The
PPA is simply the area that an individual can physically reach after one fixed activity ends while still arriving in time for the next fixed activity, and therefore results from the individual's own activity schedule and capability, coupling, and authority constraints.
35 Each individual will create their own PPAs through the course of daily activities, resulting in what Fishman (1990) has described as their own 'personal city*. When aggregated these individual daily paths or 'personal city* would make up a space-time aquarium of overlapping personal cities, perhaps the best realization in geography of the dynamism inherent in postmodern urbanism.
While early accessibility applications of time geography evaluated the feasibility of particular activities by showing whether they were possible within the constraints facing individuals (Pred and Palm, 1978; Lenntorp, 1978; Forer and Kivell, 1981; and
Miller, 1982), most accessibiUty work using time geographic concept has focused on the specification of space-time accessibility measures based on individuals' PPAs (Bums,
1979; Villoria, 1989). The PPA has natural appUcations for the study of accessibility, as not only does it show the area an individual can reach, it also allows an individual's activity pattems to define the space relevant to them, eliminating problems of defining meaningful distances common to proximity-based accessibility measures. PPAs can also be aggregated to create surfaces similar to those of conventional accessibility measures
(Kwan, 1998; Huisman and Forer, 1998b; Forer and Huisman, 1998).
While the potential path area of individuals can be defined relatively easily within
Euclidean space (Bums, 1979; Villoria, 1989; Newsome, Walcott and Smith, 1998), this is not a realistic depiction of urban space and movement possibilities, as in actual urban areas movement is likely to be constrained by street or transit networks. In order to take full advantage of the ability of space-time measures to precisely evaluate and visualize individual accessibility movement possibilities should be shown with as great a precision as possible. Methods for calculating network-based PPA have therefore been developed
36 (Miller, 1991; Kwan and Hong, 1998). Instead of calculating a true PPA (which would
take the form of an ellipse) these methods identify only those street network links that
can be reached within the available time. Because it is assumed that all activities and
movement takes place along the street network this limitation is not a problem, and the
set of feasible links can be taken to represent a network-based PPA (with size measured by the number or total length of the links). Studies implementing network based PPAs have been carried out and have shown both the feasibility of these methods as well as uncovering important gender and employment differences in accessibility among individuals (Kwan, 1998,1999a; Kwan and Hong, 1998). These differences were invisible using conventional accessibiUty measures, proving the value not only of space time measures but also of incorporating the extra detail of network PPAs. The use of these measures has also revealed that commuting distance is unlikely to have much relationship to individual accessibility as the distance traveled to work, the amount of free time an individual possesses, and the number of opportunities available along the commuting route may all differ considerably (Kwan, 1998, 1999a).
The use of space-time measures of individual accessibility has a number of advantages when compared to conventional measures. Accessibility is an attribute possessed by individuals, who create it through their daily activities and movements. It is not necessary to specify the importance of distance to individual behavior, and these measures can incorporate interpersonal differences, household responsibilities, and time of day. Rather than being distance based measures they can be thought of as context based measures, incorporating both the individual's own activities and constraints as well as characteristics of the individual's urban environment.
37 2.3.2 Geographical context and space-time accessibility measures
Time geography and space-time measures of accessibility are well suited for
showing spatial variations in individual accessibihty, as geography is built into these
methods by its influence on mobility, the local availabihty of activities, and contextual
influences on the individual's knowledge and behavior. The importance of place has been shown by several early applications of time geography that used estimated
characteristics of individuals to show how the feasibihty of their activity schedules varied
at different locations within a city (Lenntorp, 1978; Forer and Kivell, 1981; and Miller,
1982).
With the exception of Miller (1982), who found that women's social status was the most important determinant, the results of this body of research showed that the ability of individuals to carry out activities depended in large part on the relative location of an individual and the proposed activity. This is not only because of the lengthy travel times required from certain peripheral areas but also because of the structure of the transit networks, which were focused on the city center, making this area easy to get to but disadvantaging individuals wanting to get from one suburban area to another. This shows that the structure of transport networks (and changes to this network) can be expected to have important consequences to accessibility patterns in a city (as does the location of activities, residences, and workplaces). However, because each study was based on the mobility constraints of housewives and made use of public transportation only, these represent worst case scenarios of limited mobility. More sophisticated uses of time geography are needed to better understand spatial variations in accessibility.
38 More recent applications of space-time measures have made use of individual's
PPAs as a measure of accessibility. The importance of location within these studies can be shown by the extent to which individual's living (or working) in different areas possess different accessibilities. The largest study that carried out such an approach made use of eleven transport modes (each with its own estimated travel speed) and 63 zones within a city located in the Philippines (Villoria, 1989). Although this study used
Euclidean distances and is of limited relevance for North American cities, accessibility variations between zones (and population groups) were identified. The median accessibility values of inner suburbs showed the highest values (explained by the presence of higher income individuals and good public transportation) while the central city zones had the lowest values (the poorest individuals were found here along with poor public transportation service). More generally, households owning automobiles and those with higher incomes had higher accessibility levels, as would be expected. The correlation between location and socioeconomic group makes statements about the influence of location on accessibihty problematic, as it cannot be known for sure whether it is the characteristics of the location or of the people's activities (or both) that have created this difference. So while households living in a certain area may have above average accessibility, it is difficult to directly determine whether this is because of their behavior or because their location has allowed them high mobility and provides them with a large choice o f potential activities. More recent space-time studies have focused on auto travel within street networks for North American cities, but have used small samples and have not focused explicitly on the importance of place, so this question remains unresolved (Kwan, 1998, 1999a, Kwan and Hong, 1998).
39 This issue is important because even with the use o f space-time accessibility measures there is still the need to isolate the effects of people's behavior and that o f the urban environment. While the link between individual travel and activity behavior and the urban environment has been heavily studied for aggregate populations (Handy, 1992;
Ewing et al. 1994; Frank and Pivo 1994; Friedman et al, 1994; Kitamura et al. 1997), there remains the need to examine this relationship at the individual level. An emerging methodology that offers considerable promise for doing this is multilevel modeling, an extension of multivariate regression in which makes use of both area and individual information (Paterson and Goldstein, 1991; Jones, 1991a, 1991b; Jones and Duncan,
1996; Bullen, Jones, and Duncan, 1997). This method allows the isolation of variations resulting from variability in the population (compositional effects) from those resulting from differences between areas (contextual effects), while also avoiding the risk of ecological fallacies by using disaggregate data. Multilevel modeling not only allows different average values in different areas but also different relationships between different areas. The methodology therefore incorporates local contexts into more general relationships by treating variations as normal rather than error. If applied to the study of accessibility multilevel techniques should help overcome the limitation of viewing accessibility as a purely personal characteristic or as a purely environmental characteristic. The multilevel approach instead more appropriately treats accessibility as a combination of both, with individual accessibility being mediated or influenced by the characteristics of particular areas. Multilevel modeling methods will therefore be used in this research to assess local variations in accessibility, as discussed in chapter 5.
40 2.4 Conclusions
This chapter has shown that contemporary urbanization processes and patterns
require a rethinking of both notions of urban form as well as measures of accessibility.
The logic of monocentric and polycentric cities, largely reflected by conventional
accessibihty measures, treats accessibihty as an attribute of places, suppresses the role of
individual characteristics or differences, and emphasizes the role of distance in creating
accessibihty. These urban models suffer a number of limitations in describing
contemporary American urbanization due to their emphasis on the power of distance and
the homogeneity of the urban population. Distance is clearly of declining importance in
shaping cities as it is of questionable influence on land uses or individuals. Neither is
human behavior as regular or homogenous as would be expected, in part due to diversity
among households and individuals and also because the relative lack of mobility these
households experience.
These limitations strongly suggest that alternative perspectives on cities and
human behavior within them are needed. While emerging perspectives have asserted that
the there is no longer any possibility for universal explanations of urban form and behavior, this has been hotly contested by those who maintain that cities continue to be
shaped by the same forces as before. It is not necessary to accept cities as postmodern in
order to appreciate that this perspective offers a considerably different viewpoint for
interpreting cities and households. The importance of geographical contexts within cities
is highlighted, including the ways in which political jurisdictions influence land use, mobility, and housing characteristics. The removal of assumptions about a normal suburb to center commute allows for recognition of the ways in which individual travel
41 patterns shape their experience of the city, and how people create their own accessibilities through their daily behavior. This accessibility is clearly dependent on the contexts within which people live, and it can be expected that these contexts are dependent on the geographies of the cities and neighborhoods in which people hve, work, shop, and carry out other activities.
As shown above, conventional aggregate measures largely mirror the assumptions of the monocentric and polycentric urban models. With their emphasis on measuring accessibihty for points or zones rather than households or peoples, their inabihty to incorporate daily activity schedules and interpersonal differences, these measures contain numerous limitations that prevent them from being suitable for evaluating individual accessibility within contemporary cities. Only space-time measures appear suitable for the challenge of measuring individual accessibility within the dynamism and complexity of contemporary cities, though they have not yet been used to their full potential with large individual data sets and dynamic temporal constraints. However, even when this representation of individual accessibility is used, there is still no direct way of evaluating the effects of place on accessibility. The question of whether uneven accessibility is due to the presence of a population with high mobility and/or few time constraints or the availability of a large number of potential activities and/or a transportation network that allows easy mobility remains to be resolved. Given the attention directed at the relationship between land use and travel behavior, this is an important question.
Fortunately, the availabihty of multilevel modeling techniques allows this issue to be explored.
.42 The next chapter will discuss how space-time measures can be implemented with a large individual activity data set while incorporating information on the dynamic nature of mobility within cities and the uneven spatial and temporal availability of potential activities. Doing so requires the consideration of a wide range of issues, including the evaluation of the fixity of individual activities, measuring travel times through a street network under varying traffic conditions, and an innovative use of time to limit accessibility at certain times of the day. The spatial patterns of individual accessibility produced by these measures will be explored and explained in Chapters 4 and 5.
43 CHAPTER 3
DATA AND METHODOLOGY
3.1 Introduction
The importance of place and contexts to intraurban individual accessibility will be
tested using the example of Portland, Oregon. A wide range of data is necessary for this
research, beginning with a highly detailed and disaggregate source of daily travel
behavior for individuals from which fixed activity locations and durations can be
obtained. In order to represent the availability of potential activity opportunities to
individuals a detailed spatial data set containing the location and business hours of
employment, retail, and other services in the Portland urban area is also required.
Finally, a digital street network that includes estimates of both free flow and congested
travel times is necessary to allow the mobility of individuals to be accurately modeled at
different times of the day. These data were obtained from local planning agencies. Field
observations and interviews with local transportation planners in Portland carried out in
November 2000 provided additional knowledge of driving and mobility conditions in the
Portland area, helped identify major employment centers (discussed in more detail in
Chapter Four), and provided additional details on the goals of transport and land use planning (also discussed in Chapter Four).
44 An existing algorithm (developed by Kwan and Hong, 1998) was used to compute
accessibility, but because this algorithm does not make use of the temporal dimension
when measuring travel times or the availability of opportunities, modifications were
made to allow the incorporation of this additional detail. These modifications and the
characteristics o f the data sources used will be described below, beginning with a general
overview of the study area, along with a discussion of the accessibility measures created
using this algorithm.
3.2 Study Area and Data
The study area for this research comprises parts o f Mulmomah, Washington, and
Clackamas counties in Oregon and Clark County in Washington. These four counties as
well as two additional counties in Oregon (Columbia and Yamhill) make up the Portland-
Vancouver Primary Metropolitan Statistical Area (PMSA), which in turn is part of the
Portland-Salem Consolidated Metropolitan Statistical Area (CMSA). The four counties
included in the study area contained a population of 1,595,128 people in 1995, making up
93.15% of the population of the Portland-V ancouver PMSA, as Columbia and Yamhill
counties are largely rural areas outside the Portland urban area (U.S. Census Bureau,
2001). Much of this area is included within the urban growth boundary of the Portland
Metropolitan Service District (commonly known as Metro), which is the local metropolitan planning organization (MPO) for the Portland urban area, as well as being the nation’s only elected regional government (Cambridge Systematics, 1996; Metro,
1997). The boundary for this organization includes most of Mulmomah county and the urban sections of Clackamas and Washington counties (Figure 3.1). This boundary was
45 created around the Portland urbanized area in 1979 in an attempt to limit urban sprawl and preserve farmland (Nelson, 1994; Metro, 1997). The boundary is required to include all land necessary for 20 years of growth, and must be reevaluated every five years for possible increases, which have so far been minor despite strong pressures to extend the boundary (Kloster, 2000). Because it bounds the urbanized area it provides an effective study area boundary and will be used as such for this research.
Washington
WA
OR
■ Portland CBD /\y Freeways Oregon ggg Rivers m i Portland 0 5 10 15 20 Miles MHI Major Suburbs n r# ] Metro Boundary
Figure 3.1: Portland, Oregon, study area
46 The Portland urban growth boundary encompasses a total of 24 municipalities and a population of approximately 1.3 million residents (Metro, 2001). The city of Portland is by far the largest city in the study area, with 498,747 people in 1995. Although it dominates the study area, Portland includes only 31.27% of the total study area population. A considerable portion of the study area population actually resides within a large number of suburban municipalities, including Gresham (79,431 people) to the east of Portland, Beaverton (62,573), Hillsboro (47,309) and Tigard (35,054) to the west in eastern Washington County, and Lake Oswego (33,606), Milwaukie (19,977), and
Tualatin (19,353) to the south of Portland (U.S. Census Bureau, 2001). Vancouver,
Washington, is a sizable (62,771) community lying just outside the study area.
This area includes a range of geographic features that can be expected to have an influence on accessibility (the following paragraphs are based in part on field observation carried out in November 2000). The study area is traversed by two major rivers, which are potentially important barriers to mobility. The Columbia River crosses firom East to
West along the north edge of the study area and is the state line between Washington and
Oregon as well as part of the urban growth boundary. Only two highway bridges cross this river, both carrying fireeways (Figure 3.1). 1-5 crosses on the westernmost bridge and serves as a major north-south route as well as cormecting the Portland area to Vancouver and Clark County, Washington, while 1-205 crosses farther east. There are no other highway bridges over the Columbia in the vicinity of the study area, and no plans exist to build additional bridges (Kloster, 2000). The Willamette River bisects the study area as it flows firom south to north and into the Columbia northwest of the study area. This is a much smaller river and is crossed by 14 highway bridges. Of these, eight are in or near
47 the Portland CBD, including 1-5 and 1-405 (part of the downtown freeway loop) (Figure
3.1). As the Portland CBD is located on the west side o f the Willamette River these
bridges are extremely important for cormecting it to the east side o f the metropolitan area,
though all but the freeway bridges were originally built in the early 20th century to serve
local trafBc (Abbott, 1983). Only one bridge exists downriver (north) o f the downtown
area providing the only crossing point in the approximately 11 river miles between downtown Portland and the mouth of the Willamette River. There are five bridges upriver (south) of the Portland CBD and within or just outside the urban growth boundary, including 1-5 and 1-205 freeway bridges (Figure 3.1).
The terrain within the study area varies considerably, also potentially reducing mobility. The major landform feature is the West Hills, which are a chain of low mountains running along the west side of the Willamette River and generally increase in elevation and significance to the north. These hills disrupt the street pattern of the
Portland area, particularly to the northwest o f downtown (and those roads that do cross the hills are narrow, steep and winding). To the east of the West Hills the terrain is flat to rolling, and the street pattern within the city of Portland tends to be a regular grid, though with deviations around hills, major drainage features, and railroad lines. Farther to the east and southeast newer suburban areas include many subdivisions with curvilinear streets, and the overall street density decreases. Scattered volcanic peaks also break up the terrain and street pattern, especially in Clackamas County. To the west of the West
Hills (in eastern Washington County) the terrain is also generally flat, but because the area has become urbanized primarily in the postwar era the street pattern includes many curvilinear streets.
48 As in most American cities, mobility within tlie Portland area is heavily auto dependent. In 1990, 72.6% of workers drove alone, 12.5% carpooled, 6% relied on public transit, and the remaining 8.9% used other means (U.S. Census Bureau, 2001).
The Portland area includes several freeways that are important to auto mobility. 1-5 is one of the more important, as it runs North-South through the study area and passes the downtown area. This freeway serves commuting trafBc within the metro area as well as being the main North-South highway along the Pacific coast. It therefore includes considerable amounts of through traffic. 1-84 provides the main highway to the east from downtown Portland, while US 26 runs west from the CBD through the West Hills into eastern Washington County. On the east side of Portland 1-205 is a major link. This freeway connects with 1-5 on the south side of the study area and to the north in Clark
County, Washington. It runs through Oregon City and provides a north south link within the east side of Portland. In eastem Washington County Oregon Route 217 provides a north south link and connects US 26 and 1-5. Finally, 1-405 loops around downtown
Portland and provides a central hub, with connections to 1-5 and US 26. In addition to allowing faster travel, these freeways are important for mobihty in the Portland area because they provide the only bridges over the Columbia River and several important bridges over the Willamette River. Given the hub and spoke pattern centered on the
Portland CBD, it can be expected that individuals living or working near this location will enjoy high accessibility.
Although mobility remains focused on the use of autos, transit and especially rail systems are important components of transportation planning in Portland. In 1986 the
MAX light rail line running from the Portland CBD eastwards to Gresham was opened,
49 followed by a western line to Hillsboro in 1998 (Tri-Met, 2001). These lines provide a
continuous 33 mile east-west line running the length of the urban growth boundary, and
accounted for about 25% of all transit trips in 2000. This ridership is increasing and the
presence of these rail lines and stations are seen as catalysts for land redevelopment in
older areas of the city as well as providing an opportunity for creating high density transit
oriented development along undeveloped stretches of the line west of the CBD. (Tri-met,
2001). The importance of these development patterns to accessibility will be discussed
further in the next chapter. Additional MAX Unes will serve the Portland airport and
parallel 1-5 north to the Columbia River, along with streetcar rail service beginning
within the CBD in 2001, which will further increase the centrality of the Portland CBD
within the regional rail transit network (Tri-Met, 2001). The CBD can therefore be seen
as a hub for both auto and rail networks.
3.2.1 Individual accessibility data
The use of space-time measures of individual accessibility requires disaggregate
individual activity data showing the kinds of activities individuals engage in (regardless
of whether travel is required), where they take place, and the means of travel to and from
these activities. For this type of information researchers are increasingly turning to diary
surveys (Axhausen, 1995; U.S. Department of Transportation, 1996). This data was
therefore obtained for the Portland study area from the Household Activity and Travel
Behavior Survey, a travel-activity diary survey collected in 1994 and 1995 (Cambridge
Systematics, 1996). This survey collected activity and travel information from 4,451 households and 9,471 individuals throughout the Portland area over a two-day period for
50 purposes of transportation planning by the Portland Metropolitan Service District (Kim,
2000). Because data was collected for activity based modeling, particular attention was
given to capturing information on household structure and location and the use of non
automobile modes. The survey therefore obtained a wide range of data from the
respondents, including household and personal characteristics, and details of all out of
home activities as well as in-home activities lasting at least 30 minutes. These activities
were collected in a travel diary format, so that individuals were asked to record activity
information as they engaged in them (including the location of each activity, mode of
travel, the time activities began and ended, and the purpose of the activity). Each
individual in the sample recorded their activities and movements during a two-day period
assigned them. The travel-activity data set therefore provides detailed and disaggregate
information about the activities and movements of a large number of individuals during
the course of several days. After the data was collected and assembled by Metro, the household and activity location was geographically referenced to an accuracy o f200 feet, making possible the accurate representation of individual’s movements through space.
Although the Portland travel diary survey included individuals traveling by a variety (and combination) of transportation modes, only those traveling by private
automobile were used in this research. This ensures that variations in accessibility are not due to transport mode while greatly simplifying the computation of accessibihty, as no scheduling considerations are involved. The individuals used for this research were
further restricted to those adults living inside the Metro urban growth boundary with at least one fixed activity (in or out of the home) each day, and with all activities taking place within the Portland MSA counties on weekdays. 755 individuals from 598
51 households met these conditions and were used in this research. These individuals engaged in an average o f six activities each day. This sample is split roughly equally between men and women (52.7% male) and is overwhelmingly (94.7%) white.
Employment status of the individuals sampled is slightly more diverse, with 77.6% employed full time, 13.7% part time, 2.6% are full time homemakers, 1.5% retired, and
4.3% are not employed. The residential location of these individuals is shown in Figure
3.2.
■ Portland CBD / \ / Freeways Rivers ■ Individuals ( I I i l l i | Metro Boundary
Figure 3.2: Residential location of individuals in sample
52 To represent activity opportunities the location and characteristics of all potential
employment, retail, entertainment, and other opportunities in the Portland urban area was needed. To provide this a data set containing all property parcels in the Portland area was
assembled from local land use data. Data from the Oregon counties was obtained from
Metro's Regional Land hiformation System (RLIS), an extensive GIS database collection maintained for planning purposes. Parcel data for Clark County, Washington, was obtained from the county planning agency for that area. Unfortunately, the two data sets contained different land use classification schemes, and these varied considerably in the attribute information available for the parcels. The Oregon RLIS data employs less detailed classification of land uses than that for Clark County, with all parcels categorized as either agriculture, commercial, forest, industrial, multi-family residential, single family residential, public/semi-public, rural, or undeveloped/vacant. Those parcels classified as commercial or industrial were used to represent potential activity opportunities. Parcels classified as public were excluded from the analysis because these tended to include a wide range o f land uses, including parks and cemeteries that do not represent employment locations. The two data sets were combined and labeled according to the
Portland RLIS land use classes to provide a single classification scheme.
Following Kwan (1998, 1999a), this polygonal parcel data set was converted to point data by computing the centroids of each polygon using a standard Arc View GIS procedure. While the results may not be as precise as geocoding each building to its street address, this procedure was not feasible given the very large size of the database, and would be of limited use for large shopping centers or office parks, in which multiple establishments share the same building or are built around private streets, driveways, or
53 parking lots which do not appear on the Portland network. The smaU average size of
parcels and the dense street network ensures that the centroids will be close to the street.
The resulting point data set contained a total o f27,749 parcel centroids.
3.3 Street Network and Estimation of Travel Times
The calculation of individual accessibility also requires a detailed street network
showing not just the major arterial highways but also the local streets where individuals
live and work. Further, these streets must be represented with a high degree of spatial
accuracy in order to accurately measure the spatial relationship between streets and
activity locations, which is necessary for the creation of space-time accessibility measures. The digital street network that best satisfies these requirements is the network component of the U.S. Census Topologically Integrated Geographic Encoding and
Referencing (TIGER) spatial database. This database was developed in the 1980s
following several decades of work by the U.S. Census Bureau to better represent the geographic location of decennial census data (Marx, 1986, Sperling, 1995; Cooke, 1998).
For this purpose digital representations of street networks (as well as many other geographic features) were created for the entire country, with sufficient topological information to allow them to be readily used within GIS for network applications. While considerable effort has been directed at creating more specialized street networks for transportation applications (Fletcher, 1987; Shaw, 1993; Mainguenaud, 1995; Miller and
Storm, 1996; Fohl et al., 1996; Kwan and Hong, 1998; Choi and Jang, 2000) the TIGER network is widely available for all cities and has essentially become the standard GIS
54 urban street network representation. While it is not without its limitations (Fohl, et al,
1996) it possesses the detail and spatial resolution needed for creating space-time accessibility measures, and so is used in this research.
TIGER street networks for the study area counties were obtained from Metro and the Clark County planning agency. These were joined together to create a single network theme for the study area with 130,141 links and 104,048 nodes. Because this network has a planar structure, allowing turns to be made from any link to any link, it was edited to prevent turns from freeways to or from other streets in locations where these roadways crossed but no freeway interchange was present. Because space-time measures require detailed representations of driving times through the street network in order to more accurately movement possibilities within the city, this information must be integrated into the TIGER network database. While some have estimated travel times from speed limits
(Muraco, 1972; Wang, 2000) or used interzonal travel time data or estimates (Wachs and
Kumagi, 1973; Black and Conroy, 1977; Knox, 1978; Handy, 1993; Geertman and Van
Eck, 1995; Helling, 1998; Scott, 1998), time geographic studies, and especially space time accessibility measures, are dependent on point to point driving times and so require estimates of travel times at a high degree of spatial resolution. This has been accomplished by making assumptions about driving speeds along different types of roadways, such as freeways, arterial streets, and local streets (Brainard et al, 1997, 1999;
Kwan, 1998,1999a; Kwan and Hong, 1998). Although speeds, and therefore travel times, could be estimated in Portland by functional class of roadway, it is possible to make use o f additional data for this area. This allows for the fact that travel times cannot be expected to be homogenous or uniform even within different classes of roads, let alone
55 between densely urbanized areas and the rural periphery. Travel time information was
therefore supplied from a specialized street network based on the EMME/2 transport
planning model. This network model was not directly suitable for use with space-time
accessibility measures due to the fact that it includes only primary arterial streets and
freeways (with only 19,638 links) and represents them with less spatial precision than the
TIGER network, but it does include useful travel time data. As can be seen, the spatial
pattern of driving speeds within this network favors freeways, but many streets located in
peripheral areas within the Metro boundary, as well as rural roadways in Clark County
and eastem Washington County also have high speeds (Figure 3.3). The slowest free
flow speeds are largely concentrated among streets in central Portland, though they also
appear in rural streets in Clackamas County as well as Vancouver, Washington.
In order to transfer this travel time data from the EMME/2 database to the TIGER
database, each link in the Portland street network was classified by both function and
location in order to isolate spatial variation of travel times. This was carried out by using
the CFCC coding contained within the TIGER network (supplemented by reference to road maps) to first classify all roadways by function (freeway, primary street, secondary
street, or other street) and then by location (inside downtown, outside downtown, outside
Metro boundary). Each freeway was classified individually (as 1-5,1-205,1-84, US 26,
OR 217, or the downtown freeway loop, which includes 1-405 as well as a short portion of 1-5) as well as whether located inside or outside the Metro boundary. A total of 15 classes of roadways were obtained from this procedure.
56 Free Flow Speeds < 12.5 MPH A / 12.5-25 / V 25-37.5 A/ 37.5-50 A / > 50 MPH T I Metro Boundary A
Figure 3.3: Free flow speeds through Portland street network
After first classifying the EMME/2 network according to the same fimctional and locational groupings as the TIGER file, link-specific free flow driving speeds contained in the EMME/2 network were transferred to the corresponding TIGER links and used with the TIGER link lengths to generate free flow travel times. Because the EMME/2 network does not contain local and many secondary streets (which were present in the
TIGER network) a speed of ten miles per hour was assigned to these streets. Following
Kwan and Hong (1998) travel times for all roadways was further increased by 25% to allow for intersection delays as well as time needed to walk to/from opportunities after parking.
57 In addition to spatial variation within the network, there will also be variations in driving times by time o f day due to congestion. Because the effects of congestion are not included within the TIGER database, the EMME/2 network was again used to supply needed travel time data. In this case, congested speeds were first calculated within
EMME/2 using the standard Bureau of Public Roads (BPR) speed-flow equation, which uses firee flow speeds, traffic capacity, and peak period traffic volumes to estimate link- specific peak period driving speeds through a network (Dowling Associates, 1997). The
BPR equation is:
s = sf I (U a{v / cf)
Where S = predicted speed, sf is firee flow speed, v = volume, c = capacity, and a and b are parameters usually set to 0.15 and 4, respectively. In this equation speed is largely a function of the ratio between traffic volume and roadway capacity, with speeds falling rapidly once the roadway volume/capacity ratio exceeds 1.0. The application of congestion produces strong declines in driving speeds within the Portland street network, particularly on major arterial streets within the Metro boundary (Figure 3.4). Freeways appear to be only mildly affected by congestion, though central Portland (including the downtown area) and Eastem Washington County, an area of rapid growth and lagging infirastructure improvements, contain many streets with peak period speed reductions of
80% or more. The effects of congestion on driving times appears to be highly localized, with many streets showing relatively little reduction during evening congestion. The effects of these reductions on individual accessibility will be shown in Chapter Four. 58 These congested speeds were transferred to the TIGER network in the same
fashion as free flow speeds, and were applied to all travel taking place between 4:00 PM
and 6:00 PM. Free flow speeds were used during all other times of the day. Applying
congested driving speeds during the evening hours only more easily allows the isolation
of the effects of congestion on individual accessibility, especially as most discretionary
travel takes place during this time period (Kwan, 1999b). Because the Kwan and Hong
(1998) algorithm uses constant travel times this was modified to allow for the use of peak
period travel times during evening hours. This was done by identifying the temporal
midpoint of each PPA and applying congested travel times if this midpoint falls within
the 4:00 PM to 6:00 PM congested period. If it does not, free flow travel times are used.
While this results in an unrealistic assumption of "instantaneous congestion" (Bums
1979), it is computationally efficient and is consistent with other space-time accessibility
research (Huisman and Forer, 1998).
3.4 Space-Time Accessibility Measures
Space-time measures of accessibility are based on showing the area an individual
can move about in (and the potential activities which exist inside that area) within the time available to him or her. This concept can best be explained through the example of
an individual with a daily activity schedule. His or her schedule will contain a number of
activities that must be carried out over the course o f the day at varying locations. Some of these activities will have to be carried out at a particular place at a definite time (and often for a certain length o f time), and so can be considered to be fixed activities. The individual must accept the time and place of such fixed activities, which commonly
59 include work, school, medical appointments, or child care responsibihties. These fixed
activities provide the spatial and temporal firamework for the individual’s day as they
determine where and when he or she must be, and for how long. Other activities can only be engaged in during the time available (if any) between these fixed activities. If
successive fixed activities are not at the same location then the time spent moving between these activity locations will further reduce the time available to engage in other
activities (and the slower the mode of transportation, the less time will be available).
Other activities will allow more fi-eedom, as the individual can choose among a range of locations or times to engage in that particular activity, or skip it altogether. These can be considered to be flexible activities, and could include grocery shopping, choosing a gas station, visiting a post office, or renting a video. However, an individual’s abihty to choose among locations or times for flexible activities will still be limited by the time available to them between fixed activities and the limits of their mobility.
The time and mobility available to an individual is therefore vital to their ability to engage in a range of flexible activities. As discussed in Chapter Two, the area they can potentially reach within these limits is their Potential Path Area (PPA). This contains all possible combinations of routes he or she could traverse while traveling between successive fixed activities within the time available. Only those potential activities that can be found within the PPA are available to them. Because these PPAs are based on movement through an urban street network they will reflect varying speeds over street and fi-eeways and therefore tend to be highly irregularly shaped, rather than the ellipses common when calculating them with straight-line distance (Figure 3.5).
Computing PPAs based on movement through a street network is considerably more
60 challenging than with Euclidean distance, as it must take into account movement through a detailed network and the proximity and characteristics of potential opportunities to this network space, and so can only be carried out within a GIS environment. Several algorithms exist to carry out this procedure (Miller, 1991, 1999; Kwan and Hong, 1998), and the Kwan and Hong algorithm was used due to its computational efficiency when mn in ArcView 3.2 GIS on a standard PC.
Speed Reduction A /> 8 0 % / A y 60 - 80% / \ / 40 - 60% / V / 2 0 - 40% < 20% I I Metro Boundary
Figure 3.4: Peak period speed reductions in Portland street network
61 Because the accessibility of individuals is constrained by the need to take part in
fixed activities throughout the course of the day, these must first be identified. While
specialized activity data collection efforts (Kwan, 1998,1999b; Kwan and Hong, 1998)
allow respondents to designate the level o f fixity of an activity, the Portland activity set
does not contain this information. For this research it was necessary to designate each of
the 28 categories of activities hsted within the Portland data set as fixed or flexible.
Research within space-time accessibility has commonly focused on work and school
activities because of their daily nature and lack of flexibility (Villoria, 1989; Huisman
and Forer, 1998a), and so activities coded as 'work' and 'school' in the Portland data set were treated as fixed activities. Additionally, activities coded as 'medical care', and
'professional services' were treated as fixed because of their importance to the individual and the likely inflexibility in scheduling. Finally, activities labeled 'pick up/drop off passengers' were treated as fixed because this category will include work, school, or day care related activities that individuals will have little scheduling control over.
One limitation of the Portland activity data set is that it is ambiguous about each individual's first and last activities of the day. In the event the first and last activities of the day are not carried out at home there is no way of knowing when the individual left home in the morning or returned home in the evening. While others have assumed that no activities are possible for individuals between a designated evening hour (such as 7:00
PM) and a designated morning hour (such as 7:00 AM), such a strategy is only reasonable when dealing with activities that must necessarily take place during the day
(such as attending school) (Huisman and Forer, 1998a). The Portland data set does not allow for this as individuals may take part in activities at any time of the day, and in fact
62 several individuals have employment or other activities lasting through the night. While information is present in the data set that provides the start and end times of each trip there is no variable that could be used to identify what time they returned home after the last activity (if they were not already home). It was therefore assumed that individuals did not carry out any activities before their first fixed activity, then traveled to that location by the shortest path and so have no time available to engage in other activities on the way there. In the evening it was assumed that individuals would travel directly home after their last fixed activity of the day and engage in no other activities for the remainder of that day.
• Origin ■ Destination • Other Activities Network PPA / \ / Streets
N A
<5n îB A k H h 1 0 O j6 1 2 1.8 Mites
Figure 3.5: Example of network Potential Path Area (PPA)
63 Because of a tendency to round off reported starting and ending times (and
therefore travel times), as well as the difference between estimated and actual travel
times, shorter trips often presented problems for the algorithm used because reported
travel times were sometimes less than that measured through the network. Certain procedures were put into place to handle these trips. In the event that the reported travel
time between two activity locations was less than the shortest path calculated between those locations through the street network, no network PPA was calculated. In such cases it is assumed that the individual had no time to make any stops, and so had no PPA within which to move around in. Also, the algorithm was set up to use five minute intervals for prisms lasting longer than 10 minutes, and one minute intervals for less than
10 minutes. While Kwan and Hong (1998) uses two minute intervals for less than 10 minutes, experience with the Portland network revealed that one minute intervals were preferable.
3.5 Specifying Accessibility Measures
Five space-time accessibihty measures were implemented and tested in this research, all derived fi-om the sum of each individual’s daily PPAs (or DPP A). These
DPP As are then averaged over the two days for which activity data was collected for each individual. The simplest of these measures is the number of miles of street present within an individual's daily potential path areas (labeled MILES in this research). This measure assumes that the greater the street mileage an individual can move around in between subsequent fixed activities, the greater that individual's accessibility. While this
64 measure clearly emphasizes mobility, as the faster an individual can drive within the available time the greater their accessibility will be, it also reflects location. Individuals moving about in older, more densely settled areas of Portland will potentially be able to cover a greater mileage of streets than an individual in an outlying area with a sparser street network. Access to nearby freeways will also be important to higher accessibihty.
However, this measure is also sensitive to highway congestion, so that individuals moving about in areas of greater congestion, and/or carrying out activities at times of the day when congestion exists, will suffer disproportionate accessibility reductions from this congestion.
A second measure counts the number of opportunities present within the daily
PPA to allow the uneven distribution of opportunities within a city to influence individual accessibihty (Kwan and Hong, 1998; Kwan, 1999a). As in previous research, a 0.25 mile buffer was defined around each street link in the PPA to represent a reasonable walking distance from the street, with only those opportunities present within this buffer were counted as accessible to the individual. This measure (labeled OPPORTUNITIES) therefore counts the number of opportunities available to the individual and takes into account the fact that an individual with high mobility will not necessarily be able to reach a large number of opportunities, especially if they are traveling in an area with limited commercial development. This will likely help distinguish between individuals traveling within the densely built up central areas of Portland and the outlying areas of the city.
As in other space-time accessibility analyses (Kwan, 1998, 1999a; Kwan and
Hong, 1998) the relative importance and attractiveness of opportunities was also taken into account by weighting them according to their size. First, the square footage of
65 opportunity parcels within the PPA was used as a measure to represent the differing size
of individual opportunities (floor area of buildings on the lot was not consistently
available and so this could not be used). This measure (called AREA) takes into
consideration that some activity opportunities are considerably larger and therefore more
important or attractive to individuals than are others. This is likely to have the effect of
weighting accessibility in favor of areas with large shopping centers (especially shopping
malls) or business establishments. Because buildings located in downtown Portland
commonly have multiple floors and higher building to parcel size ratios, the square
footage of these parcels was weighted by a factor of 10. Lesser concentrations of multistory buildings were identified outside the Portland CBD in the Lloyd Center area of
Central Portland and the Washington Square area of eastern Washington County were identified by the author by field observations and are distinguishable fi'om surrounding areas using the criteria outlined by Giuliano and Small (1991). This procedure for identifying employment centers requires a total employment within Traffic Analysis
Zones (TAZ) of at least 10,000 and a minimum employment density of 10 workers per acre. Opportunity parcels within these areas were weighted by a factor of 1.5 to reflect their greater importance relative to surrounding opportunities. This weighting created a second measure of the attractiveness of opportunities (here called WEIGHTED AREA).
The density of weighted opportunities in the Portland study area can be seen in Figure
3.6.
Finally, a fifth accessibility measure was devised in an attempt to represent the temporal availability of potential opportunities. This measure used the WEIGHTED
AREA measure but also incorporated business hours, so that opportunities that were
66 closed at the time an individual was traveling were not counted as part of that person’s
accessibility. Time geography research has shown that the authority constraint of limited
business hours can be as crucial to the feasibihty of carrying out an activity as the time
budgets or mobility of individuals, by limiting the availabihty of activities to certain (and
perhaps very brief) times of the day (Forer and Kivell, 1981; Lenntorp, 1976; and Miller,
1982). These are often times when individuals must be engaged in employment or when
travel is restricted, greatly reducing access to these activities. Given that men and women
tend to carry out different amount and kinds of activities at different times of the day
(Kwan, 1999b) it can be expected that the accessibihty impact of business hours will not be distributed evenly among the population. Business hours have only recently been
incorporated into accessibility studies using space-time measures, but it has been shown that their use does have a considerable effect on the accessibihty of individuals to opportunities, and that accessibihty measures that do not incorporate temporal availabihty may therefore exaggerate accessibihty (Weber and Kwan, 2000). This temporally weighted measure (called TIMED AREA) should therefore more realistically portray the accessibihty of those individuals who engage in many activities at night or very early morning.
While restricting the calculation of accessibihty to activities carried out during the daytime hours (as by Huisman and Forer, 1998a) minimizes the need to incorporate business hours, the Portland travel diary data set includes activities carried out at all hours of the day. Because the land use data used to construct the opportunity data set does not contain business hours, all parcels representing potential activity opportunities were assumed to be available from 9:00 AM to 6:00 PM. If the temporal midpoint of an
67 individual’s PPA fell between these hours parcels located within an individual's PPA would be counted towards their accessibility, but if the midpoint was outside these hours they were not. This recognizes the fact that while individuals may have considerable nighttime mobility (as measured by the number of street miles within their PPA) there will actually be very few out-of-home activities that they can carry out during these hours, and so they will have a very low accessibility. The ability to schedule activities at favorable times of the day will therefore be as important to accessibility as mobility, but again there is no reason to expect this ability to be constant throughout the city or among the population (Kwan, 1999b).
Figure 3.6: Weighted opportunity density surface of Portland study area 68 3.6 Conclusions
This chapter has discussed the collection of data and procedures required for the computation of individual accessibility measures within a GIS environment. The study area of Portland, Oregon, provides an ideal case for evaluating the importance of place and geographical context to an individual’s access to employment, shopping, and other services. While this moderately sized metropolitan area has experienced considerable suburban development, it still possesses a strong downtown area that accounts for about
20% of all metropolitan area employment (Metro, 1997) and is a freeway and transit hub.
Because the CBD is important to future planning efforts, the extent to which proximity to this location influences accessibihty is an important indicator of the extent to which monocentric forces are still operating within American cities. Portland is therefore an ideal setting for testing the continued relevancy of the monocentric model. Further, the presence of the urban growth boundary helps provide a coherent and meaningful study area boundary and makes the Portland area ideal for modeling mobility and access to unevenly distributed services.
As discussed in Chapter Two, individual accessibility is best evaluated through the concept of the Potential Path Area. Although this is conceptually simple, a considerable amount of highly disaggregate geographic and attribute data is required in order to implement this measure. This data, including a household level travel-activity diary, a detailed digital street network, and a comprehensive parcel database was collected from local planning agencies and edited to provide the information necessary to evaluate accessibility. This data includes the ability to model travel at varying times of the day, which represents an advance over previous space-time research. A total of five
69 individual accessibility measures were created for 755 individuals living and traveling within the Portland urban area exclusively by car. Despite the attention given to transit and other modes, Portland remains an auto dependent city.
Each of these five measures provides an alternative view of what constitutes accessibility. While these measures all incorporate mobility in creating daily PPAs, they differ in the extent to which they include opportunities in the urban environment and how those opportunities are represented. It can therefore be expected that each will identify a different dimension o f accessibility in Portland by revealing how accessibility is dependent on location within unevenly developed urban environments, as well as the time of day at which activities and travel are carried out. The spatial patterns revealed by these accessibility measures, as well as the importance of distance firom the CBD and other locations in explaining them, will be the subject of the next two chapters.
70 CHAPTER 4
ACCESSIBILITY IN PORTLAND
4.1 Introduction
This chapter will examine the spatial patterns of individual accessibility in
Portland, Oregon, using the five accessibility measures calculated with the procedures and data discussed in the previous chapter. Because there are a variety of ways in which these spatial patterns can be visualized and interpreted, several approaches will be utilized. Point data, such as that used here, are ideally suited for visualization in the form of three-dimensional surfaces. The accessibility values produced by the five measures will therefore be interpolated to create surfaces that allow these highly variable accessibihty patterns to be seen, as well as the relationship of the measures to each other.
But because of the need to better understand and interpret these patterns, especially in light of the claims of the monocentric and polycentric models, accessibility will also be represented by distance fi'om the CBD as well as a specified set of polycentric centers. Doing so reveals that accessibility does not vary with distance in a uniform or predictable manner fi'om downtown Portland, undermining the monocentric model. That this is so given the importance of the Portland CBD is a very significant
71 finding. Accessibility patterns are more consistent with the polycentric model, but the patterns visible are modified by (and partly the result of) the time of day that activities and travel are engaged in. The effects of the time of day not only vary spatially within
Portland but interacts with each other in unpredictable (and sometimes unexpected ways).
The importance of both time and individual behavior further undermines support for both urban models.
Because visualizing spatial patterns does not provide a conclusive test of the monocentric or polycentric logic, this chapter will also attempt to explain spatial patterns and variations in individual accessibihty in the context of these models. Distance, as well as additional variables likely to be of importance to accessibihty, will therefore be used with multivariate regression to attempt to identify the source of variation in each of the five accessibihty measures. While some distance variables are indeed significant in explaining individual accessibihty variations, the results show that the importance of certain central locations differs for the five accessibihty measures, and likely also varies by time of day. Other variables representing household and individual characteristics
(and can be related to time constraints) are also important in explaining accessibility variations, despite their invisibihty within the monocentric and polycentric models. The ways in which these variables are related to the five accessibihty strategies also point out possibilities for individuals to increase their accessibihty or avoid the effects of congestion and limited business hours.
The final section of this chapter will evaluate the relative contributions of the traditional monocentric and polycentric models to explaining accessibihty in Portland, as well as the consistency of the findings with the ideas of postmodern urbanism. The
72 findings will also be discussed in the context of ongoing urban planning efforts in
Portland, including the role of the urban growth boundary and the intended development of major employment centers.
4.2 Individual Accessibility in Portland
Average accessibility values, and the importance of interpersonal accessibility differences can be seen in Table 4.1. Average raw values are shown for each measure, but because these are in different units (MILES represents mileage of streets,
OPPORTUNITIES shows the number of parcels, and the area measures all provide values for the square footage of parcels accessible to individuals) these values have been standardized to a mean o f 100. This allows meaningful comparisons between the measures and can be used to discuss differences between particular groups of individuals as well.
As can be seen, males and females within the study sample have very similar levels of accessibility, except for the TIMED AREA measure, for which women have a greater (and above average) value. This is surprising, because other research has shown that men tend to possess higher accessibility than women (Kwan, 1998), but in this case the temporal nature of the measure must be considered. For women to have a higher access with this measure they must be engaging in a higher proportion of trips during daytime hours than men. The incorporation of limited facility hours does not disadvantage them as much as men, though conversely, they may be more likely to experience congestion. A similar situation exists with race, as non-whites possess significantly greater (and very high above average) access than whites, especially for the
73 TIMED AREA measure (though it should be noted that the average accessibility for men is only slightly below average). With both race and gender, it is time of day variation in travel behavior that accounts for this difference.
N Miles Opportunities Area Weighted Area Timed Area Avg Personal Accessibility 755 2587.4 1203524 13533.53 14432.31 5514.57
Standardized Accessibility 100.00 100.00 100.00 100.00 100.00 Gender Males 398 100.63 99.19 99.17 99.21 86.46 Females 357 99.30 100.91 100.92 100.88 115.10
Race Wfiite 715 98.97 99.31 99.04 99.11 96.96 Non-White 40 118.48 112.34 117.20 115.93 154.41
Employment Status All Full Time W orkers 606 94.61 95.28 95.16 95.20 76.96 All Part Time Workers 104 119.67 118.10 118.54 118.21 179.37 All Retired/Unemployed 45 127.14 121.78 122.28 122.54 226.81
Full Time Males 347 96.01 96.14 95.66 95.72 7273 O ther Males 51 132.04 119.91 123.08 122.98 179.83 Full Time Females 259 92.73 94.11 94.49 94.50 8Z63 Otfier Females 96 116.66 118.85 117.90 117.72 200.91
Place of Residence Portland City Limits 424 113.33 119.21 115.50 115.80 117.67 O ther Multnomah County 83 72.44 75.16 83.29 81.43 63.05 Washington County 101 95.91 85.07 91.19 90.44 115.98 Qackamas County 147 79.92 68.88 70.77 71.48 58.92
Note: Bold indicates differences are significant at p < 0.05; Italics indicates différences are significant at p < 0.01
Table 4.1: Accessibility characteristics of sample individuals
74 Employment status also makes an important difference to accessibility. Full time workers have less (and below average) access than part time workers according to all of the measures, and those not working have the highest accessibility of all. This is due to the importance of time constraints within space-time measures and time geography, as those working longer hours have less time to move around, and so tend to possess less accessibility. This has been observed in other space-time research as well (Villoria,
1989). When men and women are distinguished on the basis of employment status gender differences become more visible, with men not employed on a full time basis having the highest level of access, followed closely by part time or unemployed women.
However, with the exception of TIMED AREA, full time women have a lower accessibility than full time men (as well as other men and women), which appears to confirm expectations that employed women face greater (or additional) time constraints than do men as a result of household responsibilities. Although full time employment leads to lower accessibility, it does not do so equally for both genders.
Finally, location is also clearly important to accessibility, with significant differences existing among individuals living in different political jurisdictions in the
Portland area. Those living within the Portland city limits have the highest accessibility by all measures, suggesting a higher level of mobility and greater density of potential activities than available elsewhere (though it could also indicate differing travel behavior, which will be addressed in a later section). Residents of eastern Washington County, a prosperous and rapidly growing area, were second in levels of access but except for
TIMED AREA, these values were below the study area average (largely due to the overwhelming number of people living within Portland). Values for those living in
75 Clackamas County and those outside of the city of Portland but still within Multnomah
County were considerably lower. Residents of Clackamas County had the lowest level of accessibility, except for MILES, suggesting a much lower density of potential activity opportunities.
Individual accessibihty patterns in the Portland area can be best visualized as a continuous surface. Each of the measures was therefore interpolated from points
(representing individual’s home locations) using inverse-distance weighting to a surface with a resolution of 52,104 cells o f250,000 square feet each. In order to allow meaningful comparisons all accessibihty values were again standardized to a mean of
100, so that heights show deviations from this average accessibihty value (the vertical exaggeration is 20 times). However, because of the nature of space-time accessibihty measures, these surfaces cannot be interpreted in the same manner as those for conventional accessibihty measures. Rather than the elevation of a point on the surface representing the accessibihty of that location, it represents the accessibihty of an individual living at a particular location, though of course his or her access to services may derive from travel throughout the city. Such surfaces therefore reveal variations among individual attributes.
Though each of the measures is different, the surfaces all show a strikingly similar pattern, as seen by the surface for WEIGHTED AREA (Figure 4.1). Accessibility values vary greatly within short distances, and although individuals with high access to employment and services can be identified, there is no clear pattern of accessibihty variation by distance from the CBD or other central locations. Neither are the urban growth boundary nor the role of terrain, rivers, or freeways in constraining accessibility
76 evident from these surfaces. The lack of high access on the part of individuals living quite
close to the Portland CBD is quite striking, given its dominance within the distribution of
potential activity opportunities and its importance as a job and activity center. Some of
the highest accessibility values are actually observable among individuals living on the
periphery of the study area. Many of the same individuals can be identified in each of the
surfaces, showing that these people’s mobihty and access remains quite similar regardless
of which measure is used. Some variations are nonetheless apparent, and can provide the
basis for making statements about the influence of distance and location on each of the
five measures.
The most striking difference between the surfaces can be seen when comparing
the WEIGHTED AREA and TIMED AREA surfaces. Although the latter measure
evaluates potential activity opportunities in the same fashion, the elimination of nighttime
activities generally produces lower values. Many individuals will therefore have a lower
TIMED AREA accessibility than for WEIGHTED AREA. However, the TIMED AREA
surface shows higher peaks than WEIGHTED AREA, representing more extreme
variations from the mean access level than found for WEIGHTED AREA (or any of the
other measures). This indicates that while eliminating evening and overnight activities results in a lower average accessibility, it does not do so for all individuals. Those people who engaged in activities exclusively during daytime hours have unchanged access as
evaluated by the TIMED AREA measure, while others now have a lower accessibility.
The exclusion of nighttime activities therefore heightens differences among individuals and produces a surface with more extreme deviations from the mean. In fact, many of the individuals with above average accessibility according to the WEIGHTED AREA
77 measure appear to have risen even higher above average with the TIMED AREA
measure. Given that individuals with severe daytime mobility or time constraints will
need to carry out any discretionary activities in the evening or at night, this should not be
surprising.
Portland CBD
Figure 4.1: Weighted opportunity individual accessibility surface for Portland study area
4.3 Accessibility and Distance
The relationship between distance and accessibility can be more clearly visualized by charting individual levels of access according to distance from the Portland CBD. As shown in the previous chapter, downtown Portland is an important concentration of
78 employment and potential activities. This area actually contains about 20% of all employment within the Metro urban growth boundary and serves as a hub for the freeway network as well as transit and light rail systems (Metro, 1997). For these reasons the
CBD can be expected to be very important to individual’s mobility as well as access to activities, very hkely to a degree not found in most other middle sized American metropolitan areas. Given the unusual importance of this area, Portland can be considered an ideal city for assessing the importance of distance to the CBD, as the monocentric model predicts that accessibility should decline with distance from this location. Because this model also assumes that the effects of congestion should be distributed evenly throughout the city, congestion should also be plotted by distance from the CBD. This can be shown by comparing accessibility values computed under exclusively free flow conditions with that identified for this research using both free flow and congested travel times (as appropriate for the time of day). Doing so will provide a fiurther test of the importance of monocentric logic on accessibility as well as evaluating the utility of including congested travel times in the analysis.
Because of the slow speed of the Arc View GIS algorithm, a subsample of 200 individuals was randomly selected for visualizing accessibility under variable and exclusively free flow conditions (Weber and Kwan, 2001). The composition of this subsample is similar to that of the full 755 person sample, with 50.5% being male and almost all individuals being white. Because accessibility had already been measured for these individuals using variable travel times (free flow or congested, as appropriate for time of day), it was reevaluated using only free flow speeds. This recomputation of the five accessibility measures for 200 individuals took approximately two months.
79 4.3.1 Accessibility Patterns and the Monocentric Model
The resulting relationship between distance from the CBD and the home locations of individuals can be seen in Figure 4.2 and Table 4.2. Distance to the CBD was measured by driving time (in minutes) through the street network. Accessibility values were computed using only free flow times and then averaged within five minute driving time intervals to show how access to employment and services varied according to individual’s home locations. As before, accessibility values have been standardized to a mean of 100 in order to allow meaningful comparison between the measures. Values above 100 represent above average access for individuals living at a particular location, while values below 100 represent below average accessibility.
As can be seen, there is no consistent relationship between distance and accessibility. With the exception of the TIMED AREA measure, each of the five accessibility measures shows a similar pattern, with people living within 20 minutes of the CBD possessing access levels that are close to average. Individuals hving at 20 to 25 minutes driving time of the CBD possess the highest accessibility (136% of the citywide average), with values falling beyond that distance until the periphery of the study area is reached, where individuals possess an average accessibihty that is only about 40% of the citywide mean. While this latter finding is consistent with the monocentric expectation that accessibility should decrease with distance from the CBD, the peak values at 20 to 25 minutes is not. Instead, suburban areas of Portland tend to contain the home locations of individuals with the highest accessibility. Whether this is due to the behavior or characteristics of those individuals, reflecting higher mobility or fewer time constraints, or the attributes of the surrounding urban environment, perhaps because of the presence
80 of uncongested freeways or major opportunity concentrations, cannot be directly determined from this chart. There are clearly a variety of explanations that require further analysis, and these will be addressed in Chapters Five and Six.
140 g '
Miles Opportunities Area Weighted Area ® Timed Area
Figure 4.2: Average individual accessibility by distance from the Portland Central
Business District (CBD)
As noted, the TIMED AREA shows a considerably different pattern. Individuals closest to the CBD now have below average access, while those living ten to 15 minutes from downtown having the highest above average values. This measure shows a suburban decrease in access rather than the increase shown by the other four measures.
81 Mies Opportunities fifea WeigitedABa Timed Area Axessitniiity Aooessibility /teoessit)ility AxessitDiiity Axessitsiiity N ReeRow Raducdon FreeRow Réduction ReeRow Reduction ReeRow Reduction ReeRow Reduction Avg Personal Axessitiiiity 200 3580.45 -28.52 15718.93 -22.12 18489.19 -27.22 19604.13 -26.52 6690.16 -16,15
Standarcfzed Acoessiinility 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
Driving time from CBD less tfian 5 minutes 13 10&29 111.75 9503 81.72 10420 102.65 103.22 100.88 84.17 101.30 5-10 min 44 93.07 92.03 104.06 88.69 9032 97.95 97.59 96.82 90.90 113.82 10-15 min 49 97.32 84.23 106.25 85.41 100.79 86.18 101.44 85.63 128.89 56.86 1520 min 21 103.56 110.33 97.80 105.62 100.25 108.68 100.71 107.63 116.45 144.44 2525 min 33 13&00 101.57 120.12 109.45 128.21 95.02 127.27 97.17 89.20 82.69 2530 min 18 117.77 120.52 110.44 133.86 119.45125.11 117.60 126.70 75.72 126.37 S)00 3535 min 17 44.83 145.02 41.59 16&21 4475 148.07 4395 150.38 74.06 226.89 more ffian 35 minutes 5 42.03 24.55 3679 32.83 4434 33.51 4296 33.48 115.78 34.26
Note: Bdd indicates dfferenœs are significant at p < 0.05; Italics indicates differences are significant at p < 0.01
Table 4.2: Average accessibility under monocentric expectations though at the farthest distances from the CBD TIMED AREA shows an increase. The accessibility of individuals living on the periphery of the city is now above average.
These differences are striking because this measure is the same as WEIGHTED AREA except that evening and nighttime activities are excluded (and the absolute values o f the
TIMED AREA measure are everywhere less than WEIGHTED AREA). Higher values of TIMED AREA therefore show areas that had the least reduction in access when limited facility hours were imposed. As the areas with the highest values are ten to 15 minutes from downtown at the edge o f the city, individuals living in these locations engaged in the lowest proportion of nighttime activities. So while suburbanites living 20 to 25 minutes driving time from the CBD had above average accessibility with the
WEIGHTED AREA measure, much of this access stemmed from nighttime activities.
When these were not counted, accessibihty fell to well below average values.
Incorporating facility hours therefore shows that the time of day is very important to evaluating accessibility, as it allows an important component of activity behavior to be included in the analysis. This behavior (and time of day variations) is invisible within the monocentric model.
The importance of congestion to accessibility, and its relationship to distance from the CBD, can also be visualized in the same manner (Figure 4.3). In this case the chart shows the percentage reduction in accessibility when congested speeds were applied during the evening peak period. These values have again been standardized to a mean of 100 in order to allow comparison between the five accessibility measures.
Values above 100 represent above average reduction in accessibility (so that congestion has had a greater impact in this area), while those below 100 indicate a less than average
83 reduction (meaning that these individual’s accessibihties under variable travel time
conditions remains similar to that under free flow conditions).
250 C -O- g o200 3 "O 150 ■o Miles Opportunities Area • Weighted Area Timed Area Figure 4.3: Average percent reduction in individual accessibility by distance from the Portland Central Business District (CBD) Once again there is no clear pattern of variation by distance, and again the TIMED AREA measure shows a different pattern from the other four measures. While accessibihty has been reduced at all locations, the four similar measures show the effects of congestion are consistent with increasing distance from the CBD, but then increase towards a peak reduction at 30 to 35 minutes driving time. Outer suburban areas therefore contain the individuals whose mobility has been most heavily reduced by 84 congestion. These include some of the same areas with well above average access under free flow conditions, so the advantages of living in these areas (or traveling during peak hours) must be offset by the existence of greater congestion. However, as before, it is not possible to directly determine whether these reductions exist because of behavior or the influence of the urban environment. These individuals may be engaging in a high proportion of their travel during evening rush hour, or they may be attempting to travel on extremely congested roadways. Those people living farther towards the periphery of tlie city actually possess the least reduction (only 25% o f the citywide average), indicating that they either engage in travel during non-peak hours or are able to move around on relatively uncongested roadways. However, it must be remembered that these individuals had the least accessibility to begin with, so that the existence of congestion appears to be reducing variations in accessibility by location. This pattern is altered somewhat by the TIMED AREA measure. As before, lower values for this measure indicate greater reduction in access as a result of excluding nighttime activities. Individuals living at ten to 15 minutes from the CBD and at the periphery of the city again show the smallest reductions. While this is to be expected because it has already been established that they tend to carry out activities during daytime hours, it is also striking in that it is during the late afternoon hours (when these individuals are traveling and activities are still available) that congestion has been applied. This shows that these individuals are therefore not only carrying out activities during the daytime but that they are also avoiding evening congestion. In contrast, people living at 15 to 20 minutes driving time experienced similarly high levels of TIMED AREA accessibility under free flow conditions, but have above average 85 reductions because of congestion. These individuals therefore engage in daytime travel as well, but face rush hour traffic (or congested roadways) to a greater degree than those hving closer to the CBD. Even worse, those individuals hving at 30 to 35 minutes firom downtown possessed below average TIMED AREA accessibihty with fi*ee flow speeds but also suffer fi'om well above average reductions. So while they tend to carry out a higher proportion of activities during evening or night hours, these activities and their associated travel overlaps to a high degree with the evening peak rush hour (and hkely also makes use o f highly congested streets or fi-eeways), further disadvantaging them. Accessibihty and congestion are therefore highly locationally sensitive, and also depend to a great extent on subtle variations in travel behavior and the urban environment. Again, these characteristics are not present in the monocentric model. Despite the importance of the Portland CBD as a center with a very high concentration of potential opportunities, variations in accessibihty and the effects of congestion do not directly support the logic of the monocentric model. Instead, individuals hving in suburban areas possess the highest access to employment and service locations, as well as suffering the most firom congestion. This accessibility peak at a considerable distance fi'om the CBD can be related to other work that has evaluated the importance of distance within contemporary American cities (Hoch and Waddell, 1993; Waddell, Berry, and Hoch, 1993). These have found generally weak or nonexistent relationships between housing values or rents and distance fi'om a range of urban features (including shopping centers, fireeways, airports, and employment centers). When distance relationships can be found, these relationships may take the form of an ‘inverse U’, suggesting that there is a penalty or disadvantage to living too close to an otherwise 86 desirable feature. For that reason rents or housing values may peak at a distance of several miles rather than adjacent to the center. Although the idea of an ‘inverse U’ appears to have utility for explaining accessibility in Portland (except perhaps for the TIMED AREA measure), it must he remembered that considerable distances are present from the CBD to the home locations of individuals with the highest access to services. Though there may indeed be a disadvantage to living closer to the CBD, this would more likely be related to housing conditions and neighborhood desirability, perceptions of school quality, and the presence of amenities in suburban areas. These strongly favor the importance of location over that of distance, and suggest that the notions of postmodern urbanism may be useful in explaining accessibility variations according to location rather than distance. It can also be argued that these individuals with high accessibility could be carrying out a high proportion of their activities in the CBD (and so have access to the high concentration present there), but this does not directly support the monocentric model, and in fact points out that commuting patterns may differ considerably based on residential location. This is again much more consistent with postmodern urbanism, which highlights residential and workplace segregation by race, age, class, occupation, and other lifestyle attributes, rather than attempting to render them invisible. These results also run counter to Abbott's (1983) and Marshall’s (2000) discussions of Portland, which stress the importance of the CBD area and the inner residential neighborhoods. While these places may indeed be vibrant and offer a range of services and employment opportunities, it is clear that the suburban realm of Portland is a key to understanding both accessibility as well as the effects of congestion. Because these areas contain 87 numerous retail or employment concentrations, as well as possessing a well developed freeway network, it can be argued that the polycentric model, reflecting late 20th century American cities, is better able to reveal the importance of distance in these areas. This model will therefore be evaluated as well. 4.3.2 Accessibility Patterns and the Polycentric Model The relationship between distance and individual accessibility in Portland can also be assessed within the context of the polycentric model. In this case accessibility would again be expected to decline with distance, but now from multiple centers. Although a substantial literature exists regarding the identification of suburban employment centers (Gordon, Richardson, and Wong, 1986; McDonald, 1987; Heikkila, et al, 1989; McDonald and McMillen, 1990; Giuliano and Small, 1991; Waddell and Shukla, 1993; McDonald and Prather, 1994; Small and Song, 1994; Forstall and Greene, 1997; Bogart and Ferry, 1999), for this research the 2040 Growth Plan for the Portland area was used to identify major suburban centers (Metro, 1997). This plan designates a hierarchy of centers throughout the metropolitan area, including the Portland CBD, regional centers, and smaller town centers (Figure 4.4). Regional centers are intended to be large mixed land use developments, representing a concentration of employment, retail, and recreational opportunities (Metro, 1997). When fully developed, they will have higher densities than surrounding areas, and will strongly connected to one another and to the Portland CBD by highway and transit (including light rail) links. They are to be pedestrian firiendly walking environments, and should possess a jobs/housing balance that allows a considerable population to live within 88 easy walking or biking distance of their workplace (as well as shopping and entertainment facilities). These areas should therefore possess a high accessibility within the Portland area (though not as high as the CBD). The emphasis on non-auto transport is expected to lead to non-single occupancy vehicle (SOV) travel to and within these areas making up only 45 to 55% of all trips in 2040 (Metro, 1997). By comparison, non- SOV trips to and within the CBD should make up 60 to 70% of all trips in 2040. The 11 regional centers include the downtowns of several major subiurbs, including Milwaukie, Gresham, Hillsboro, Oregon City, Beaverton, and Vancouver, Washington. Others are large suburban employment and retail areas, including the Washington Square Center, Clackamas Town Center, the 1-84/205 center, and the Vancouver Mall. The remaining center, Salmon Creek in Washington, is much smaller. The development of these regional centers is an ongoing project (Metro, 1997), but at the present time many are important retail or office concentrations and so are already significant areas within Portland. To show the importance of distance to these centers, accessibility values produced using only firee flow travel times were again plotted by distance, but this time it was by the distance of individual’s home locations to the nearest of the 11 regional centers or the CBD. (Figure 4.5 and Table 4.3). Because these centers are well distributed throughout the study area (and in Clark County, Washington) all 200 individuals live within a fairly short driving time firom the nearest center. As a result, there are only four driving time intervals, but some interesting patterns are nonetheless apparent. Most significantly, the average accessibility of individuals living closest to the nearest center is the highest, with values tending to fall with increasing distance away firom each of the centers. The 89 TIMED AREA measure again shows a somewhat different pattern, but in this case shows even more strongly the higher accessibilities of those living adjacent to the centers. Due to the nature of this measure, this must be because these people tend to engage in a higher proportion of daytime activities. Individuals living from five to 15 minutes now show a lower level of access, though those living farthest from the centers show very little reduction when limited faciUty hours are imposed. The support these results offers for the polycentric models is therefore undermined by the importance of individual and household behavior, with this model actually allows no consideration of. ■ Portland CBD • Major centers /\/ Freeways 0 0 Rivers (iTTiiTI Metro Boundary 0 6 12 18 24 Miles Figure 4.4: Major polycentric centers in Portland study area 90 When congestion is applied the patterns are reversed. Reductions in accessibihty are almost constant (and at the city wide average) except at the farthest distances, where extreme reductions are present for all five measures (Figure 4.6). This reduction is most dramatic for TIMED AREA, which is surprising because it was at these distances where TIMED AREA showed the most similarity with other measures. This indicates that while these individuals are engaging in daytime travel these trips tend to be concentrated 160 9 . \ ’co 140 (O \ Q) Ü o (Ü 120 \ ■! •O Miles Opportunities Area Weighted Area Timed Area Figure 4.5: Average individual accessibility by distance firom twelve regional centers in the Portland metro area 91 Mies Opportunities Area Weighted Area Timed Area Aooessitxiity Axessibility Aooessibiiity Aooessibiiity Aooessibiiity N FreeFkw Raduction FreeFkw RaducBon FreeFkw FteduoHon FreeFkw Fteducüon FreeFkw Radudion Avg Personal Aœessilâiity 200 3580.45 -28.52 15718.98 -22.12 18489.19 -27.22 19804.13 -26.52 6690.16 -16.15 StandardzBd Axessibility 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Diving distance from centers Less ttian 5 minutes 45 116.63 92.40 106.83 86l21 115.49 88.21 113.97 88.18 153.89 98.97 5-10 m'n 121 93.32 102.21 97.55 100.15 94.49 10274 94.86 10291 82.27 98.33 10-15 min 28 107.41 89.37 105.14 10258 103.84 93.46 104.78 9269 95.82 72.56 geater*Tan20min 6 75.39 203.55 74.09 228.01 77.04 205.26 76.68 20016 72.78 322.19 g Note; Bold indicates dHferenœs are significant at p < 0.05; Italics indicates differences are significant at p < 0.01 Table 4.3: Average accessibility under polycentric expectations in the late afternoon hours, and so are subject to rush hour traffic. A minor time of day difference therefore has the potential to produce major differences in individual accessibility, which in turn are distributed differently by area. Again, this runs counter to the expectations of the polycentric model. As with the monocentric results, it is also not possible to isolate behavior firom the characteristics of the surrounding urban environment, so there are a variety of reasons why accessibility may decline with distance from the centers. To the extent that individuals do not work, shop, or carry out other activities in these centers, or select housing on the basis of distance from these c 350 1 ! i 300 1 1 1 g 250 1 / ■■ 1 "S 2 0 0 _ ! / X I ISl 1 / ^ I 150 1 1 1 I 100 1 1 1 S 1 1 1 I" 5<10 10<15 >20mln Driving time from centers Miles Opportunities Area Weighted Area ® Timed Area Figure 4.6: Average percent reduction in individual accessibility by distance from twelve regional centers in the Portland metro area 93 centers, these patterns may be purely coincidental. However, they do strongly suggest the need for additional testing of the original sample of 755 individuals selected from the travel/activity diary data set to identify the importance of distance from each home location to the regional centers and the Portland CBD. This will be carried out in the next section. 4.4 Explaining Individual Accessibility Patterns in Portland The ability of distance to explain individual accessibility patterns was tested by carrying out a multivariate regression analysis. The goal was to use distance to multiple centers, as well as socioeconomic characteristics, to help explain the observed accessibihty variations within Portland. A range o f variables was used in this analysis. One set of variables measured distance from each individual's home location through the street network using free flow travel times to the CBD and to each of 11 major regional centers within the Portland metropolitan area. This allows evaluation of the assumption that accessibihty should decline with distance from the CBD or, within a polycentric city, major suburban centers. While some relationships with distance were observed in the previous section, this was for a subsample of data that had been aggregated into a set of distance intervals, and so showed only the relationship between access and distance to the nearest center. A more sophisticated analysis of distance is therefore warranted. A range of socioeconomic variables was also included to help isolate important characteristics of individuals and their households that can be expected to be important influences on mobility, and therefore accessibihty (Villoria, 1989, England, 1993; Blumen, 1994; Kwan, 1998). These include gender, age, race, whether the individual is 94 head of their household, relation to the head of their household, household size, number of children in the household, the individual's employment status (not working, part time, full time), number of hours worked per week, and household income (within $5,000 intervals). Each of these variables can be expected to have an influence on accessibility. For example, accessibility can be expected to be less for females, the very young and elderly, and non-whites due to their lesser mobility and the importance of household responsibihties for women. Increasing household size and especially the number of children will likely reflect those responsibihties, leading to a decrease in accessibihty with larger famihes. The monocentric and polycentric models assume one worker in each household (Hanson and Pratt, 1988). Although these models ignore the characteristics of individuals, it may be that one’s status and role within a household can make a difference to accessibihty as those who are not the head of the household (assumed to be women within these urban models) may be less likely to have access to a vehicle and more likely to be constrained by household responsibihties. The role of employment status in accessibihty is in some ways counterintuitive. Although earning an income through employment (or living within a household with at least one income earner) is obviously essential to the support of the household and continued access and operation of a vehicle, and therefore accessibihty, it can be expected that accessibihty should decrease with an increasing number of hours worked or a shift to a more intensive employment status. This is due to the nature of space-time accessibihty measures, which show that employment produces strong constraints on time and mobility and therefore limit the opportunity to engage in flexible activities or discretionary travel. An unemployed 95 individual may therefore have very high accessibility (Villoria, 1989). While the monocentric model implies that lower income groups will have high accessibihty because of their central location (McLafferty, 1982), the nature of space-time measures means that income will have uncertain effects on accessibihty. Accessibihty has been shown to increase with income (Villoria, 1989), but it may also be that income serves as a surrogate for the amount of time spent at work (at least in one-person households), so that increases in earnings may be related to a decline in accessibihty. These variables were used to predict each of the five accessibihty measures using forward stepwise regression with a tolerance o f0.0001. As before, the accessibihty values were standardized to a mean of 100. The results of the best fitting models are shown in Table 4.4. The use of stepwise regression eliminates the likelihood of multicollinearity among the independent variables, and the possibility of non-linear relationships was taken into account by examining scatterplots of the relationships between the five accessibihty measures and the explanatory variables. No strong indication of non-linearity was apparent, so the following discussion is based on the assumption that all relationships can be adequately modeled linearly in this and following chapters. In addition to the significant parameter estimates, F tests for each of the models indicate that they are all significant at the 99% confidence level. The small values indicate that the independent variables account for only a small portion of the variation in standardized accessibihties, indicating that there is a high degree of variation among individuals. However, several important relationships exist between these variables and access. As can be seen, only three of the distance variables were present in the final models, representing distance to the Portland CBD, the Clackamas Town Center, and the 96 Dependent Independent Standardized Significance Adjusted Variable Variable Coefficient Std Error Coefficient Level Rsquare Rsquare MILES Constant 201.11 16.41 0.000 0.055 0.051 Distance to Clackamas Center -1.63 0.35 -0.17 0.000 Hours worked -1.25 0.28 -0.16 0.000 Household size -5.90 2.74 -0.08 0.031 OPPORTUNITIES Constant 183.19 10.70 0.000 0.094 0.091 Distance to Portland CBD -1.19 0.41 -0.13 0.004 Distance to 1-84/205 Center -1.37 0.37 -0.17 0.000 Hours worked -0.93 0.22 -0.15 0.000 AREA Constant 175.40 11.58 0.000 0.066 0.063 Distance to 1-84/205 Center -1.94 0.31 -0.22 0.000 Hours worked -1.02 0.25 -0.14 0.000 WEIGHTED VO Constant 174.08 11.33 0.000 0.067 0.064 AREA Distance to 1-84/205 Center -1.91 0.31 -0.22 0.000 Hours worked -1.00 0.24 -0.14 0.000 TIMED AREA Constant 281.89 22.27 0.000 0.103 0.1 Distance to Clackamas Center -1.83 0.57 -0.11 0.001 Hours worked -3.90 0.44 -0.31 0.000 Table 4.4: Regression results for individual accessibility 1-84/205 center. Given the strongly monocentric distribution of potential activities, the importance of the Portland CBD was surprisingly small as of the five measures, only the number of activities accessible (OPPORTUNITIES) declined with distance fi'om this location. This decline in the number of parcels available could be due to a reduction in the density of parcels or mobility towards the periphery of the city, or to a greater level of time constraints with increasing distance from the CBD. The first interpretation is the most consistent with monocentric model and suggests that individuals living nearer the CBD do in fact possess higher accessibility. However, this conclusion must be modified by the significance of distance to other locations. OPPORTUNITIES, AREA, and WEIGHTED AREA all decline with distance from the 1-84/205 regional center. The relationship with OPPORTUNITIES indicates that the density of activities declines both with distance from the CBD as well as the 1-84/205 junction. This again is consistent with the monocentric model, though it is clear that additional centers must be incorporated to model accessibility in Portland. For AREA and WEIGHTED AREA the results indicate that the size of activity parcels also declines with distance from a central location, though of course this is not the CBD. Given the peripheral location of the CBD within Portland, the importance of the 1-84/205 center likely reflects its greater geographic centrality within the city, especially on the highway network. Travel times may favor this location over the CBD as well. The logic of the monocentric and polycentric models are weakly supported, though again it is not possible to directly identify whether this is because of the distribution of activities, mobihty, or travel behavior. 98 Distance from the Clackamas Town Center is important for MILES and TIMED AREA. For MILES this shows that mobility declines with distance from this location. Individuals living farther from the Clackamas Town Center tend to be able to move over a smaller street mileage than those living closer. This could be due to differences in mobihty or behavior. Likewise, the area of potential daytime activities tends to decrease with distance from this center. The fact that the TIMED AREA measure is linked to this center while the other two area measures respond to the 1-84/205 center clearly shows the importance of travel behavior. The Clackamas Town Center has a greater influence on daytime activities while the 1-84/205 center has a greater influence on all activities. This shows that incorporating temporal constraints on the availability of activities is important not just for the way it reduces accessibility but because it alters the geography of this accessibility. In this case, eliminating nighttime activities from consideration changes the center from which distance must be measured to explain accessibihty. As the WEIGHTED AREA and TIMED AREA measures are identical except for the absence of nighttime activities, this difference must be due to travel behavior (but not mobility). These geographic and time of day variations are not allowed to exist within the monocentric and polycentric models, but are clearly vital to individuals. Rather than distance determining access or behavior, the reverse actually appears to be the case, with the activity and travel choices of individuals strongly influencing the importance of distance in Portland. While the importance of centrality is apparent, it is actually proximity to several centers, and these vary according to accessibility measure. Despite the widespread distribution of individuals in the sample there are no outlying centers (particularly in 99 Gresham or eastern Washington County) that have any influence on accessibility. The area of greatest activity density is not necessarily the area of greatest mobility, the area of the largest activities, or of the greatest daytime travel. This strongly suggests that these centers have little real influence on accessibility, and are simply surrogates for centrality within Portland. It is likely that a number o f other points could have been picked from along the freeway network within this area to produce similar results, though of course it must be remembered that these centers represent important and well known concentrations of retail and employment activities within Portland. So while centrality has a role in explaining accessibihty, it is highly sensitive to automobile mobility, travel behavior, density of potential activities, and the area of those activities. Each of these factors involved in accessibility appears to respond differently to distance, so that it is not possible to point unambiguously to a single center as being important to all elements of accessibility. The importance of centrahty within the city also likely changes considerably over the course of the day, depending on congestion and changing travel and activity behavior patterns. Because distances were measured to the centers using uncongested driving speeds, this does not necessarily reflect their proximity at certain times of the day. This may help explain the unevenness of the importance of centrality to accessibility. So while the results superficially support the monocentric and polycentric models, there is actually little evidence for these models, or for the importance of distance in general. 100 4.4.1 Socioeconomic Characteristics and Accessibility Two socioeconomic variables were also useful. The results with these variables are striking for the importance of the number of hours worked per week and household size to accessibility. Accessibility declines with an increasing number of hours worked for all five accessibility measures, the only explanatory variable which does so, while the number of MILES available declines with increasing household size. Both results are consistent with other work on space-time measures and time geography in general. The importance of hours worked shows clearly that employment status is fundamental to accessibility (and in fact when hours worked was excluded the employment status variable did appear in the model). Given the additional time constraints imposed by more time spent working, it is not surprising that accessibility should tend to decrease with additional hours spent working. It is particularly interesting that this effect holds true for mobility (MILES) as well as access to both a high density (OPPORTUNITIES) and area (AREA and WEIGHTED AREA) of potential activities. Increasing hours spent working is therefore a universal reducer of accessibility, suggesting that individuals have little ability to overcome a reduction in accessibility by altering their behavior. That increasing the hours worked would reduce the daytime accessibility (as evaluated by TIMED AREA) is to be expected, as it is most likely to be daytime hours which are taken up with additional work hours. However, while employment status and hours worked reduce accessibility, it must also be remembered that low accessibility could hinder the search for new employment and therefore limit a change in employment status (as the 101 spatial mismatch literature points out). Accessibility and employment status can therefore be expected to have a mutually reinforcing effect rather than one of unidirectional causality. That accessibility also decreases with increasing household size is also to be expected, as many explanations for travel behavior highlight the constraints of child care responsibilities, whether by having to spend time at home caring for children or providing transportation to and from school or other activities. Such responsibilities appear to be resulting in lower accessibilities for individuals in Portland. However, it is interesting that while increasing household size is associated with a reduction in mobility (and therefore the number of MILES reachable), it is not associated with a reduction in the density or area of activities reachable. This indicates that reductions in mobihty are not necessarily associated with a reduction in the number of activities to which individuals can gain access. Whether this is because individuals with larger households tend to live close to areas with a greater density and size of activities (such as those living near the Clackamas town center) and therefore see no significant reduction in access to those activities on the periphery of their PPAs, or whether these individuals are altering their travel and activity behavior patterns so as to maximize what they have access to despite their low mobility, remains unclear at this point. If it is the latter possibility than it indicates a crucial difference in how household size and hours worked per week affect accessibihty. While individuals may be able to offset a reduction in mobility resulting from heavy childcare responsibilities and so retain access to desired services, individuals working longer hours may have no such options. This is likely because the time constraints of working very long hours offer less temporal flexibility than the 102 responsibilities involved in caring for larger households (such as child care, engaging in frequent trips to pick up or drop people off, or making more frequent and varied shopping trips). It is also interesting that status within the household was not significantly related to accessibility, so that all members of a large household appear to suffer the decrease, not just the heads o f the households. This is also significant in terms of identifying strategies that may be adopted to increase accessibihty (or overcome the threat of lower accessibihty following changes in hours worked or household size). The absence of gender in the regression results is surprising (and especially so since separate ANOVA testing did reveal significant differences in TIMED AREA accessibihty for men and women), given the results of other space-time research that has shown significant differences between the accessibihties of men and women hving within the same households (Villoria, 1989; Kwan, 1998, 1999a). It is also interesting that race, age and income were not significantly related to accessibihty. The absence of race is undoubtedly due to the racial homogeneity of the sample population, but age and income have been commonly found to be related to travel behavior. These results may be due to the characteristics of the Portland sample used in this research (for example, the use of automobile travel may have eliminated gender, age, and racial differences present in the full population), but it may also be that the effects o f gender or race on accessibihty are already captured in the models by the use of hours worked or household size (due to additional responsibihties for females) and so carmot be directly isolated. Other methodologies, such as path modeling (McLafferty and Preston, 1992) or structural equations (Kwan, 1999b) may be more useful for identifying the contributions of gender, race, or other personal characteristics on individual accessibihty. However, the 103 importance of household size and hours worked does show that individuals are not homogenous within the sample, and that interpersonal or household differences do make a difference to accessibihty, further undermining the monocentric and polycentric models. These are important findings because they emphasize that individual and household characteristics are very important to accessibihty, and carmot be assumed away without consequence, as in the monocentric and polycentric models. The significance of these interpersonal and household differences also highlights the need to examine location fi'om a different perspective, one that is sensitive to such variations. 4.5 Discussion and Conclusions While the Portland CBD is clearly a major employment and service center, as well as a multimodal transport hub, this dominance is not reflected in observed accessibihty patterns. Rather than individual access to employment or other services declining away firom the CBD, there is actually a distinct peak at some distance firom the CBD for each of the five measures, at least for a subsample of data. Individuals hving at this location have the highest accessibihty of those within the Portland study area. When limited business hours are incorporated accessibihty patterns become even more complex, with individuals hving on the periphery of the study area also now possessing well above average accessibihty due to their travel behavior. The effects of congestion are most severe in peripheral subiurban areas, but the inclusion o f limited facihty hours shows that the interaction between congestion, business hours, and individual travel or activity patterns vary considerably by location. Both location within the city and activity behavior are therefore important to individual accessibihty, and neither fit the 104 expectations of the monocentric model. The case for the polycentric model is stronger, but there remain inconsistencies. The importance of the time of day and travel behavior raises the question of exactly why proximity to centers appears to be related to accessibility. The effects of congestion also do not fit polycentric expectations, as it shows that these effects will be greatest at farthest distances, rather than evenly distributed. The regression testing performed with the full sample of individuals further demonstrates the limits of relying on distance as an explanation for accessibility patterns or human behavior within cities. Not only was distance to the CBD and major regional centers within Portland of limited explanatory ability, but the significance of these centers fluctuated according to which measure was being tested. Time of day variations in carrying out activities and travel appear to account for some of these variations, leading to the conclusion that travel behavior is strongly influencing not only accessibility but also the importance of distance within Portland. This reverses the assumptions of the monocentric and polycentric models that distance determines accessibility and further emphasizes the importance of time to accessibility. Because these urban models are static and timeless, as well as ignoring characteristics of individuals and households that are of fundamental importance to accessibility, their usefulness is highly questionable. These findings should also be evaluated in the context of the urban growth boundary and the 2040 growth plan. Because an important goal of the growth boundary is to prevent low-density sprawl and help direct attention towards infill developments and increasing residential and commercial densities (Metro, 1997), it may have significant 105 consequences on accessibility. The presence of the growth boundary may in fact be reducing variations in accessibility by confining peripheral development. This is especially the case because this development would likely favor certain areas, such as along US 26 in eastern Washington County, producing greater heterogeneity than currently exists within the boundary. Comparisons with other metropolitan areas that have not attempted to limit growth in this fashion would be very interesting for helping to reveal the significance of a growth boundary on accessibihty. Given the debate over urban sprawl and increasing support for growth controls, the potential impacts of limiting urban growth or influencing land uses should be studied so that they do not reduce access to employment or services or exacerbate intraurban accessibility variations. Although only two of the 11 regional centers were of any significance in explaining accessibihty in Portland, the regional centers identified in the 2040 growth plan are in the process of being developed into higher density residential and commercial centers, and are to some degree provisionary. Some, such as the Clackamas Town Center or the Washington Square area (which was not significant in the regression results) are already large and highly visible areas. Others, including the Salmon Creek center in northern Clark County and the Oregon City CBD, would appear to have significantly less promise for future development. Should the 2040 plan succeed in transforming these areas into major employment and activity centers (which some clearly already are) than over the next few decades accessibility patterns may indeed come to reflect proximity to these centers. Metro is therefore attempting to counter the prevailing trend toward decentralized (and sprawling) cities and increasing the importance of distance within the city by establishing major nodes around which a polycentric city can 106 develop. While the results for the 11 regional centers tend to support some (weak) polycentrism in Portland, it is unclear exactly what relationship a well developed polycentric hierarchy will have on actual travel and activity patterns. For example, although Atlanta has provided an excellent example of a polycentric city, actual commuting patterns are not clearly organized into polycentric commutersheds (Fujii and Hartshorn, 1995). There remains a considerable degree of cross commuting, and it is not clear that distance has a causal relationship with commuting behavior. The implicit assumption that every center is functionally equivalent to one another and the CBD is not supported as each has become somewhat specialized with different employment structures. Nor are shopping trips any more clearly organized around individual shopping centers, as each draws customers from the entire metropolitan area (and most shoppers regularly patronize several malls). This suggests “that the entire Atlanta metropolitan area is highly interconnected and that whereas multiple specialized work, shopping, and cultural/entertaimnent centers exist, they do not constitute independent realms, but rather offer many choices for living, work and entertainment that are associated with many overlapping and complicated action spaces for urban residents” (Fujii and Hartshorn, 1995, 705). Attempts at forcing polycentric commutersheds into existence through planning efforts have not so far produced better results (Pickus and Gober, 1988). In Phoenix plarmed ‘urban villages’ have been designated which “contain a mixture of housing types and a variety of employment, shopping, recreation, and educational facilities. The spatial structure of each village consists of a core, gradient, and periphery, all defined in terms of the intensity of land use. The core contains a mixture of the village’s most intense land 107 uses, including office, retail, and high-density residential activity. The gradient is characterized by lower but sufficient land-use intensity to support some neighborhood- level services. The periphery is the outer boundary of the village, containing low-density residential land uses, agriculture, and open space” (Pickus and Gober, 1988, 86). Even though these villages were centered on existing employment or retail centers there was Uttle trace of polycentric commuting or shopping behavior by the residents of an urban village, except for grocery and clothing shopping (explained by the presence of a large regional mall and numerous grocery stores in the roughly 100 square miles of a village). This remained true even when socioeconomic characteristics such as age, gender, income, and length of residence within the village were taken into consideration. The authors see Uttle likelihood of polycentric patterns developing in the future because of the highly dispersed locations for employment, recreation, and medical facilities within the metropolitan area. Another issue is that the Portland 2040 plan makes heavy use of non-auto modes, particularly walking and bicycling within centers and bus and light rail between centers. The effect of the continued development of these centers on individuals who continue to rely on their personal cars (and who will continue to make up the majority of the city’s residents) is therefore less certain. And, as central locations have traditionally been favored by transit systems (Adler, 1986, 1987, 1988; Hodge, 1990, 1995; Ruthheiser, 1996), the increased focusing of the bus and Ught rail transit networks on these 11 centers (and the CBD) could actually diminish the accessibility o f transit users. For these reasons the 2040 growth plan has mixed and uncertain consequences for automobile 108 based mobility and accessibility. The planned collection of a new travel-activity diary data set in 2004 by Metro (Kim, 2001) should allow the evaluation of changes brought about by the 2040 plan, increasing congestion and the extension of the light rail system. This chapter has examined the spatial patterns in individual accessibility to potential employment, retail, medical, educational, and other services within the Portland study area. The patterns revealed by the five accessibility measures do not support the role of distance in the monocentric model, and while they do appear to provide superficial support for the polycentric model, the reasons for the observed spatial patterns actually strongly undermine both models. Distance does not determine accessibility, and in fact the importance of distance appears to be more a result than cause of travel and activity behavior (an issue which will be raised again in Chapter Seven). Travel behavior and especially interactions between activities and time of day are neither invisible nor irrelevant. One striking finding of this chapter is that the time of day that individuals engage in travel is clearly important due to whether services are obtainable and the likelihood of encountering congestion. For these reasons (and in part because of the interactions between them) the time of day has contributed to spatial variations in accessibility. This is not possible within the timeless monocentric and polycentric models. Quite unlike the assumptions of these models, residents in cities such as Portland must schedule their activities within the constraints of their jobs, educational requirements, medical needs, and household responsibilities. Because most daily trips, including considerable discretionary travel, occurs in the afternoon and evening (Kwan, 1999b), these individuals will most likely also encounter spatially uneven traffic congestion and the need to reach certain activities before they close for the day. And 109 time can be expected to be even more important because many travel characteristics, including the temporal scheduling of trips, varies by gender, age and employment status (Blumen, 1994; Kwan, 1999b; Tacken, 1998). Men and women, the elderly, or fidl and part time workers are therefore unlikely to experience employment or household-related time constraints, congestion, or the inconvenience o f limited business hours, in the same way. The results strongly support the need to examine the importance of location within the city in order to reveal the interactions between the characteristics or behavior of individuals and households on one hand, and the urban environment in which those people must work, shop, and carry out other activities on the other. This will be examined in the following chapter using multilevel modeling, an extension of multivariate regression that allows accessibility to be examined at a variety of scales and within multiple spatial contexts. 110 CHAPTERS ACCESSIBILITY AND SCALE IN PORTLAND 5.1 Introduction The previous chapter attempted to explain individual accessibihty by use of distance and socioeconomic characteristics of individuals. While distance to the CBD and major centers did provide some explanation for accessibility variations, there are strong reasons to conclude that distance is of little importance to accessibility, especially when compared to individual time constraints and time of day variations in activity scheduling. These time constraints, and time of day variations, also appear to be linked to the influence of distance on accessibility, so that rather than distance determining accessibility, individual behavior and household characteristics appear to be influencing the extent to which distance is important to individuals (and how it is expressed). But there remains the question of the relationship between accessibility and both behavior and the urban environment. Accessibility variations clearly exist within Portland, but the analysis in the previous chapter was not able to directly determine whether these variations are the result of individual and household variations or due to differences in the distribution of activities or high travel speeds. Higher average 111 accessibilities may be present within a particular area of Portland because individuals there tend to have freedom from time constraints (or tend to carry out travel during uncongested times of the day) and therefore have more freedom to move about, or it could be do the presence of large number of potential opportunities in that part of the city, as well as uncongested roadways. More likely, the observed accessibility variations are due to some combination of these factors, but again, the previous chapter was not able to identify their relative contributions. This chapter wtill attempt to resolve these issues by examining individual accessibility among various areas in Portland at a range of spatial scales. While a number of variables were found to be useful in the previous chapter, the use of these variables in a multivariate regression equation assumes that the entire metropolitan area is the most appropriate scale of analysis. However, the effect of these explanatory variables may be scale dependent and of less importance at more local scales, such as at the neighborhood level. The extent to which the importance of distance and socioeconomic characteristics, which were observed to be useful explanatory variables in the previous chapter, vary among locations and at different spatial scales will provide an indication of the relative importance of behavior and the urban environment. These possibilities will be addressed in this chapter using multilevel modeling for several sets of geographic zones representing political jurisdictions, polycentric commutersheds, and neighborhoods. The results of the models will be compared to identify those variables that are useful for explaining accessibility at which scales, and whether certain variables are independent of scale and influence accessibility at all scales. 112 This is an important question, as there are a range of reasons to expect that accessibility will vary among locations and at different scales. This topic will be discussed in the next section. 5.2 Individual Accessibility and Scale While individual accessibilities were related to both socioeconomic variables and distance in the previous chapter, these relationships were evaluated at the scale of the entire metropolitan area. Individuals living throughout Portland were therefore expected to have the same relationship between (for example) the number of hours worked per week and accessibility, regardless of where they lived and worked. However, given the clearly uneven distribution of potential activity opportunities and accessibility revealed in the last chapter, this assumption may have to be revised. Those living within opportunity-rich areas of Portland may have considerably different relationships between hours worked and accessibility compared to those in areas where fewer opportunities exist (or where mobility is less due to slower driving speeds and congestion). There is therefore reason to expect that the relationships between individual or household characteristics and accessibility may vary depending on the geographic scale at which it is measured. This issue has become an important topic because of interest in using land use to influence travel and activity behavior. Ih contrast to traditional low density, suburban, automobile oriented development, which limits the efficiency and usefulness of non-auto modes while forcing people to travel considerable distances to employment or activity locations, planners who favor Neo-Traditional Development (or the New Urbanism, as it 113 is also called) hold that designing pedestrian and transit friendly urban environments can (among other things) improve accessibility to employment and especially retail activities (Bookout, 1992; Handy, 1992; Boamet and Crane, 2001). Designing residential areas using grid street patterns with shopping opportunities located within the neighborhood is thought to reduce the number of trips required by households, as well as shifting travel away from a dependence on autos (Handy, 1992). Though there are monocentric notions (such as the importance of distance minimization and the invisibility of individual and household characteristics) within these concepts, there is also a great deal of attention given to the characteristics of neighborhoods and activity spaces. While the effects of Neo-Traditional Development, and land use patterns in general, on travel behavior are hotly contested (Ewing, Haliyur, and Page, 1994; Friedman, Gordon, and Peers, 1994; Steiner, 1994; Crane, 1996; Handy, 1996a, 1996b; Boamet and Crane, 2001), there has also been interest in examining accessibility at multiple geographic scales. One approach has been to measure access to neighborhood retail and grocery stores (local or neighborhood accessibility) separately from major employment centers and large shopping centers (regional accessibility) (Handy, 1992, 1993; Handy and Niemeier 1997). This approach recognizes that not all potential activity opportunities are identical in their desirability or the frequency with which they are visited. However, studies in which local and regional access have been measured separately have used conventional aggregate accessibility measures, and so suffer from the limitations discussed in Chapter Two. Although the goal of local/regional accessibility studies is to evaluate the influence of distinct urban forms (traditional grid streets with dispersed retail vs. suburban curvilinear streets and large shopping centers) 114 on accessibility, there is no direct way to evaluate the role of individual behavior on accessibility levels. While two neighbors may share similar accessibilities by virtue of their residence within a Neo-Traditional Development, this does not take into account different activity schedules, time constraints on their mobility or the temporally varying availability of activity opportunities (these were shown in the previous chapter to be important components of individual accessibility). Finally, there is no easy way to combine local and regional access into a single measure, or to determine how one may substitute for the other. The importance of scale to individual accessibility in Portland will therefore be examined to identify whether the observed relationships vary by scale. Multilevel modeling will be used with the explanatory individual and contextual variables found to be useful for the single level regression models in the previous chapter. This methodology is an extension of multivariate regression in which macro (zonal) and micro (individual) level models are estimated separately and then together in a final model, allowing the use of both area and individual information (Paterson and Goldstein, 1991; Jones, 1991a, 1991b; Goldstein, 1994,1998; Jones and Duncan, 1996; Bullen, Jones, and Duncan, 1997; Kreft and de Leeuw, 1998). This method allows the isolation of variations resulting fi'om variability in the population (compositional effects) from those resulting from differences between areas (contextual effects), while also avoiding the risk of ecological fallacies by using disaggregate data (Jones, 1991a). These models produce linear relationships but allow for different areas to have different intercepts (reflecting different average values in different areas) or even different intercepts and slopes (showing different relationships between different areas). 115 Variables can be added to any level to represent characteristics of individuals or areas. The result is that “instead o f reducing the world to one universal equation, there can be different relationships for different places” (Jones, 1991b, 7). The model therefore incorporates local context into general relationships by treating variations as normal rather than error. The use of multilevel modeling also avoids the risk of ecological fallacies by using both areas and individuals, though of course if all individuals within a zone are homogenous no multi-level structure is required. This avoids the problems introduced by the use of zonal data in aggregate accessibility analysis. While most multilevel research uses only two levels (individuals and a set of zones) it can be used with additional levels. Both separate intercepts as well as separate slopes for each zone can potentially be computed, so that areas may have both different average values for the feature being modeled as well as different relationships between this feature and explanatory variables. Multilevel models are usually treated as hierarchical, in which individuals are nested within a zone that may in turn be nested within larger zones, but non-hierarchical levels are also possible (Jones, 1991b; Jones, Gould, and Watt, 1998). Variances are produced for both the individual (showing variation between individuals) and zonal levels (showing variation between areas). Although standard regression techniques could be used to study variations between zones, multilevel modeling produces better results. Calculating a slope for each zone separately would not be as efficient as using multilevel modeling to pool the individual data (Jones, 1991b). Likewise, dummy variables could be used to represent each zone individually but would require creating unwieldy equations if there were more than a few zones. 116 The output and interpretation of multilevel models is similar to that of multivariate regression, with a constant, parameter coefficients, and indicators of goodness of fit. Significance testing is generally of httle importance within multilevel modeling literature, though such measures can be calculated for parameter coefficients, variance at each level, and the entire model. For parameter coefficients a ratio of the estimate to standard error that is greater than two is usually accepted as indicating significance at the traditional 95% confidence level (Jones, 1991b; Coombes and Raybould, 1997). Similar standards apply for the ratio of variance to standard error for each level, showing whether there are significant variations between areas. Likewise, the deviance (equivalent to the —2 log likelihood) for the entire model can be used to show a reduction in model variance firom a null model (one with no explanatory variables). The significance of the difference in model deviances can be evaluated by a chi-square test, though again a rule of thumb that this difference “should be at least twice as large as the difference in the number of estimated parameters” can be used to estimate significant improvements in the level of model fit (Kreft and Leeuw, 1998). However, multilevel research within geography has generally focused on the identification of variance at the individual level or at varying geographic scales, and what effect on this variance (and parameter coefficients) is produced by the introduction of additional variables or scales (for example, Jones and Bullen 1993, 1994; Coombes and Raybould, 1997). For example, if the introduction of explanatory variables to the model reduces zonal variations then the model has a better fit. However, in cases where 117 considerably more variance exists at the individual level (level 1 in most examples) than at the zonal level (level 2) there is likely to be very little geographic variation within the phenomena being studied. Geographic applications of this methodology have so far been limited, but studies of school performance (Jones, 1991b; Coombes and Raybould, 1997), housing markets (Jones, 1991a; Jones and Bullen, 1993,1994; Bullen, Jones and Duncan, 1997), pubUc health (Jones and Duncan, 1995; Duncan, Moon, and Jones, 1996) and voting behavior (Jones and Duncan, 1996; Jones, Gould and Watt, 1998) have been conducted. If applied to the study of accessibility, multilevel techniques could help overcome the limitation of viewing accessibility as a purely personal characteristic or as a purely environmental characteristic. The multilevel approach instead more appropriately treats accessibility as a combination of both, with individual accessibility being mediated or influenced by locations. 5.3 Incorporating Scale and Contextual Influences on Individual Accessibility A crucial step in evaluating the important of scale and geographical context is the selection of appropriate geographic units within (and between) which to evaluate relationships. The majority of multilevel studies have used formal, govemmentally designated geographic areas such as schools (Coombes and Raybould, 1997), electoral wards (Jones and Duncan, 1995; Duncan, Jones, and Moon, 1996), and local government areas (Jones and Bullen, 1993; Jones, Gould, and Watt, 1998). However, functional regionalizations have also been used to capture areas of demographic and economic 118 similarity (Jones, Gould, and Watt, 1998). Given that the multilevel literature in geography has generally heed concerned with national or regional level the use of existing political jurisdictions is reasonable at these scales. Unfortunately, these provide little guidance to selecting geographic units at the intraurban level. Multiple criteria were therefore used in this research to identity a set of zones at a range of scales that can be expected to be of importance for individual accessibility. Rather than using a single set of zones and exploring the effects of adding new variables to models, as is common in most multilevel modeling research, the goal here is to examine relationships between a relatively small set of variables at a range of spatial scales. The use of zonal units clearly raises the possibility of the Modifiable Areal Unit Problem (MAUP), a fundamental problem in geography (Openshaw, 1996). Because this problem exists largely due to the use of zonal schemes it has been argued the continuous representation of variables in the form of surfaces is the best way to avoid scale and resolution issues (Martin, 1996). However, it has also been argued that simply selecting meaningful zones would eliminate the problem (Openshaw, 1996). Not only are zones fundamental to spatial analysis, there are limitations with other representations of data. The use of point data "ignores the important role of a zoning system as a generalization operator that can serve a useful spatial analysis function, as well as add noise. Many geographical patterns of interest have an innate scale dependency to them. Finer resolution data mainly increases flexibility in usage and storage, yet this is not really the problem" (Openshaw, 1996, 66, 67). Surface representations may also be of little help, as "the continuous approximation of discrete information merely reintroduces other sorts of error, inaccuracy, and distortion. Geographical analysis is about the variation between 119 and within places, and places are discrete lumps of space" (Openshaw, 1996,67). According to this perspective the MAUP (and ecological fallacy) exist in geography only because researchers have had little ability to select their zones, so the solution is therefore to allow the user to select an appropriate set of zones. A functional approach would be used by "defining or identifying areal entities that make some sense in a particular substantive context" (Openshaw, 1996, 67). This has hopefully been achieved with the use of the Portland neighborhood and other zonal units, which will be described in the next section. 5.3.1 Identifying spatial scales A number o f spatial scales and contexts can be expected to be important to accessibility within cities. One scale is created by political boundaries, which are likely to be related to socioeconomic composition as well as housing and land use characteristics (Lewis, 1996). For these reasons, several zonal schemes were created to attempt to represent the importance of these units. These boundaries were used to construct the first set of zones, called CITY. Individuals were grouped according to residence within the city of Portland, other Multnomah county jurisdictions, Clackamas county suburbs, unincorporated areas within Clackamas County, Washington County suburbs, and unincorporated areas within Washington County. Because of the need to interpolate socioeconomic and other characteristics into discrete zones in the next chapter, individuals living outside of any municipal limits were grouped within the neighborhood units defined by Metro (which are discussed below). Incorporating county boundaries can also help differentiate areas in terms of mobility, as, for example, eastern 120 Washington County suffers from poor street connectivity and therefore limited mobility (Kloster, 2000). Another zonal scheme was used to test for the importance of location within particular school districts. This scheme (named SCHOOL) groups individuals according to location within the Portland school district, other Multnomah county districts, Clackamas County districts, or Washington County districts. Only those school districts with sampled individuals present in them were used for interpolating socioeconomic and land use characteristics. In addition to political boundaries, economic divisions of Portland should also be recognized. While the use of local labor markets could likely produce important differences, there is no clear guide to the identification of these relatively intangible spaces (Hanson, Kominiak, and Carlin, 1997; Wyly, 1998). An attempt was therefore made to capture contextual variations present within different areas of the city using a polycentric framework. Because the polycentric model assumes cities contain several discrete commutersheds, each centered on a separate center, this was used to isolate fimctional regions within Portland. Commutersheds were created by first measuring the free flow driving time from the centroid of all census block group to the Portland CBD and each of the 11 regional centers (discussed in the previous chapter), and then assigning each block group to the nearest center. The aggregations of block groups by the centers were treated as commutersheds, and were used to classify individuals within the commutershed of the closest regional center or the CBD, which is consistent with the assumptions of distance minimization inherent within the polycentric model. This set of zones was named COMMUTERSHED. Two regional centers located in Clark County, Washington (Salmon Creek and the Vancouver Mall) did not contain any individuals in 121 the study sample and so do not show up in this zonal scheme. Because several of these commutersheds are focused on suburban CBDs, their use could also help capture variation between suburban areas. The use of these areas is also important given the attempt at creating a polycentric city in the 2040 growth plan for Portland. 5.3.2 Representing neighborhoods As noted in Chapter Two, there is considerable support for the idea of using neighborhoods as local contexts for discussing travel behavior and accessibility, but there is httle agreement on how to identify these spatial entities. One approach defined neighborhoods based on the shared spatial familiarity of areas on the part of local residents (Talen, 1999; Aitken, et al, 1993; Aitken and Prosser, 1990). Local schools, shopping centers, or commonly traveled streets are likely to be famihar to residents while other areas will be almost unknown. However, attempts at defining neighborhood boundaries based on the assumption that they would be visible as homogenous areas of high famiharity/experience are inconclusive. Instead, the resulting pattern shows a very uneven surface representing almost constant "discontinuities in the familiarity surface such as between neighborhood parks, shopping centers and schools on the one hand, and residential areas on the other" (Aitken, 1993, 9). Others have defined local areas based solely on individual experiences. Hanson, Kominiak, and Carlin (1997) used a circle centered on each individual's residence (with the radius defined by commuting distance) to define each individual's community. 122 The use of multilevel modeling in this research requires the definition of discrete zones, and so two distinct approaches were used to construct useful neighborhood zones. The first set of zones is based on an interpretation of neighborhood types within Portland developed by Abbott (1983). Abbott distinguished four distinct types of neighborhoods within the city of Portland, beginning with the ‘stopover’ neighborhoods, in the vicinity of the CBD. These originally provided cheap housing for recent arrivals to Portland before they moved on to more desirable neighborhoods. In more recent decades they have been transformed by changing land uses, including fi-eeway construction and urban renewal projects, but still retain a residential component. Beyond the stopover neighborhoods were the ‘everyday’ neighborhoods, large residential areas in eastern Portland and to the northwest of the CBD. These were originally developed in the late 19th and early 20th centuries and tend to consist of high-density single-family homes, and have remained stable communities during the automobile era. Highland neighborhoods are present in a few scattered areas, including the West Hills and Mt. Tabor east o f the CBD. These were developed in the early 20th century and are stable high-income residential enclaves within the city. Finally, beyond the everyday neighborhoods are the postwar automobile era neighborhoods, which include mainly areas outside the city limits of Portland in Mutlnomah, Clackamas, and Washington counties. Neighborhood units corresponding to the areas defined were aggregated to create four different neighborhood units (called ABBOTT). A second approach to representing neighborhoods was to make use of those designated by Metro. These comprise 259 subdivisions of the area within the Portland Urban Growth Boundary, and can be expected to be more sensitive to variations within 123 the city of Portland as well as suburban areas (which Abbott largely ignored). Because sampled individuals were present in relatively few of these neighborhoods, these units were aggregated using physical features, freeways and political boundaries to isolate relatively homogenous areas. This took into account the presence of barriers to mobility such as the West Hills and the Willamette River, as well as the tendency of several freeways to isolate neighborhoods, and was based in part on field observations of the Portland area. Two separate sets o f neighborhoods were created with these aggregations. The first (NBOl) contains 21 zones, while the second (NB02) has 12 (Figure 5.1). Figure 5.1: Nesting of 21 NBOl (fine lines) neighborhood zones within 12 NB02 (bold lines) zones 124 The zones used for this analysis have been defined by a variety of criteria and so can be expected to capture accessibility variations at a range of scales. Political, functional, and socioeconomic units are used, reflecting not just different criteria but a range of scales as well. ANOVA testing identified significant differences in individual accessibihty for all measures within each sets of zones, except for ABBOTT and SCHOOLS, for which TIMED AREA was not significantly different. This indicates that average accessibilities within at least some geographic areas in each set of zones differ considerably firom the means accessibility in other areas. Although this does not necessarily indicate that each geographic area within Portland has a significantly different level of individual accessibility, it does show that these zones are capable of capturing geographic variations in individual accessibility. To the extent that these differences reflect substantial differences between areas, the use of these contextual variables for geographic zones can be created to help explain individual accessibility. This will be discussed in the next section. 5.4 Multilevel Modeling of Individual Accessibility In order to identify the importance of geographic scale on individual accessibihty these geographic areas were used with multilevel modeling. The relationships found between individual accessibihty and individual characteristics with stepwise regression in Chapter Four were replicated as multilevel models using the MLwin software package (version 1.02). The five original regression models (one for each accessibihty measure) were tested for each of the six sets of zones, resulting in a total of 30 models. Following Jones and Bullen (1993) these multilevel models can be represented as: 125 Accessibility of individual i in neighborhood j = average accessibility across the city + fixed effects of attributes of individual i in neighborhood j + random term for neighborhood j + random term for individual i The intercept shows the average individual accessibility within Portland, while the random terms show variations among individuals and neighborhoods. The results for the five accessibility measures with the six sets of zones used are shown in Table 5.1. The interpretation of the parameter coefficients is the same as for the regressions in the previous chapter, though now there are six different models for each of the accessibility measures. At each scale, and for each measure, the values of the constant and coefficients have remained surprisingly similar to their single level regression values, indicating that the effects of distance, hours worked, and household size are independent of spatial scale. Parameter coefficients are remarkably stable among all the sets of zones, despite their variations in size and the parts of the city they include. The major exception is the TIMED AREA measure, for which there is some difference in parameter values at the scale of political jurisdictions (CITY). Distance to the Clackamas Town Center is considerably more important when accessibility is evaluated at the scale of political jurisdictions. Likewise, the number of hours worked has less than half the effect on accessibility using this set of zones than with all others. All parameters have a ratio of estimate to standard error greater than two, the value commonly used as a rule of thumb for accepting the significance of these parameters within multilevel modeling (Jones, 126 CITY SCHOOL COMMUTERSHED Estimate/ Estimate/ Estimate/ Estimate Std. Error Estimate Std. Error Estimate Std. Error MILES Intercept 197.47 10.89 201.11 12.29 201.42 11.83 Distance to Clackamas Center -1.60 -4.03 -1.63 -4.65 -1.60 -4.17 Hours worked -1.24 -4.53 -1.25 -4.57 -1.26 -4.61 Household size -5.92 -2.17 -5.90 -2.16 -6.14 -2.24 Level 1 variance 10371.04 19.37 10414.78 19.43 10369.09 19.33 Level 2 variance 53.80 0.63 0.00 0.00 44.09 0.57 Deviance 9126.54 9127.09 9126.43 Reduction from null 34.48 41.00 36.75 OPPORTUNITIES Intercept 183.11 17.17 183.10 17.17 183.40 16.71 Distance to CBD -1.19 -2.89 -1.19 -2.89 -1.21 -2.79 Distance to 1-84/205 Center -1.37 -3.76 -1.37 -3.76 -1.35 -3.52 Hours worked -0.93 -4.23 -0.93 -4.23 -0.93 -4.25 Level 1 variance 6925.19 19.43 6925.19 19.43 6915.62 19.33 Level 2 variance 0.00 0.00 0.00 0.00 9.64 0.24 Deviance 8819.00 8819.00 8818.94 Reduction from null 39.25 48.01 46.92 AREA Intercept 175.40 15.17 175.40 15.17 175.94 14.79 Distance to 1-84/205 C enter -1.94 -6.22 -1.94 -6.22 -1.95 -5.95 Hours worked -1.02 -4.08 -1.02 -4.08 -1.02 -4.09 Level 1 variance 8839.74 19.43 8839.74 19.43 8826.65 19.33 Level 2 variance 0.00 0.00 0.00 0.00 13.09 0.25 Deviance 9003.29 9003.29 9003.21 Reduction from null 35.68 42.85 37.69 WEIGHTED AREA Intercept 174.08 15.40 174.08 15.40 174.69 14.92 Distance to 1-84/205 Center -1.91 -6.26 -1.91 -6.26 -1.92 -5.93 Hours worked -1.00 -4.10 -1.00 -4.10 -1.00 -4.12 Level 1 variance 8453.43 19.43 8453.43 19.43 8437.94 19.33 Level 2 variance 0.00 0.00 0.00 0.00 15.59 0.30 Deviance 8969.56 8969.56 8969.45 Reduction from null 34.61 42.44 37.76 TIMED AREA Intercept 277.71 12.00 281.89 12.68 277.33 11.89 Distance to Clackamas Center -3.89 -8.89 -1.83 -3.23 -1.69 -2.76 Hours worked -1.73 -2.95 -3.90 -8.91 -3.90 -8.92 Level 1 variance 27289.74 19.39 27316.32 19.43 27219.54 19.33 Level 2 variance 29.11 0.24 0.00 0.00 94.18 0.49 Deviance 9855.08 9855.10 9854.66 Reduction from null 75.12 79.70 78.27 (continued) Table 5.1: Results of multilevel models with individual characteristics 127 Table 5.1: (continued) ABBOTT N B O l N B 0 2 Estimate/ Estimate/ Estimate/ Estimate Std. Error Estimate Std. Error Estimate Std. Error MILES Intercept 201.11 12.29 200.69 11.84 201.83 11.84 Distance to Clackamas Center -1.63 -4.65 -1.58 -4.16 -1.63 -4.22 Hours worked -1.25 -4.57 -1.26 -4.60 -1.26 -4.59 Household size -5.90 -2.16 -6.16 -2.25 -6.02 -2.20 Level 1 variance 10414.79 19.43 10337.41 19.18 10361.74 19.29 Level 2 variance 0.00 0.00 74.78 0.68 54.66 0.62 Deviance 9127.09 9126.24 9126.62 Reduction from nuli 41.34 36.64 37.11 OPPORTUNITIES Intercept 183.10 17.17 181.36 16.17 182.65 16.01 Distance to CBD -1.19 -2.89 -1.11 -2.35 -1.13 -2.39 Distance to 1-84/205 Center -1.37 -3.76 -1.37 -3.31 -1.39 -3.27 Hours worked -0.93 -4.23 -0.93 -4.24 -0.93 -4.23 Level 1 variance 6925.19 19.43 6866.64 19.18 6880.39 19.28 Level 2 variance 0.00 0.00 56.19 0.75 46.52 0.73 Deviance 8819.00 8817.92 8818.29 Reduction from null 51.05 47.57 47.20 AREA Intercept 175.40 15.17 174.64 14.37 176.16 14.24 Distance to 1-84/205 Center -1.94 -6.22 -1.91 -5.60 -1.98 -5.53 Hours worked -1.02 -4.08 -1.02 -4.11 -1.02 -4.11 Level 1 variance 8839.74 19.43 8767.93 19.18 8779.04 19.28 Levei 2 variance 0.00 0.00 68.62 0.72 61.93 0.76 Deviance 9003.29 9002.26 9002.42 Reduction from null 46.61 38.28 34.02 WEIGHTED AREA Intercept 174.08 15.40 173.49 14.58 174.94 14.42 Distance to 1-84/205 Center -1.91 -6.26 -1.88 -5.63 -1.93 -5.49 Hours worked -1.00 -4.10 -1.01 -4.14 -1.00 -4.13 Level 1 variance 8453.43 19.43 8382.24 19.18 8392.22 19.28 Level 2 variance 0.00 0.00 67.38 0.73 62.55 0.78 Deviance 8969.56 8968.42 8968.60 Reduction from null 46.00 38.44 34.60 TIMED AREA Intercept 281.89 12.68 279.02 11.55 282.27 12.23 Distance to Clackamas Center -1.83 -3.23 -1.73 -2.68 -1.84 -3.03 Hours worked -3.90 -8.91 -3.90 -8.92 -3.90 -8.91 Level 1 variance 27316.33 19.43 26956.57 19.18 27228.43 19.29 Level 2 variance 0.00 0.00 352.86 1.03 89.68 0.43 Deviance 9855.10 9853.02 9854.91 Reduction from null 81.67 78.74 79.43 128 1991b). The existence of random slopes, which allow the relationships between individual accessibility and explanatory variables to vary among areas, was tested for, but none were found. Because the individual and contextual variables used were selected with a single level regression model that treats the entire Portland metropoUtan areas as a single zone, they may not reflect relationships that exist at more local scales. Relationships between accessibihty and household income, for example, maybe present at the scale of neighborhoods but not at the level of the entire metropolitan area. The imposition of single level model on the data may therefore be missing important relationships. For this reason, additional models were estimated for the NBOl and NB02 neighborhood zones to identify whether other variables appear at more local scales. Because the Mlwin software package does not include any equivalent of stepwise regression, each of the individual level variables was tested separately to identify useful variables, which were then used in combination for each of the five accessibility measures to identify the best fitting models for each measure. However, no models were found that provide a higher level of explanation than those found using stepwise regression. In addition to the parameter estimates the tables for these models also include the level 1 (individual) and 2 (zone) variances. These show variation between individuals as well as between areas. At all scales each accessibihty measure has considerably more variance at the individual level than at the neighborhood level, so that there is more variation between individuals within areas than there is among individuals in different areas. Given the accessibihty patterns visible in the surfaces in the previous chapter, this 129 is not surprising. Accessibility in Portland is highly variable, and in fact for several of the measures there is actually no variation at the zone level, meaning that the individual level explanatory variables have accounted for all variance. Although multilevel models with significant level 2 differences can be found (for example, with OPPORTUNITIES at the NBOl and NB02 levels, using only the number of hours worked as an explanatory variable), the addition of other explanatory variables eliminates these differences. This suggests that differences between areas were the result of individual differences, and adding variables eliminates these geographic differences. The results in Table 5.1 also include the deviance for each model along with the reduction in deviance firom the null model (one with no explanatory variables) for that accessibility measure. These values indicate that the models shown here do represent a significant improvement (at the traditional 95% significance level) over their respective null models. These results again confirm that when distance and individual characteristics are used to explain accessibility, a single level regression model is adequate. Any geographic variation in individual accessibility that is present in these models is therefore the result of variation among individuals. If people living in a particular area tend to have higher accessibility, than this is due to their own personal or household characteristics, not because o f the attributes of the surrounding urban environment. This is clearly an important geographical statement, as it fiuther undermines the attention given to distance within the monocentric and polycentric models. The importance of distance to the CBD, the Clackamas Town Center, or the 1-84/205 Center do not vary 130 significantly within Portland despite the expectations of the polycentric model. Likewise, the importance of the time constraints resulting from the number of hours worked per week or from being part of a large household does not vary appreciably within Portland. 5.5 Conclusions A striking result o f the use of multilevel modeling with individual accessibility data is the scale independence of the observed relationships. With the limited exception of the effects of explanatory variables on the TIMED AREA measure of accessibility, the observed relationships show no substantial variation at different geographic scales. Given the importance of individual and household attributes to accessibility values, this is not particularly surprising, yet it also suggests that because accessibility is clearly a property linked to individual characteristics and activity behavior, it is unlikely to differ among geographic areas within the city. The relationship between the number of hours worked and the area o f opportunities accessible is therefore constant across the city. This relatively constant relationship provides some support for the monocentric model, though the scale independence of relationships between distance and particular employment centers weakens the arguments in favor of polycentric models. Interestingly, these results are quite different than would be expected from postmodern urbanism, as people living throughout the city share the same relationships between their household characteristics and accessibility. So while individuals may make their own city, as Fishman (1990) suggests, they are all subject to the same relationships between employment status and accessibility. This also undermines arguments for Neo- Traditional Developments, as there is no evidence that accessibility varies among areas, 131 as would be expected if land use and urban forms were a strong influence on individual accessibility. Accessibility does not vary by scale, so that the entire metropolitan area provides a valid study area for the analysis. While the idea of measuring local and regional accessibility is appealing, the use of space-time measures renders this approach superfluous, as people define their own spaces and tie together different spatial scales through their daily movements and activities, which may be independent of any areal units (as has also been observed for feasible opportunity sets (Kwan and Hong, 1998)). Space-time measures appear to be independent of the spatial ‘frames’ used, and this framelessness suggests that the MAUP is not a problem for accessibility analysis. Due to this scale independence, it also appears that the single level stepwise regression adequately identified useful variables for the analysis. Finally, this scale independence also suggests that the spatial configuration of zones would likely play a small part, if any, in MAUP issues. Selecting different neighborhood boundaries would therefore be unlikely to affect the parameter estimates or model fit. The use of multilevel modeling to explain accessibility offers the opportunity to find geographic variations in accessibility previously invisible with single level regression modeling. This research has also identified scale dependence among individual or distance variables with the TIMED AREA measure. The importance of time constraints, identified in the previous chapter, remains unchallenged within multilevel modeling. In part this may be explained by the size of the data set, as this research uses what can be considered a relatively small sample size for multilevel research. It is likely that a much larger data set (containing several thousand individuals) 132 would greatly improve the utility of multilevel modeling as well as allowing for a much finer spatial resolution of the contextual zones. Nonetheless, there are significant variations in accessibility among neighborhoods when only the number of hours worked is used, so there is some evidence that the distance variables selected by regression models appear to have actually masked the importance of location within the initial multilevel models. The only significant differences between areas found were with neighborhoods. This strongly suggests that such differences are more likely to be found with smaller, more compact areas, than with larger groupings based on political jurisdictions, school districts, or commutersheds, as well as supporting the use of the Metro defined neighborhood zones. So while land uses and even socioeconomic characteristics can vary tremendously and quite visibly between political jurisdictions and school districts, these differences need not always result in accessibility variations. The presence of large cities or school districts (such as Portland) and the large aggregations used in this chapter have likely reduced much of the variation present in Portland. Even with a larger sample, the size of Portland may prevent political jurisdictions fi-om having important effects within multilevel modeling. The same is also true of the ABBOTT neighborhood divisions. Because Abbott (1983) characterizes the majority of the metropolitan area as auto era suburbs, this effectively eliminates any possibility for variations within the majority of the study area. This may be consistent with the implied monocentric bias that Abbott uses, but it is clearly of little use for explaining accessibility variations in Portland. The scale of polycentric commutersheds has also failed to produce significant 133 variations in accessibility. This perhaps should be expected given the failure of the polycentric model in other contexts (Pickus and Gober, 1988; Fujii and Hartshorn, 1995) and the provisional nature of certain regional centers in the 2040 regional growth plan. This chapter has shown that there are no substantial accessibility variations when individuals are grouped within discrete zones or accessibihty is measured at different scales in Portland. Given the frameless nature of space-time accessibility measures, and the highly individual nature of these measures, this is perhaps not surprising. However, the analysis in this chapter only examined variations between areas and scales by grouping individuals within particular areas. Any differences that could be found would therefore be due solely to individual characteristics, such as the amount of travel, the location of activities, and the temporal scheduling of those activities. There remains the possibihty that the characteristics of those zones may also make a difference to accessibility. This issue will be explored in the next chapter. 134 CHAPTER 6 ACCESSIBILITY AND GEOGRAPHICAL CONTEXT IN PORTLAND 6.1 Introduction The previous chapter grouped individuals within several sets of zones in order to identify variations in accessibility among those zones as well as between different spatial scales. Multilevel modeling was used to allow the identification of the extent to which differences were due solely to individual characteristics, such as the amoimt of travel, the location of activities, and the temporal scheduling of those activities. However, there were no substantial differences except when models using only the number of hours worked were used. This indicates that individual accessibility within cities does not vary because individual behavior differs among political boundaries, commutersheds, or neighborhoods. While there are clearly variations in individual activity and travel behavior, these differences do not correspond to discrete areal units. As a result, space time accessibihty measures appear to be essentially frame-independent, with differences in access to employment or retail opportunities varying more due to individual characteristics or the ways in which people define their own activity spaces. This largely ignores the possible influences of the urban environment on accessibility. 135 However, because the previous chapter examined only variations due to grouping individuals within zones, there remains the possibility that accessibility may vary due to the characteristics of these areas. Individual accessibility may be influenced or mediated by the local areas within which people live or work. As will be discussed, this may be because land use characteristics have created greater concentrations of opportunities within certain areas, allowing for greater access within these areas. Or these local areas (whether based on legal jurisdictions and/or neighborhoods) may also form basic urban spatial units within which people will have much in common, including employment characteristics, commuting patterns, or residential preferences. These neighborhood effects maybe reinforced by political attempts at attracting/excluding certain land uses or types of people. Residence in such spatial units (and the characteristics of these units) should therefore be able to account for more variation in accessibility than distance by reducing the micro-level variance (that which the individual level variables cannot explain) within the multilevel models. This chapter will therefore examine the influence of the local geographic contexts introduced in the previous chapter to individual accessibility in Portland. The goal will be to identify whether the characteristics of the areas in which people live and work can help explain spatial variations in accessibility, and whether this contextual influence can provide a higher level of explanation than distance or individual characteristics. This analysis will again make use of multilevel modeling, and will focus on two neighborhood scales (NBOl and NB02) for which significant area differences were found in the previous chapter (when using only the number of hours worked). The extent that contextual characteristics are useful explanatory variables will provide additional 136 evidence on the relative importance of behavior and the urban environment to accessibility. This may also help confirm the utility of alternate conceptualizations of urban form and human behavior. The reasons to expect that accessibility will vary due to contextual characteristics will be discussed in the first section, followed by the evaluation of the relationship between contextual characteristics and accessibility at a single and at multiple scales. 6.2 Incorporating Geographical Contexts Several reasons exist to consider the influence of geographical context on individual accessibility. First, the use of monocentric or polycentric models leads to the expectation that accessibility should be higher in areas with greater population and housing density, as these areas should be closer to the center(s) of the city. According to these models density would serve as a proxy for distance from the center(s), but because measuring density directly takes into account variations within the built form of the city this could potentially provide a higher level of explanation for accessibility variations. Attributes representing population density, housing density, the proportion of homes that are detached (representing single family homes), and the density of potential activity opportunities within each zone (the weighted area of opportunity parcels divided by acreage of the neighborhood) were therefore selected for testing. Because some households would select a distance/housing cost tradeoff that allows them to live in suburban locations, accessibility should also decrease with higher average housing 137 values, a larger proportion of housing units that are owner occupied, and newer homes (represented here by the proportion of housing built before 1970), variables measuring these characteristics were also included. The proponents of Neo-Traditional Development and the New Urbanism also make strong claims about the influence of different types of urban forms on travel behavior and accessibility (Boamet and Crane, 2001). As with the monocentric and polycentric views, higher densities should be associated with higher accessibilities because individuals would be expected to hve much closer to a greater range of potential activity opportunities (though there would not necessarily be any expectation that these opportunities would be concentrated in a downtown location), and so should be able to reach more of these locations. Older residential areas would be indicative of higher densities and more dispersed employment and retail opportunities (representing actual ‘traditional’ developments rather than Neo-Traditional imitations), and people living in these areas should therefore have higher accessibilities. In contrast, the postmodern urbanism conceptualization leads to an expectation that accessibility would be greatest in areas of higher income and status, regardless of location within the city. Accessibility should therefore increase with higher average housing values, a larger proportion of housing units that are owner occupied, and newer homes (represented here by the proportion of housing built before 1970). Higher population and housing densities would likely be associated with lower incomes and less desirable areas, so would have lower accessibility. Greater densities of opportunities would certainly be expected to result in higher accessibilities, though with the expectation that these opportunity densities exist in part because of the desirability of 138 these areas of the city. Areas with higher household incomes and occupied primarily by white collar workers and those with greater educational attainment would also be expected to have higher accessibihties (and it would be expected that the city would be sorted by occupation and income as well as race or other social divisions). Additional variables were created to measure these characteristics. Secondly, geographic contexts may be important to accessibihty because places are not just empty spatial containers but are instead active entities influencing human behavior and knowledge (Hanson and Pratt, 1988; Wyly, 1998). This possibility has been explored particularly in the context of women’s employment possibilities and the ways in which the community can be important in providing information about jobs and support services such as day care, transportation, or recreation opportunities (Hanson and Pratt, 1988; Wyly, 1998). These services can be crucial to an individual’s (especially a woman’s) ability to work outside the home, and communities can also be important as a source of information about jobs. Because women work closer to home and acquire more information about the local labor market from within their community than men they could be expected to be more influenced by location than men (Hanson and Pratt, 1988). The residential community is also important because of “the socialization process by which certain work-related attitudes, skills, and goals are passed from one generation to the next” (Hanson and Pratt, 1988, 309). These happen not only within the home “but also within the neighborhood via schools and social interaction” (Hanson and Pratt, 1988, 309). As a result, “the neighborhood is an important ‘agent’ of socialization that structures work/career aspirations and work-related attitudes” (Hanson and Pratt, 1988, 309). Because certain occupations (professional and managerial jobs) have higher 139 mobility and different residential selection processes than others, the importance of local communities could vary among social groups (Hanson and Pratt, 1988). hiformation representing occupation and income could therefore be used as proxy variables for social values within neighborhoods. Living within a predominantly blue-collar neighborhood may have an influence on an individual’s travel behavior independent of their own personal characteristics, and similarly income levels may represent or covary with attitudes toward work and education. Those individuals living and interacting within contexts where a larger proportion of the population that is employed, and a greater proportion of workers are employed within the Portland city limits, may have more information about the city and be more willing or able to make use of employment opportunities at a farther distance from their home location. Areas that are predominantly blue collar (represented by the proportions working in primary or manufacturing sectors) may be more geographically rooted in their knowledge of the city as well as limited in their mobility and abihty to find work outside their local areas than white collar workers (represented by areas dominated by workers in finance, insurance and real estate (FIRE) and executive/professional occupations). Additionally, areas with a higher proportion of worker commuting to work in single occupancy vehicles may have greater mobility and knowledge of the city, and therefore potentially higher accessibility. This could also serve as a surrogate for the lack of transit dependence, which would certainly be expected to result in lower accessibility. Although the sample used in this research is exclusively auto based, certain individuals may have been selected from areas with relatively low auto usage and therefore lower average accessibility. Dual-eaming households may however become more involved in 140 work-based social contacts than neighborhood contacts and so become less influenced by the neighborhood context (Hanson and Pratt, 1988). Areas that are predominantly occupied by married child-raising households may therefore be expected to have little effect on accessibility, while those with large proportions of single-person households. Finally, low residential turnover might suggest a lack of residential mobility, which Hanson and Pratt (1988) have shown to be important for the influence of home environment on decisions about work, and therefore accessibility. To identify and test the importance of geographic contexts within Portland, a set of contextual variables that capture these land use and socioeconomic characteristics was created by interpolating socioeconomic data from census block groups to the NBOl and NB02 neighborhood zones. The most straightforward means of transferring the values of these source zones to the target neighborhood zones is the overlay method, which “superimposes the target zones on the source zones. The values of the target zones are then estimated from the size of the overlapping zones” (Lam, 1983, 139). However, this approach requires homogeneity within the census block group source zones, as “if the value of each source zone is unevenly distributed within its domain, estimation of target zone values from the amount of overlapping areas may not be reliable” (Lam, 1983, 140). Within a city such as Portland, it would be unrealistic to assume that that population, income, housing densities, or other characteristics will not vary within block groups, a problem that is especially severe because the target zones are themselves at a fairly fine scale. 141 This problem can be overcome using pycnophylactic interpolation to allow values from adjacent zones to influence the values of a source zone (Lam, 1983), replacing the assumption that phenomena are evenly distributed within block groups with the expectation that they are constantly varying across space. Further, pycnophylactic interpolation is a volume preserving method, meaning that the total value of the variable present within each source zone remains the same within the interpolated surface (Lam, 1983). The original source population (or other) values can therefore be reconstructed from the target zones once the variable has been interpolated. This method was implemented within ArcView GIS by first using the census block group polygons to create a raster surface (with 500 foot cells) for each of the socioeconomic variables, then using a specialized script (Riedl, 1998) to pycnophyactically interpolate the value of each variable within the boundaries of the original source zone. The neighborhood zones were then overlaid onto each surface and the values transferred to the target zones. 6.2.1 Accessibility and Geographical Context In order to identify the most useful contextual characteristics, each of the five accessibility measures was tested using multivariate stepwise regression. The useful variables identified in Chapter Four were used, but with the addition of the contextual variables described above. This will reveal the extent to which contextual characteristics are useful at the metropolitan level. The results for the best fitting models are shown in Table 6.1. As before, all models and parameter coefficients are significant at the 99% significance level. As can be seen, relatively few contextual variables remained within the final models, though these varied considerably according to the particular 142 Dependent Independent Standardized Significance Adjusted Variable Variable Coefficient Std Error Coefficient Level Rsduare Rsquare MILES Constant 201.11 16.41 0.000 0.055 0.051 Distance to Clackamas Center -1.63 0.35 -0.17 0.000 Hours worked -1.25 0.28 -0.16 0.000 Housetiold size -5.90 2.74 -0.08 0.031 Constant 171.08 21.64 0.000 0.06 0.055 Distance to Clackamas Center -2.26 0.46 -0.23 0.000 Hours worked -1.31 0.28 -0.17 0.000 Housetiold size -6.00 2.73 -0.08 0.028 Proportion working in Portland (CITY) 83.87 39.52 0.10 0.034 Constant 388.12 91.29 0.000 0.06 0.055 Distance to Clackamas Center -2.22 0.45 -0.23 0.000 Hours worked -1.29 0.28 -0.17 0.000 Household size -5.51 2.74 -0.07 0.045 Proportion living in same house (SCHOOL) -408.60 196.23 -0.10 0.038 Constant 382.67 91.44 0.000 0.055 0.051 Distance to Clackamas Center -2.31 0.45 -0.23 0.000 Hours worked -1.21 0.27 -0.16 0.000 Proportion living in same house (SCHOOL) -435.79 196.16 -0.10 0.027 OPPORTUNITIES Constant 183.19 10.70 0.000 0.094 0.091 Distance to Portland CBD -1.19 0.41 -0.13 0.004 Distance to 1.84/205 Center -1.37 0.37 -0.17 0.000 Hours worked -0.93 0.22 -0.15 0.000 Constant 150.18 17.76 0.000 0.101 0.096 Distance to Portland CBD -1.44 0.43 -0.16 0.001 Distance to 1.84/205 Center -1.29 0.37 -0.16 0.000 Hours worked -0.89 0.22 -0.14 0.000 Proportion of homes detached (NBOl ) 54.79 23.64 0.84 0.021 Constant 143.89 21.20 0.000 0.1 0.095 Distance to Portland CBD -1.39 0.42 -0.15 0.001 Distance to 1.84/205 Center -1.28 0.37 -0.16 0.000 Hours worked -0.89 0.22 -0.14 0.000 Proportion of homes detached (NB02) 62.78 29.33 0.08 0.033 Constant 210.16 15.44 0.000 0.096 0.092 Distance to 1.84/205 Center -2.44 0.30 -0.31 0.000 Hours worked -0.95 0.22 -0.15 0.000 Proportion working in manufacturing (SCHOOL) -157.26 50.80 -0.12 0.002 Constant 120.80 27.59 0.000 0.101 0.097 Distance to Portland CBD -1.93 0.51 -0.21 0.000 Distance to 1.84/205 Center -0.94 0.40 -0.12 0.020 Hours worked -0.96 0.22 -0.15 0.000 Proportion of homes detached (COMMUTERSHE 111.09 45.38 0.11 0.015 AREA Constant 175.40 11.58 0.000 0.066 0.063 Distance to i.84/205 Center -1.94 0.31 -0.22 0.000 Hours worked -1.02 0.25 -0.14 0.000 Constant 243.37 29.45 0.000 0.074 0.07 Distance to 1.84/205 Center -2.38 0.36 -0.27 0.000 Hours worked -1.07 0.25 -0.15 0.000 Proportion working in Portland (COMMUTERSHE -183.89 65.32 -0.10 0.012 WEIGHTED Constant 174.08 11.33 0.000 0.067 0.064 AREA Distance to 1.84/205 Center -1.91 0.31 -0.22 0.000 Hours worked -1.00 0.24 -0.14 0.000 Constant 240.79 28.80 0.000 0.074 0.071 Distance to 1.84/205 Center -2.34 0.35 -0.27 0.000 Hours worked -1.05 0.24 -0.15 0.000 Proportion working in Portland (COMMUTERSHE -160.84 63.88 -0.10 0.012 TIMED AREA Constant 281.89 22.27 0.000 0.103 0.1 Distance to Clackamas Center -1.83 0.57 -0.11 0.001 Hours worked -3.90 0.44 -0.31 0.000 Table 6.1: Regression results with individual and contextual characteristics 143 accessibility measure and the scale of the contextual variable. The proportion of employed persons working within the city of Portland was a useful variable for explaining MILES, AREA, and WEIGHTED AREA, though for MILES it was the proportion within each CITY grouping, and for the area measures it was measured at the COMMUTERSHED level. When this variable was measured at the CITY level the effect of this variable on individual accessibility was positive, indicating that accessibility tends to be higher with an increasing proportion of workers commuting into Portland. Given the dominance of the city of Portland within the study area, this variable may be distinguishing between those who live and work in Portland and those living and working in suburban locations. The inclusion of this variable tends to increase the negative effects of distance and time constraints. However, when measured at the COMMUTERSHED level the effect of this variable is negative, so that accessibility tends to be lower in those areas where the number of people commuting into Portland is greater. The inclusion of this variable also tends to increase the negative effects o f distance and time constraints. Residential turnover is related to accessibility, but only when measured by MILES and only at the level of SCHOOLS. In this case the regression modeling identified two useful models involving this contextual variable, in each of which the effect is negative. Those areas with lower residential turnover (a greater proportion of people living in the same home they lived in 10 years earlier) tend to have lower accessibility. Including this variable tends to increase the negative effect of distance to the Clackamas center but has little effect on the other variables. The proportion of housing within each zone that are detached (representing single family homes) is related to the number of OPPORTUNITIES accessible to individuals at the level of NBOl, 144 NB02, and COMMUTERSHEDS, as well as to the TIMED AREA accessible at the scale of NBOl. In all cases the effects are positive, so that individuals living within areas with a larger proportion of single-family homes tend to have higher accessibihty. Two employment variables were of use as well. The proportion of workers in manufacturing occupations was negatively related to the number of OPPORTUNITIES accessible at the scale of SCHOOL districts, and the TIMED AREA available was positively related to the proportion employed in executive or professional occupations within political jurisdictions (CITY). The importance of occupation likely represents the effects of residential segregation. The availability (or size) of potential activities, or the mobihty needed to reach these opportunities, appears to be less within blue-collar communities than elsewhere. Incorporating the context in which individuals live and work provides additional explanation for individual accessibility in Portland, though it also raises further questions. The scale at which context is measured is clearly important, with employment contexts most visible when using larger groupings, such as political jurisdictions, school districts, and commutersheds. Characteristics of housing are visible with smaller neighborhood units, though this variable, as well as residential turnover, are also useful when measured at the scale of COMMUTERSHEDS and SCHOOLS. It is therefore not possible to make simple statements about which scales are most important, and in fact the effects of context may change at different scales (as appears to be the case with the proportion of workers commuting to jobs in Portland). In most cases the effects of these contextual variables (as assessed by the standardized coefficients) are similar to those of the distance 145 variables and the effects of time constraints on individuals, though contextual variables are however of greater strength than household size in those models where the latter variable is present. A number of variables are noteworthy for their absence. These include the population density of zones, the proportion of housing built before 1970, and the density of potential opportunities, which might be expected to be very important influences on individual accessibility because of their relation to land uses. The income of neighborhoods or political jurisdictions also failed to appear, so that accessibihty does not appear to vary between wealthier and less prosperous areas of Portland. The absence of these variables from the models highlights the importance of behavior to individual accessibility, and the fact that much of an individual’s access to opportunities is influenced by where they work or engage in out of home activities, so that the attributes of their residential area may be less important. However, because these relationships were estimated at the metropolitan level, it may be that substantially different relationships may exist at different scales. Although the previous section identified the metropolitan scale as appropriate for evaluating individual level relationships to accessibihty, this may not be the case with contextual relationships. The effects of the proportion of single-family homes on accessibility may be substantially different at different scales or in different parts of the city, yet these variations are invisible in the traditional single level model. The use of multilevel modeling to help isolate differences between individuals and locations should therefore prove useful, and this will be discussed in the next section. 146 6.3 Contextual Characteristics and Multilevel Modeling In order to identify the importance of geographic context on individual accessibility the contextual variables created for these geographic areas were used with multilevel modeling. The original relationships found between individual accessibility and individual characteristics identified with stepwise regression in the previous section were replicated in multilevel form. These models can be represented as: Accessibility of individual i in neighborhood j = average accessibility across the city + fixed effects of attributes of individual i in neighborhood j + fixed effects of attributes of neighborhoods + random term for neighborhood j + random term for individual i The results for these contextual models are shown in Table 6.2. Only those models with contextual variables (and are therefore different from the individual level models in Chapter Five) are shown in the table. Two models were calculated for the MILES measure, the first using political jurisdiction and including the proportion of workers within each zone commuting to work in Portland, and the second at the level of SCHOOLS. In all three models the constant is now considerably larger (though the standard error has risen considerably as well) but the individual level parameters remain reasonably similar to the individual level model as well as the regression models. For OPPORTUNITIES four models were estimated, one each at the scale of COMMUTERSHEDS, SCHOOLS, NBOl, and NB02. Similar results are apparent for the three area measures. Except for larger constants, the parameters remain fairly 147 CITY SCHOOL COMMUTERSHED Estimate/ Estimate/ Estimate/ Estimate Std. Error Estimate Std. Error Estim ate Std. Error MILES Intercept 171.08 7.93 388.09 4.27 Distance to Clackamas Center -2.26 -4.93 -2.22 -4.94 (no contextual Hours worked -1.31 -4.77 -1.29 -4.72 variables) Housetiold size -6.00 -2.21 -5.51 -2.02 Proportion working in Portland 83.87 2.13 Proportion in same tiouse -408.57 -2.09 Level 1 variance 10352.63 19.43 10354.92 19.43 Level 2 variance 0.00 0.00 0.00 0.00 Deviance 9122.57 9122.74 Reduction from null 39.45 45.35 OPPORTUNITIES Intercept 210.16 13.65 120.81 4.39 Distance to CBD (no contextual -1.93 -3.79 Distance to 1-84/205 C enter variables) -2.44 -8.05 -0.94 -2.34 Hours worked -0.95 -4.31 -0.96 -4.38 Proportion in manufacturing -157.26 -3.10 Proportion tiomes detached 111.08 2.46 Level 1 variance 6913.28 19.43 6870.30 19.43 Level 2 variance 0.00 0.00 0.00 0.00 Deviance 8817.70 8813.00 Reduction from null 49.31 52.86 AREA Intercept 243.37 8.29 Distance to 1-84/205 C enter (no contextual (no contextual -2.38 -6.68 Hours worked variables) variables) -1.07 -4.31 Proportion working in Portland -163.89 -2.52 Level 1 variance 8766.26 19.43 Level 2 variance 0.00 0.00 Deviance 8996.99 Reduction from null 43.91 WEIGHTED AREA Intercept 240.79 8.38 Distance to 1-84/205 C enter (no contextual (no contextual -2.34 -6.72 Hours worked variables) variables) -1.05 -4.32 Proportion working in Portland -160.84 -2.52 Level 1 variance 8382.66 19.43 Level 2 variance 0.00 0.00 Deviance 8963.21 Reduction from null 43.99 TIMED AREA Intercept 23.55 0.43 Hours worked -3.76 -8.65 (no contextual (no contextual Proportion in exec/professional 645.12 4.05 variables) variables) Level 1 variance 27105.73 19.43 Level 2 variance 0.00 0.00 Deviance 9849.26 Reduction from null 80.94 (continued) Table 6.2: Results of multilevel models with contextual characteristics 148 Table 6.2: (continued) NB01 NB02 Estimate/ Estimate/ Estimate Std. Error Estimate Std. Error OPPORTUNITIES Intercept 150.34 8.40 144.19 6.77 Distance to CBD -1.43 -3.29 -1.38 -3.23 Distance to 1-84/205 Center -1.29 -3.47 -1.28 -3.45 Hours worked -0.89 -4.05 -0.89 -4.05 Proportion homes detached 54.17 2.26 62.36 2.11 Level 1 variance 6869.12 19.19 6878.92 19.29 Level 2 variance 6.80 0.12 4.24 0.10 Deviance 8813.60 8814.40 Reduction from null 51.89 44.93 TIMED AREA Intercept 208.36 5.32 Distance to Clackamas Center -1.77 -2.96 (no contextual Hours worked -3.82 -8.74 variables) Proportion of homes detached 109.48 2.27 Level 1 variance 26956.15 19.19 Level 2 variance 133.99 0.51 Deviance 9848.50 Reduction from null 83.26 consistent, and all have estimate/standard error ratios in excess of two. The reduction in deviance for two models for OPPORTUNITIES (that for SCHOOLS and NB02) are not significantly different fi-om the previous models estimated in Chapter Five (using individual level variables only), so that there is no improvement in fit of the model gained by incorporating contextual variables. With the exception of the model for OPPORTUNITIES with NBOl neighborhoods, the use of contextual variables has reduced individual level variances, though also reducing neighborhood level variations. As a result, there are again no significant neighborhood differences. This indicates that there are no significant differences present between individuals in different zones, and therefore the relationship between accessibility and explanatory characteristics can be 149 adequately modeled at a single level. The possibility for random slopes (in which different neighborhoods have not only different average levels of accessibility but different relationships between individual or neighborhood characteristics and accessibility) was tested for but no useful relationships were found. The models for OPPORTUNITIES show some scale dependency with the negative effect of distance to the 1-84/205 center being greatest at the level o f school districts and smallest within polycentric commutersheds. The effect of a higher proportion of detached homes also varies between commutersheds and neighborhood units, with stronger effects found using larger zones. In contrast, the results for the three area measures are similar to their single level counterparts. Except for larger constants, the parameters remain fairly consistent, and all have estimate/standard error ratios in excess of two. With the exception of the model for OPPORTUNITIES with NBOl neighborhoods, the use of contextual variables has reduced individual level variances, though also reducing neighborhood level variations. As a result, there are again no significant neighborhood differences. This indicates that there are no significant differences present between individuals in different zones, and therefore the relationship between accessibility and explanatory characteristics can be adequately modeled at a single level. These models do however provide a slightly greater level of fit than models with individual level variables only, as shown by the lower deviances. However, as with the individual level models tested earlier, it may be that the contextual characteristics that show up within a single level regression model are not those that are important within a multilevel model. Combinations using each contextual variable were therefore again tested to find any useful multilevel models at neighborhood 150 levels exist that were not identified with single level regression modeling. Again, however, no models were found which provide an equal or better level of fit when compared to the single level modeling. One important potential limitation of multilevel modeling is the reliance on zones, especially as previous sections have shown that individual accessibility is largely frameless and independent of areal units. While most multilevel research has been carried out at interurban or regional scales with poUtically designated boundaries, this is not necessarily possible at the intraurban level where relatively intangible areas such as commutersheds or neighborhoods are more relevant, hi addition to issues of MAUP and boundary effects, it can also be argued that this is an inappropriate conceptualization of neighborhoods or local contexts within cities, because it is assumed that all individuals living within a polygon will have the same experience of the city. The use of neighborhood areas or local contexts based on individual activities or experiences would therefore likely be useful. As one example, Hanson, Kominiak, and Carlin (1997) used a circle centered on each individual's residence to define each individual's activity space within which they are more likely to learn of employment opportunities. This radius was defined by the median commuting distance within the metropolitan area being studied, so is only meaningful for those who are employed. Because not all individuals in the Portland sample are employed, a test of this approach to conceptualizing local context was used with contexts defined as those areas reachable within five minutes free-flow driving time from each individual’s home location. The resulting areas differed greatly in size and shape, depending on location within the city and proximity to freeways and streets with fast driving speeds. Contextual 151 variables were created for these local areas using the same procedure discussed above for the areal zones. When used with stepwise regression to explain accessibility, only one useful model was found, showing that the number of OPPORTUNITIES accessible to individuals varied with the distance to the CBD, the distance to the 1-84/205 center, the number of hours worked per week, and the density of opportunities within five minutes driving time of each individual’s home location (Table 6.3). While goodness of fit statistics shows that most coefficients and the entire model is significant at the 99% level, the parameter coefficient for the local density of opportunities is only significant at the 95% level. The coefficients for hours and distance variables are as before, while the local density of opportunities has a negative influence on accessibility. This is very interesting, as higher densities of opportunities close to home is actually related to lower accessibilities! This is the exact opposite of what would be predicted according to the New Urbanism, and is a striking indication of the ways in which space-time measures take into account the activities and movements of people rather than simply evaluate the proximity of employment or retail activities. When represented in multilevel form at the level of neighborhoods there is very little difference in the models (Table 6.4), except that the density of opportunities has a slightly greater influence with larger neighborhood units. The lack of significant neighborhood level differences shows that there is again no need for a multilevel model, the reduction in deviances shows that there is a significant gain in model fit by adding this variable. 152 6.3 Discussion and Conclusions The use of contextual characteristics in multilevel modeling leads to several conclusions. First, the influence of geographic context on individual accessibility appears to be quite weak. Given the poor results of using distance to explain accessibility (as well as the ways in which distance and context may be linked) and the importance of individual and household attributes, this is not surprising. Accessibility is clearly a property of individuals, with strikingly little influence of urban form. The results from this section provide additional evidence against the monocentric and polycentric models as densities and other local characteristics do not provide useful surrogates for distance from the center of the city. The results also undermine the new urbanism hterature in that there is only limited support for linking accessibility to land use or urban form. Socioeconomic characteristics and social divisions are not associated with accessibility Dependent Independent Standardized Signifcanoe Atfusted Variable Variable Coeflident StdBrror CoefRdent Level Rsquare Rsquare OPPORTUMTIES Constant 201.977 13.962 0.000 0.099 0.095 Distance to CBD -1.315 0.417 -0.146 0.002 Distance to 1-64/205 Center -1.625 0.384 -0204 0.000 Housvwxked -0.939 022 -0.148 0.000 Local Density of Opporiunties -79.55 38.085 -0.084 0.037 differences, providing no strong evidence in favor of postmodern urbanism. Table 6.3: Regression results with local areas These results also show a fair amount of scale dependency, so the scale at which geographic context is included is important to the results. This therefore requires that attention be given to identifying the appropriate scale for analysis (Openshaw, 1996), yet 153 the weakness of these relationships suggest that context is of httle importance. This is consistent with what was observed earlier regarding the firamelessness o f space-time measures. People may work and carry out other activities outside their local area (however defined), they may travel considerable distances across the city in the course of their daily activities, and have extended social networks outside their local area. These movements and activities obviously do provide structure to people’s lives, but this structure is independent of neighborhoods, political boundaries, school districts, or commutersheds. The idea that travel or knowledge of the city is limited to a single discrete area is therefore not supported, and so neither is the idea of neighborhoods as ‘active agents’ shaping knowledge of the city. NB01 NBOl Estimate/ Estimate/ Estimate Std. Error Estimate Std. Error OPPORTUNITIES Intercept 200.34 13.86 202.35 13.75 Distance to CBD -1.28 -2.78 -1.32 -2.81 Distance to 1-84/205 Center -1.60 -3.87 -1.61 -3.78 Hours worked -0.94 -4.30 -0.94 -4.28 Density of Local Opportunities -77.13 -1.98 -80.71 -2.06 Level 1 variance 6846.34 19.19 6848.45 19.29 Level 2 variance 37.71 0.55 37.18 0.63 Deviance 8814.11 8814.07 Reduction from null 51.38 51.42 Table 6.4: Multilevel results using local areas 154 Contextual characteristics are also scale dependent. Certain characteristics of neighborhoods or other zones are useful at particular scales and not others, and often only for certain accessibihty measures. As examples, the proportion of detached housing is useful at the neighborhood level only, while the proportion of workers who commute to jobs located within the city of Portland is appropriate at the scale of CITYs and COMMUTERSHEDS. The OPPORTUNITIES measure appears to give the strongest results, and is the one that results in the most geographic differences. This may be surprising, given the relatively simple nature of this measure compared to the several area measures, but it again shows that the presence of large employment or retail centers near an individual’s residence will not necessarily lead to higher accessibility if the individual has no time to visit those areas. Different contexts work best for certain measures, and models that assume that variables are useful at all levels may be missing out on important relationships present. Not only is there no single best way of measuring accessibility, there may also be no single best way to evaluate relationships between the urban environment and accessibihty. Thirdly, the use of distance and contextual variables appears to lead to distinct sets of models, with little overlap. It is clear that while adding additional variables such as distance provides a higher level of explanation, it also eliminates the geography of accessibility within Portland, because including the distance to the Portland CBD or the I- 84/205 and Clackamas Town centers masks neighborhood differences in accessibility. This is likely due to the fact that distance is measured from each individual’s home, and there will be great differences in this value within each zone, while individuals living in adjacent zones may have similar distances. For this reasons adding distance to the 155 models increases within zone differences while decreasing between zone differences. The exception is the distance to the Gresham regional center, which is located in the far eastern part of the study area and is one of only three not located adjacent to a freeway (the remaining two centers (the Hillsboro CBD and Milwaukie CBD) do not show up in any models in this research). Because this center is both peripheral and requires considerable driving on city streets to reach, distances to this area may show less variation among individuals, preserving the geographic differences that clearly exist in individual accessibility. Alternatively, the location of the Gresham center may be such that considerably different routes must be used to reach it from different parts of the study area, increasing travel time differences for individuals within different neighborhoods, and so increasing between zone variations. In either case, it appears that measuring distance by travel time through the street network tends to reduce geographic variations between parts of the metropolitan area. Distance and contextual characteristics can therefore be seen as opposing approaches to explaining accessibility, and are not likely to be complementary. While the MAUP is an important issue for the use of contextual zones, identifying the correct set of points and the proper measure of distance for monocentric and polycentric approaches is a similarly difficult task, and neither approach is straightforward. Finally, this chapter not only offers additional confirmation of the limited explanatory ability of distance but also strongly suggests that contextual characteristics are of relatively little importance when compared to individual or household characteristics. Although providing additional evidence against the monocentric and polycentric models these findings also challenge the notions of postmodern urbanism. 156 Individuals living within areas with low residential turnover, populated predominantly by white-collar workers, and with a high percentage of single-family homes tend to have higher accessibility regardless of their own individual or household constraints. In contrast, people living in blue-collar areas tend to have lower accessibility independently of their time constraints or distance to major centers within Portland. However, the exact nature of the relationships between areas and accessibility remains to be determined through additional research, most likely using qualitative fieldwork to assess the ways in which people gain information about the city, make travel and activity location decisions, and therefore construct their daily accessibility. The ideas of the New Urbanism, and also the 2040 Growth Plan for Portland, are not strongly or consistently supported either. The fact that a greater proportion of detached homes in an area is associated with higher access to opportunities does not fit expectations and actually challenges the idea that increasing housing densities in Portland will be an improvement (though of course goals other than accessibility are also important to the 2040 plan). The difference relationships between accessibility and blue- collar and white-collar populations also highlight the need for examining the impact of the 2040 plan on different parts of the city and different social groups living in different areas. However, it must be remembered that the 2040 plan is explicitly multimodal, while this research deals only with automobile travel. As the importance of regional centers is based in large part on transit and walking trips, these results may underestimate the importance of these centers to individual accessibility in Portland. 157 Accessibility cannot be simply read off from the characteristics of neighborhoods or cities, or even from the local land uses around an individual’s home. This is an important point, as while individual accessibihty varies within Portland, it does so to a large extent because people and households vary, not because of distance from specified central points or even because of the uneven urban environment. This is not to deny there is a geography of accessibihty, but rather that people not only make their own accessibihty through their daily activities (within their space-time limits), but that this accessibihty will reflect their spatial behavior within cities. Using the built form of cities to attempt to alter accessibihty is not likely to be straightforward, especially if these attempts are based on assumptions about the role of distance in shaping behavior. Further, it must be remembered that (as shown in Chapter Four) individual accessibihty is dynamic, varying by time of day and also by day of week. As can be seen, accessibihty will be more hkely to tend to reflect individual and household characteristics rather than those of common spatial units. Because accessibihty is an attribute of individuals, it should be studied at this level. This is consistent with the results of the previous chapter regarding the importance of time constraints on individual accessibihty. Not only do household characteristics appear to influence the importance of distance to selected features, it may also be that these household characteristics may also influence the importance of the local urban environment, though this requires more study. These issues, and the possible impacts of new technologies on accessibihty, will be discussed in greater detail in the next chapter. 158 CHAPTER 7 CONCLUSIONS 7.1 Introduction The primary goal of this research was to identify the relevance of alternate conceptualizations of urban form and human behavior to the evaluation of individual accessibility. These conceptualizations are important because of the way in which our understanding of cities (and expectations about how people behave within them) has been structured by models of urban form, each with their own logic and imphcations for individual accessibility. The reliance on a particular urban model assumes much about what accessibility is and how it should be measured even if the user does not make this explicit. There is increasing reason to think that certain models may be obsolete as explanations of urban form, and although this topic continues to be debated, this research has shown that not all urban models are equally adequate for discussing accessibility. This chapter will discuss the implications for the use of the prevailing monocentric and polycentric models in accessibility research, as well as the related notions of Neo-Traditional Development. This research does not support the accessibility implications of these models and concepts, regardless of their utility for understanding 159 urban form. The influence of distance on accessibility has been evaluated in several different ways, yet little evidence has been found to support this influence. Neither have the effects of geographical context met the assumptions of the New Urbanism or the 2040 Growth Plan for Portland. The expectations of postmodern urbanism that there should be social and spatial fragmentation and polarization are harder to evaluate, but also do not appear to be an effective means of explaining individual accessibility within contemporary cities. Instead, individual and household characteristics such as time constraints appear to be of the greatest importance, which implies that any theoretical construct that ignores the importance of activity schedules to households will be unlikely to provide any useful explanation for accessibility variations. These issues will be discussed in greater detail in the next section. Finally, given that urban form, household organization, spatial mobility and technology are constantly changing, the implications for ongoing and potential future changes in urban form and technology to accessibility will also be discussed. Central to these discussions is the Internet, which is sometimes expected to result in the impending elimination of distance as a variable in social, economic, and urban geography (though there have of course long been reasons to doubt the determination of land uses and human behavior by distance, as discussed in Chapter Two). However, while the Internet may indeed have tremendous geographic impacts in the future, it will be argued that this research has shown that it is actually the Internet's relaxation of time constraints, rather than its effects on spatial constraints, that will likely have the greatest impacts on individual accessibility. 160 7.2 Individual Accessibility in the Late Twentieth Century City The results of this research show that there is considerable reason to question the importance given to distance in much of the literature regarding accessibility. The regression modeling carried out in Chapter Four revealed very little support for the role of distance required by the monocentric model, and while they do appear to provide superficial support for the polycentric model, the reasons for the observed spatial patterns actually strongly undermine both models. Distance does not determine accessibility, and in fact the importance of distance appears to be more a result than cause of travel and activity behavior. Travel behavior and especially interactions between activities and time o f day are neither invisible nor irrelevant. That these results exist in a city that possesses a strongly monocentric opportunity surface and which is also currently being developed under a polycentric growth plan is striking. Further, as noted in Chapter Four, rather than distance determining behavior and accessibility, it instead appears that household characteristics and time constraints on individuals are strongly influencing the importance of distance to these individuals and households. To the extent that there are limits on how far individuals are willing to travel, this likely has more to do with their time constraints than with distance. If individuals are not willing or able to drive long distances to work or shop, it is more likely that they do not have the time available to engage in this trip than it is because of any concern over the distance. This will be discussed in more detail in the next section. One striking finding of Chapter Four is that the time of day that individuals engage in travel is clearly important due to whether services are obtainable and the likelihood of encountering congestion. For these reasons (and in part because of the 161 interactions between them) the time of day has contributed to spatial variations in accessibility. This is not possible within the timeless monocentric and polycentric models. Quite unlike the assumptions of these models, residents in cities such as Portland must schedule their activities within the constraints of their jobs, educational requirements, medical needs, and household responsibilities. Because most daily trips, including considerable discretionary travel, occurs in the afternoon and evening (Kwan, 1999b), these individuals will most likely also encounter spatially uneven traffic congestion and the need to reach certain activities before they close for the day. And time can be expected to be even more important because many travel characteristics, including the temporal scheduling of trips, varies by gender, age, and employment status (Blumen, 1994; Kwan, 1999b; Tacken, 1998). Men and women, or full and part time workers, or the elderly, are therefore unlikely to experience employment related time constraints, congestion, or the inconvenience of limited business hours, in the same way. The results also strongly support the need to examine the importance of one’s location or context within the city in order to reveal the interactions between the characteristics or behavior of individuals and households on one hand, and the urban environment in which those people must work, shop, and carry out other activities on the other. In Chapter Five the use of multilevel modeling to explain accessibility showed that the relationships evident between accessibility and individual characteristics do not (with some slight exceptions) vary among spatial scales within the Portland urban area. The use of local and regional accessibility measures (as in Handy, 1992, 1993; Handy and Niemeier 1997) is therefore unnecessary, and cities appear to be an appropriate scale of analysis for individual accessibility. Nor is there any significant variation in individual 162 accessibility among a wide range of geographic areas within Portland, so that the social and spatial hragmentation and polarization expected from postmodern urbanism is not found. Although the surfaces and charts used to visualize accessibility variations in Chapter Four showed clear variations, these do not correspond to the boundaries of the areas used with multilevel modeling. Rather, space-time accessibility measures are essentially frameless in that they appear to have little correspondence or relation to common spatial frameworks such as city boundaries and neighborhood units. As was suggested in Chapter Two, "families create their own 'cities' out of the destinations they can reach (usually travelling by car) in a reasonable length of time....The pattern formed by these destinations represents 'the city for that particular family or individual. The more varied one's destinations, the richer and more diverse is one's personal 'city" (Fishman, 1990, 38). The results of this research strongly support this view. Due to the manner in which space-time measures make use of the daily activity schedules and movements of people, each individual and household will likely possess a very different personal city than their neighbors, and these personal cities will be very unlikely to correspond to traditional boundaries or categories. In Chapter Six, incorporating characteristics of the neighborhoods in which individuals live shows that the concepts of Neo-Traditional Development may be unlikely to lead to significant variations among areas, again due to the scale- independence of accessibility. Further, this chapter also shows that local context and distance are mutually exclusive approaches. Although using distance to explain accessibility variations between neighborhoods produces statistically better results, the rather weak results of relying on distance as an explanation found in Chapter Four must 163 be kept in mind. Unlike the effects of distance, the importance of contextual characteristics varies by scale as well as with the specific accessibility measure used. Characteristics such as the proportion of a neighborhood’s housing that is detached is useful when accessibility is being explained at the neighborhood level, while commuting patterns have relationships with accessibility with the larger political and commutershed units. Although contextual characteristics were usefiil, they still do not provide the level of explanation of individual or household level attributes. Given the variation observed in accessibility patterns in Chapter Four, it is not surprising that accessibility should vary more among individuals than among areas. While access to employment or services does vary within Portland, it does so to a large extent because of individual variation, not because of distance firom specified central points or due to an uneven distribution of activity opportunities. Though this may favor the ideas of postmodern urbanism over traditional distance-based urban models, the results do not strongly support the idea of social and spatial polarization that would be expected to be present in a postmodern city. Despite the complexity o f socioeconomic variations and fundamental changes in household organization since the monocentric model became a standard representation of cities, the socioeconomic patterns of contemporary cities are not necessarily clean breaks with the past (Wyly, 1998). Tlie same may be true of accessibility, as while there is only limited evidence of the importance of distance, there is no indication that access to employment or services (at least by automobile users) has become firagmented into social or spatial extremes. Only repeated space-time studies over time would yield evidence of whether access to employment or services is getting better or worse for urban residents, and how these 164 changes vary among people and places, but any such change in access clearly cannot be assumed from changes in urban form. Additional research is also needed to understand the relationships that do exist between accessibility and contextual characteristics, as the causal connections between neighborhood characteristics and the people living there are unlikely to be direct or easy to imtangle. 7.3 Time, Distance, and Accessibility An important finding of this research is that time is important to accessibility. Not only do travel times and the time of day make significant differences to accessibility, but time constraints on individuals consistently appeared as among the most useful explanatory variables, and in fact were the only variables present within all models. As mentioned in the previous section, the important role o f time constraints on individual accessibility has a nmnber of implications for many concepts of accessibility and travel behavior, but it is also consistent with a number of observations and statements about commuting times. These include the “law of constant commuting times” (Hupkes, 1982), which holds that the amount of time spent traveling, as well as the nmnber of trips made, has remained remarkably constant over time. This occurred, despite an increase in the length of trips made, because there was a shift toward faster transport modes (such as from bicycle to car), so that more distance could be covered in the same period of time. While this observation is based on European data, it confirms that travel time is more important than distance traveled, and it is within the household that decisions regarding travel time are made. These decisions are based on a tradeoff between the utility derived from travel and the need to have time available for other activities (Hupkes, 1982). A 165 related and more recent observation within American cities is the widely observed ‘commuting paradox’, in which the length of commuting trips over time has been observed to remain stable (or even decline) over time, even though cities are getting larger and becoming more congested (Gordon, Richardson, and Jun, 1991). While this lack of change in commuting trips is held to represent proof of the development of polycentric patterns, it also provides evidence in support o f the importance of time constraints to travel behavior. As noted in Chapter Two, constant commuting times have been used to support the polycentric model because people can be expected to respond to congestion or a new workplace location by rationally relocating nearer their place of work to retain the original commuting times (Gordon, Kumar, and Richardson, 1989a, 1989b; Gordon, Richardson, and Jun, 1991; Levinson and Kumar, 1994; Levinson, 1998). However, this is not supported by actual travel behavior, as commuting patterns instead suggest that very few people are acting to minimize their journey to work by relocating either their home or workplace (Hamilton, 1982, Giuliano, 1989; Small and Song, 1992). Residential location may also influence workplace location, as there will likely be significant costs associated with moving, as well as very strong ties to a particular location (Cox and Mair, 1988; Giuliano, 1989; Hanson and Pratt, 1988; England, 1993; Hanson, Kominiak and Carlin, 1997), though a lack of suitable housing near workplaces may also be a problem for those wishing to relocate (Cervero, 1989, 1996). One detailed study in the polycentric city of Los Angeles (Wachs, et al, 1993) showed that non-work factors such as neighborhood quality, schools, and safety were slightly more important than the distance to work in decisions about whether to move, and only 26% of obseived 166 relocations actually resulted in shorter commutes. These results also supported the importance of spatial fixity in that homeowners and parents of small children were more likely to have longer commutes. Yet there is no simple relationship between commutes and relocation decisions and it is concluded that “employees for whom commute time is an important factor tend to choose housing near work; those for whom commute time is less important, tend to trade off commute time for higher neighborhood quality” (Wachs, etal, 1993, 1725). Given the importance of household time constraints on individual accessibility, it can be expected that changes to those time constraints could produce corresponding changes in accessibility. One of the most significant changes in household responsibihties in recent decades has been the entrance of large numbers of women into paid employment in recent decades. Because these women have generally remained responsible for domestic chores even while engaging in work outside the home, it can be expected that they would have considerably greater time constraints than men, and therefore less accessibility, as has in fact been found in other space-time accessibility research (Kwan, 1999a). Although the sample used in this research is roughly balanced between men and women, gender has been remarkably absent from the results, and the shorter commutes of women do not help explain the ‘commuting paradox’, as female commute lengths are actually rising (Gordon, Richardson, and Jun, 1991). Another potential source of changing time constraints is new technology, and particularly the ability of workers to substitute electronic communications for travel, whether for work, shopping, or socializing. At the forefront of these technologies are the Internet and the possibilities for telecommuting. It has been common to suggest that 167 these technologies will further reduce the power of distance in human affairs, allowing workers to relocate wherever they please without regard for distance to the workplace (Graham, 1998; Kitchin, 1998; Kotkin, 2000). This is though to lead to an emphasis on places with a high quality of life and desirable amenities. These notions have however been strongly critiqued for their technological determinism (Graham, 1998; Kitchin, 1998), as have many concepts regarding the monocentric and polycentric models. These statements also presume that distance was previously important to household and firm location decisions, and typically ignore barriers to residential or firm mobility resulting in spatial fixity. Given that locations may be relatively fixed, the arrival of new technologies may have surprisingly little impact on relocation or commute lengths. Regardless of the impact of new communication technologies on distance, there are strong reasons to believe they will greatly affect accessibility, and some have developed extensions of accessibility measures that can take the Internet or other communications technologies into account (Shen, 1998; Kwan, 2000). However, this research suggests that the importance of the Internet or other technologies might not be because they reduce the importance of distance, but because of their impact on time constraints. This is because these technologies may be eroding the importnace of fixed activities outside the home (Mitchell, 1995, 1999). According to this view employment, shopping, education, and other services have been dominated until recently by the need for synchronous communication, which requires bringing people together at the same place and time for interaction to take place. This need, which is expressed within time geography by the use of coupling constraints, holds that (for example) teachers and 168 students must interact at the same time for teaching to take place, though they do not necessarily need to be in the same room. Similar statements can be made about medical appointments or most workplace situations. However, while the possibilities for telecommunications to reduce the need for all parties to be physically present in the same room for a meeting to take place are widely recognized, it is the relaxation of the need to communicate at the same time that is actually the most important breakthrough. The Internet and other new technologies are therefore opening up greater possibilities for asynchronous communication. As the name suggest, this form of communication does not require the relevant parties to be in the same place or active at the same time for communication to take place (for example, mailing a letter or sending email, in which the sending of the message may be hours or days removed from when the message is actually received and read). This means many activities could increasingly be carried out with greater freedom not only as to where, but also as to when the activities are carried out. So for example, while attention has been given to the ability to carry out banking activities through the hitemet from home (reducing the need to overcome the distance to the bank), it is therefore actually the ability to reschedule these banking activities to any time of the day that is most important. Rather than having to resolve financial matters during a very limited span of time during the weekday, banking could be carried out on weekends or during the middle of the night. Given the clear importance of household time constraints on accessibility, this ability to reschedule many activities to a more convenient time is likely to be of far more 169 usefulness than the ability to substitute electronic communication for travel. The importance of fixed business hours, which this research has shown to be an important influence on individual accessibility, is therefore greatly reduced. The ability of new technology to reduce time constraints is therefore more likely to make significant changes in accessibility than any ability to reduce the need for travel (or the importance of distance). This is in contrast to the expectations of prevailing urban models about the effects of the Internet, Intelligent Transportation Systems or telecommuting on mobility and accessibility (Hodge, Morrill, and Stanilov, 1996; Nilles, 1991; Hanson, 1998; Graham, 1998). The monocentric and polycentric perspectives would expect these technologies to reduce travel times or the need for travel, and therefore reduce the density of the city by allowing it to expand outwards (while retaining constant commuting times). The opposite would apply in the case of congestion (Hodge, 1992), but neither situations would alter the fundamental logic of these models (though of course the evidence of commuting behavior and residential mobility strongly disputes this likelihood, while the Taw of constant travel times’ suggests that a shorter commute will be offset by additional discretionary travel). The notions of postmodern urbanism would Likely lead to a very different perspective about these technologies, as rather than affecting all peoples and places equally the effect of these new technologies would be expected to further increase social and spatial inequalities by selectively benefiting certain social groups and places, and thereby increasing the gap between those with and without accessibility (Hanson, 1998). This social and spatial polarization would increase the importance of place while also making explicit the lack of equality found within contemporary cities. But again, 170 however, the emphasis would very likely be on distance rather than time. For this reason, even to the extent to which the concepts of postmodern urbanism may be useful for understanding urban form (which is debatable) or individual accessibility (which appears doubtful), these ideas may be very unlikely to successfully predict exactly whose access will benefit from these technologies, and why. As noted in the first chapter, our concepts of accessibility are based on the need to overcome distance in order to get to work or carry out other necessary activities. Given changes in the speciahzation o f urban land uses and the changing patterns o f daily life, the concept of intraurban accessibihty may be limited to the period following the interior differentiation of cities in the 19th century and before some (still hypothetical) future time when communications technology eliminates the need for movement or allows for 'virtual' travel with few impediments. In the meantime, a considerable amount of daily travel is still required by most people, so that the concept of accessibility will remain important to daily life for some time to come. For this reason it is important that the way individuals construct their accessibility, and its relation to time constraints and choices about activity patterns, be examined. 7.4 Significance of this Research and Future Directions The issues discussed in this chapter are important not just for their empirical statements about cities and accessibility but also for their theoretical and methodological contributions to accessibility studies. As discussed in section 7.1, this dissertation strongly questions the importance of distance as an influence on individual accessibility, as well as statements about the importance of land use to accessibility. Regardless of 171 which urban model is considered to best explain the form of Portland or other American cities, none appear to make adequate statements about individual accessibihty. As a result, the use of space-time measures of accessibihty is warranted despite their greater complexity and computation requirements than conventional proximity-based measures. The finding that these measures are scale-independent or frameless is especially important, as it indicates that accessibihty is not something that can be represented as existing at a specific scale or easily confined to a particular distance or spatial container. The use of time-geographic concepts has been greatly facilitated by the coupling of these ideas with travel-activity diary data sets (Villoria, 1989; Kwan and Hong, 1998) as well as the implementation of these space-time measures within GIS (Miller, 1991; Kwan and Hong, 1998). This research has built upon these foundations not only by establishing the use of network based Potenhal Path Area measures with large data sets but also by incorporating time into these measures. Time is clearly a fundamental component of accessibihty, but the importance o f time is bound up with geography. The mobility of people is dependent in part on where people live and carry out activities, and therefore the roadways available to them, while the limited availability of businesses during the day further restricts access. Congestion wiU have an effect on mobihty and the amount of time available for other activities, but again will not be distributed evenly within a city. The incorporation of both congestion and limited business hours, which have not previously been used with these measures except in hypothetical situations (for example. Miller, 1982), is therefore of considerable significance. This dissertation also provides a useful example of how time can be incorporated into GIS-based 172 methodologies, a topic that has received considerable attention in the last decade but which appears to have resisted widespread implementation (Langran, 1992; Peuquet, 1994, 1999; Peuquet and Duan, 1995). Further, making use of multilevel modeling with individual and area level data has allowed statements to be made about the relative contribution of individual activity behavior and the role of the urban environment, an important topic that has received considerable attention for its impacts on transportation management, urban planning, and the environment (Boamet and Crane, 2001). However, this research does not strongly support the expectations of this literature by finding only weak relationships between accessibility and the characteristics of the surrounding urban environment. This finding that space-time measures of individual accessibility are not dependent on particular spatial scales and are only weakly related to characteristics of particular areas within cities is significant because it confirms that these measures provide a radically different way of looking at cities than that provided by conventional accessibility measures or urban models. Several topics can be identified as promising for future research. Because individuals respond to congestion in a variety of ways (Mokhtarian, Raney and Salomon, 1997; Raney, Mokhtarian and Salomon, 2000), the issue of how individuals make use of time and engage in adaptive behavior to maintain levels of accessibility in the face of congestion and changes in facihty hours should be examined. Household time constraints will be central to this issue, as they will likely limit responses to these conditions and may further increases among individuals and potentially also among areas within the city. Examining the importance of time to accessibility could also take 173 advantage of new data sources and methodologies, such as the use of GPS technology to precisely identify the routes traveled and further refine the computation of network Potential Path Areas. More detailed opportunity data could allow the disaggregation of accessibihty to determine how access varies to specific types of services. The possibihties for incorporating temporally specific (or real time) traffic flow or speed data within GIS (for example, Claramunt, Jiang, and Bargjela, 2000; Quiroga, 2000; Taylor, Woolley, and Zito, 2000) opens up opportunities for more precisely evaluating the temporal dynamism of accessibility, and even for estimating the potential impacts on accessibility of congestion occurring at particular times and places. Given the results achieved here with relatively simple estimates of travel times and congestion, it is likely that additional data would yield considerable additional insights into personal accessibility. Doing so would also take advantage of the data sources and methodologies found within the developing field of transportation studies within GIS (or GIS-T) (Waters, 1999; Thill, 2000), as well as helping to integrate space-time accessibility concepts with other GlS-T research. It is also important to incorporate additional transport modes into the idea of individual accessibility. Allowing transit modes, bicycling, or walking to be used opens up additional possibihties for studying accessibihty fi"om a space-time perspective, but has not yet been carried out in an American context. Accessibihty within radial transit networks will almost certainly be more structured than for auto travel, and will also be far more sensitive to time due to the rehance on fixed schedules. The use of transit can also be expected to increase differences among gender, income, age, or racial lines. As these individual attributes carmot necessarily be isolated fi-om employment status, income, 174 commuting patterns, mode choice and therefore accessibihty, incorporating non-auto models is therefore necessary in order to fully evaluate accessibihty differences between individuals and households. Given that this dissertation has found minimal differences in accessibihty when using only automobile travel, extending this research to other means of travel may therefore considerably alter the results. Within the context of Portland, making use of multiple modes is also vital because the 2040 Growth Plan specifically includes and promotes non-auto transport modes. As noted in this research, while accessibihty patterns in Portland do not support certain notions o f the 2040 plan, it can he argued that this is less of a problem than it appears because many elements of this plan (such as the importance of regional centers) focus on non-auto trips. The limits of mobihty when traveling by bus, train, bicycle, or walking at various times of the day or week will likely reveal entirely new and unexpected geographies of accessibility. Given the importance of accessibihty to all urban residents in the conduct of their everyday lives, these geographies should be examined in order to better understand the present patterns of access to jobs, education, medical care, recreational facihties, or shopping (regardless of what our urban models may tell us), and more efficiently plan for the future. 175 BIBLIOGRAPHY Abbott, Carl. 1983. Portland: Planning. Politics, and Growth in a Twentieth-Century City. Lincoln: University o f Nebraska Press. Adams, John S. 1970. Residential structure of Midwestern cities. Annals o f the Association o f American Geographers 60: 37-62. Adler, Sy. 1991. The transformation of the Pacific Electric Railway. Urban Affairs Quarterly 27: 51-86. Adler, Sy. 1988. A comparative analysis of rail transit politics, policy and planning in Canada and the United States. Logistics and Transportation Review 24: 265-279. Adler, Sy. 1987. Why BART but no LART? The political economy of rail rapid transit planning in the Los Angeles and San Francisco metropolitan areas, 1945-57. Planning Perspectives 2: 149-174. Adler, Sy. 1986. The dynamics of transit innovation in Los Angeles. Environment and Planning D4: 321-335. Aitken, Stuart; Stutz, Frederick; Prosser, Rudy; and Chandler, Russell. 1993. Neighborhood integrity and resident’s familiarity: using a geographic information system to investigate place identity. Tijdschrift voor Economische en Sociale Geografie 84: 2-12. Aitken, Stuart, and Prosser, Rudy. 1990. Residents’ spatial knowledge of neighborhood continuity and form. Geographical Analysis 22: 301-325. Alonso, W. 1964. Location and Land Use: Toward a General Theory o f Land Rent. Cambridge: Harvard University Press. Archer, Wayne R., and Smith, Marc T. 1993. Why do suburban offices cluster? Geographical Analysis 25: 53-64. 176 Axhausen, K.W. 1995. Travel Diaries: An Annotated Catalogue. Unpublished draft paper. Baerwald, Thomas J. 1989. The evolution of suburban downtowns in Midwestern metropolises. In Suburbia Re-examined, ed. Barbara M. Kelly. New York: Greenwood Press. Baerwald, Thomas J. 1982. Land use change in suburban clusters and corridors. Transportation Research Record 861: 7-12. Baerwald, Thomas J. 1978. The emergence of a new “downtown.” Geographical Review 68: 308-318. Beauregard, Robert A. 1999. Break dancing on Santa Monica Boulevard. Urban Geography 20: 396-399. Berry, B.J.L., and Kim, H.M. 1993, Challenges to the monocentric model. Geographical Analysis 25: 1-4. Black, L, and Conroy, M. 1977. Accessibility measures and the social evaluation of urban structure. Environment and Planning 9: 1013-1031.A Blumen, Oma. 1994. Gender differences in the journey to work. Urban Geography 15 : 223-245. Boamet, Marlon G., and Crane, Randall. 2001. Travel by Design: the Influence o f Urban Form on Travel. Oxford: Oxford University Press. Bogart, William T., and William C. Ferry. 1999. Employment centers in greater Cleveland: evidence of evolution in a formerly monocentric city. Urban Studies 36: 2099-2110. Bookout, L.W. 1992. Neotraditional town planning: a new vision for the suburbs? Urban Land 51: 20-26. Brainard, Juli; Andrew Lovett; and Ian Bateman. 1997. Using isochrone surfaces in travel cost models. Journal o f Transport Geography 117-126. 5: Brainard, Juli; Andrew Lovett; and Ian Bateman. 1999. Integrating geographical information systems into travel cost analysis and benefit transfer. International Journal o f Geographical Information Science 13: 227-246. Bullen, N.; Jones, K.; and Duncan, C. 1997. Modelling complexity: analysing between- individual and between-place variation- a multilevel tutorial. Environment and Planning 29:A 585-609. 177 Burgess, Ernest W. 1925. The growth of the city: an introduction to a research project. In The City, ed. Robert E. Park, Ernest W. Burgess, and Roderick D. McKenzie. Chicago: University of Chicago Press. Bums, Lawrence D. 1979. Transportation, temporal, and spatial components of accessibility. Lexington, MA: Lexington Books. Cambridge Systematics, hic. 1996. Data Collection in the Portland, Oregon Metropolitan Area Case Study. Washington, DC: US Department of Transportation. Carrothers, Gerald A. P. 1956. An historical review of the gravity and potential concepts of human interaction. Journal o f American Institute o f Planners 94-102. 22: Castells, M. 1989. The Informational City: Information Technology, Economic Restructuring, and the Urban-Regional Process. Oxford: Basil Blackwell. Cervero, Robert. 1996. Jobs-housing balance revisited: trends and impacts in the San Francisco Bay area. Journal of the American Planning Association 62: 492-511 Cervero, Robert. 1989. Jobs-housing balance and regional mobility. Journal of the American Planning Association 55: 136-150. Choi, Keechoo, and Jang, Wonjae. 2000. Development o f a transit network from a street map database with spatial analysis and dynamic segmentation. Transportation Research C 8: 129-146. Claramunt, C.; Jiang, B.; and Bargiela, A. 2000. A new framework for the integration, analysis and visualization of urban traffic data within geographic information systems. Transportation Research C 167-184. Clark, W.A.V., and Burt, J.E. 1980. The impact o f workplace on residential location. Annals o f the Association ofAmerican Geographers 70: 59-67. Cooke, Donald F. 1998. Topology and TIGER: the Census Bureau’s contribution. In The History o f Geographic Information Systems: Perspectives From the Pioneers, ed. Timothy W. Foresman. Upper Saddle River, NJ: Prentice Hall. Coombes, M., and Raybould, S. 1997. Modelling the influence of individual and spatial factors underlying variations in the levels of secondary school examination results. Environment and Planning 29: A641-658. 178 Cox, Kevin R., and Mair, Andrew. 1988. Locality and community in the politics of local economic development. Annals o f the Association o f American Geographers 78: 307-325. Crane, Randall. 1996. On form versus function: will the new urbanism reduce traffic, or increase it? Journal o f Planning Education and Research 15: 117-126. Davis, Mike. 1998. Ecology o f Fear: Los Angeles and the Imagination o f Disaster. New York: Henry Holt and Company. Dear, Michael. 2000. The Postmodern Urban Condition. Oxford: Blackwell. Dear, Michael, and Flusty, Steven. 1999. Engaging postmodern urbanism. Urban Geography 20:412-416. Dear, Michael, and Flusty, Steven. 1998. Postmodern urbanism. Annals o f the Association o f American Geographers 88: 50-72. Dear, Michael, and Flusty, Steven. 1997. The iron lotus: Los Angeles and postmodern urbanism. The Annals o f the American Academy o f Political and Social Sciences 551: 151-163. Dowling Associates. 1997. Travel Model Speed Estimation and Post Processing Methods for Air Quality Analysis. Washington, DC: United States Department of Transportation. Duncan, Craig; Kelvyn Jones; and Graham Moon. 1996. Social Science and Medicine 42: 817-830. England, Kim V.L. 1995. “Girls in the office”: recruiting and job search in a local clerical labor market. Environment and Planning 27: A 1995-2018. England, Kim V.L. 1993. Suburban pink collar ghettos: the spatial entrapment of women? Annals o f the Association o f American Geographers 83: 225-242. England, Kim V.L. 1991. Gender relations and the spatial structure of the city. Geoforum 22: 135-147. Erickson, Rodney A. 1986. Multinucleation in metropolitan economies. Annals o f the Association o f American Geographers 76: 331-346. Erickson, Rodney A., and Gentry, Marylynn. 1985. Suburban nucléations. Geographical Review 75: 19-31. 179 Ewing, Reid; Padma Haliyur; and G. William Page. 1994. Getting around a traditional city, a suburban planned unit development, and everything in between. Transportation Research Record 1466: 53-62. Fishman, Robert. 1990. Megalopohs unbound. The Wilson Quarterly 14: 25-45. Fishman, Robert. 1987. Bourgeois Utopias: The Rise and Fall o f Suburbia. New York: Basic Books. Fletcher, David. 1987. Modelling GIS transportation networks. In. Papers From the Annual Conference o f the Urban and Regional Information Systems Association (URISA). Washington, DC: URISA. Fohl, Peter; Curtin, Kevin M.; Goodchild, Michael F.; and Church, Richard L. 1996. A non-planar, lane-based navigable data model. In Advances in GIS Research II: Proceedings of the Serventh International Symposium on Spatial Data Handling, ed, M. J. KraaJc and M. Mole Naar. London: Taylor and Francis. Forer, Pip, and Huisman, Otto. 1998. Space, time and sequencing: substitution at the physical/virtual interface. Paper presented at the Varenius specialist meeting on Measuring and Representing Accessibility in the Information Age, November 20- 22, Asilomar, California. Forer, P.C., and Kivell, Helen. 1981. Space-time budgets, public transport, and spatial choice. Environment and Planning 13: A497-509. Forstall, Richard L., and Richard P. Greene. 1997. Defining job concentrations: the Los Angeles case. Urban Geography 18: 705-739. Frank, Lawrence D., and Pivo, Gary. 1994. Impacts o f mixed use and density on utilization of three modes of travel: single-occupant vehicle, transit, and walking. Transportation Research Record 1466: 44-52. Friedman, Bruce; Gordon, Stephen P.; and Peers, John B. 1994. Effect of neotraditional neighborhood design on travel characteristics. Transportation Research Record 1466: 63-70. Fujii, T., and Hartshorn, Truman A. 1995. The changing metropolitan structure of Atlanta, Georgia: locations of fimctions and regional structure in a multinucleated urban area. Urban Geography 16: 680-707. Garreau, Joel. 1991. Edge City: Life on the New Frontier. New York: Doubleday. Garrison, William L. 1960. Connectivity of the Interstate Highway System. Papers and Proceedings o f the Regional Science Association 6: 121-137. 180 Gauthier, Howard L. 1968. Transportation and the growth of the Sao Paulo economy. Journal o f Regional Science 8: 77-94. Geertman, Stan C.M., and Ritsema Van Eck, Jan R. 1995. GIS and models of accessibility potential: an application in planning. International Journal o f Geographical Information Systems 9: 67-80. Giuliano, Genevieve. 1995. The weakening transportation-land use connection. Access 6: 3-11. Giuliano, Genevieve. 1991. Is jobs-housing balance a transportation issue? Transportation Research Record 1305: 305-312. Giuliano, Genevieve. 1989. New directions for understanding transportation and land use. Environment and Planning A 145-159. 2\'. Giuliano, Genevieve, and Small, Kenneth A. 1993. Is the journey to work explained by urban structure? Urban Studies 30: 1485-1500. Giuliano, Genevieve, and Small, Kenneth A. 1991. Subcenters in the Los Angeles region. Regional Science and Urban Economics 21: 163-182. Gober, Patricia, and Behr, Michelle. 1982. Central cities and suburbs as distinct place types: myth or fact? Economic Geography 58: 371-385. Goldstein, Harvey. 1998. Multilevel Statistical Models, Second Edition. London: Edward Arnold. Goodchild, Michael F., and Janelle, Donald G. 1984. The city around the clock: space time patterns of urban ecological structure. Environment and Planning 16: A 807- 820. Gordon, P.; Kumar, A.; and Richardson, H.W. 1989a. The influence of metropolitan spatial structure on commuting time. Journal o f Urban Economics 26: 138-151. Gordon, P.; Kumar, A.; and Richardson, H.W. 1989b. Congestion, changing metropolitan structure, and city size in the United States. International Regional Science Review 12: 45-56. Gordon, P., and Richardson, H.W. 1996. Beyond polycentricity: the dispersed metropolis, Los Angeles, 1970-1990. Journal o f the American Planning Association 62: 289-295. 181 Gordon, P.; Richardson, H.W.; and Jun, M J. 1991. The commuting paradox: evidence from the top twenty. Journal o f the American Planning Association 57: 416-420. Gordon, P.; Richardson, H.W.; Wong, H.L. 1986. The distribution o f population and employment in a polycentric city: the case of Los Angeles. Environment and Planning A 18: 161-173. Gordon, P., and Wong, H.L. 1985. The costs of urban sprawl: some new evidence. Environment and Planning 17: A661-666. Graham, Stephen. 1998. The end of geography or the explosion of place? Conceptualizing space, place, and information technology. Progress in Human Geography 22: 165-185. Greene, David L. 1980. Recent trends in urban spatial structure. Growth and Change 11:29-40. Guy, C.M. 1983. The assessment of access to local shopping opportunities: a comparison of accessibility measures. Environment and Planning 10: B219-238. Hagerstrand, Torsten. 1970. What about people in regional science? Papers o f the Regional Science Association 24: 7-21. Hall, Tim. 1998. Urban Geography. London: Routledge. Hamilton, Bruce W. 1982. Wastefril commuting. Journal of Political Economy 9Q: 1035-1053. Handy, Susan. 1996a. Methodologies for exploring the link between urban form and travel behavior. Transportation Research 1: D 151-165. Handy, Susan. 1996b. Understanding the link between urban form and nonwork travel behavior. Journal o f Planning Education and Research 15: 183-198. Handy, Susan. 1993. A cycle of dependence: automobiles, accessibility, and the evolution of the transportation and retail hierarchies. Berkeley Planning Journal 8: 21-43. Handy, Susan. 1992. Regional versus local accessibility: neo-traditional development and its implications for non-work travel. Built Environment 18: 253-267. Handy, Susan, and Niemeier, D.A. 1997. Measuring accessibility: an exploration of issues and alternatives. Environment and Planning 29: A 1175-1194. Hanson, Susan. 1998. Off the road? Reflections in transportation geography in the information age. Journal o f Transport Geography 6: 241-249. 182 Hanson, Susan; Kominiak, Tara; and Carlin, Scott. 1997. Assessing the impact of location on women’s labor market outcomes: a methodological exploration. Geographical Analysis 29: 281-297. Hanson, Susan, and Pratt, Geraldine. 1992. Dynamic dependencies: a geographic investigation of local labor markets. Economic Geography 68: 373-405. Hanson, Susan, and Pratt, Geraldine. 1988. Reconceptualizing the links between home and work in urban geography. Economic Geography 64: 299-321. Harris, Chauncy D., and Ullman, Edward L. 1945. The nature o f cities. The Annals of the American Academy o f Political and Social Sciences 242: 7-17. Hartshorn, Truman A., and Muller, Peter O. 1992. The suburban downtown and urban economic development today, hi Sources o f Metropolitan Growth, ed. Edwin S. Mills and John F. McDonald. New Brunswick, NJ: Center for Urban Policy Research. Hartshome, Truman A., and Muller, Peter O. 1989. Suburban downtowns and the transformation of metropolitan Atlanta’s business landscape. Urban Geography 10: 375-395. Heikkila, E.; Gordon, P.; Kim, J.I.; Peiser, R.B.; and Richardson, H.W. 1989. What happened to the CBD-distance gradient?: land values in a policentric city. Environment and Planning 21: A221-232. Helling, Amy. 1998. Changing intra-metropolitan accessibility in the U.S.: evidence from Atlanta. Progress in Planning 49: 55-108. Hoch, Irving, and Waddell, Paul. 1993. Apartment rents: another challenge to the monocentric model. Geographical Analysis 25: 20-31. Hodge, David C. 1995. My fair share: equity issues in urban transportation. In The Geography o f Urban Transportation, ed. Susan Hanson. New York: The Guilford Press. Hodge, David C. 1992. Urban congestion: reshaping urban life. Urban Geography 13: 577-588. Hodge, David C. 1990. Geography and the political economy of urban transportation. Urban Geography 11: 87-100. Hodge, D C.; Morrill, R.L.; and Stanilov, K. 1996. hnphcations of intelligent transportation systems for metropolitan form. Urban Geography 17: 714-739. 183 Holzer, Harry J. 1991. The spatial mismatch hypothesis: what has the evidence shown? Urban Studies 28: 105-122. Hoyt, Homer. 1939. The Structure and Growth o f Residential Neighborhoods in American Cities. Washington, DC: Federal Housing Administration. Huisman, Otto, and Forer, Pip. 1998a. Towards a geometric frameworkfor modeling space-time opportunities and interaction potential. Paper presented at the hitemational Geographical Union Commission on Modelling Geographical Systems Meeting, August 28-29, Lisbon, Portugal. Huisman, Otto, and Forer, Pip. 1998b. Computational agents and urban life spaces: a preliminary realization o f the time-geography o f student lifestyles. Manuscript paper. Hupkes, Geurt. 1982. The law o f constant travel time and trip-rates. Futures 14: 38-46. Ingram, D.R. 1971. The concept of accessibility: a search for an operational form. Regional Studies 5: 101-107. Jackson, Peter. 1999. Postmodern urbanism and the ethnographic void. Urban Geography 20: 400-402. Jones, Kelvyn. 1991a. Specifying and estimating multi-level models for geographical research. Transactions o f the Institute o f British Geographers 16: 148-160. Jones, Kelvyn. 1991b. Multi-Level Models for Geographical Research. Norwich: Environmental Publications. Jones, Kelvyn, and Bullen, Nina. 1993. A multi-level analysis of the variations in domestic property prices: Southern England, 1980-87. Urban Studies 30: 1409- 1426. Jones, Kelvyn, and Bullen, Nina. 1994. Contextual models of urban house prices: a comparison of fixed and random-coefficient models developed by expansion. Economic Geography 70: 252-272. Jones, Kelvyn, and Duncan, Craig. 1996. People and places: the multilevel model as a general firamework for the quantitative analysis of geographical data. In Spatial Analysis: Modelling in a GIS Environment, ed. Paul Longley and Michael Batty. Cambridge: Geoinformation International. 184 Jones, Kelvyn, and Duncan, Craig. 1995. Individuals and their ecologies: analyzing the geography of chronic illness within a multilevel modeling framework. Health and Place 1: 27-40. Jones, Kelvyn; Gould, Miles I.; and Watt, Robert. 1998. Multiple contexts as cross classified models: the labor vote in the British general election of 1992. Geographical Analysis 30: 65-93. Kersten, Earl W., Jr., and Ross, D. Reid. 1968. Clayton: a new metropolitan focus in the St. Louis area. Annals o f the Association o f American Geographers 58: 637- 649. Kim, Kyung-Hwa. 2000. Personal communication with Transportation Planning Supervisor at Metro, November 16, 2000. Kitamura, Ryuichi; Mokhtarian, Patricia L.; and Laidet, Laura. 1997. A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay area. Transportation 24: 125-158. Kitchin, Robert M. 1998. Towards geographies of cyberspace. Progress in Human Geography 22: 385-406. Kloster, Tom. 2000. Personal communication with Senior Transportation Program Supervisor at Metro, November 16, 2000. Knox, Paul L. 1997. Globalization and urban economic change. The Annals o f the American Academy o f Political and Social Sciences 551: 17-27. Knox, Paul L. 1991. The restless urban landscape: economic and sociocultural change and the transformation of metropolitan Washington, DC. Annals o f the Association o f American Geographers 81: 181-209. Knox, Paul L. 1982. Residential structure, facility location and patterns of accessibility. In Conflict, Politics and the Urban Scene, ed. Kevin R. Cox and R.J. Johnston. New York: St. Martin’s Press. Knox, Paul L. 1978. The intraurban ecology of primary medical care: patterns of accessibility and their policy implications. Environment and Planning 10: A415- 435. Krakover, Shaul, and Casetti, Bmilo. 1988. Directionally biased metropolitan growth: a model and case study. Economic Geography 64: 17-28. Krefr, Ita, and De Leeuw, Jan. 1998. Introducing Multilevel Modeling. London: SAGE Publications. 185 Kwan, Mei-Po. 1998. Space-time and integral measures of individual accessibility: a comparative analysis using a point-based framework. Geographical Analysis: 30:191-217. Kwan, Mei-Po. 1999a. Gender and individual access to urban opportunities: a study using space-time measures. Professional Geographer 51: 210-227. Kwan, Mei-Po. 1999b. Gender, the home-work link, and space-time patterns of nonemployment activities. Economic Geography 370-394.75: Kwan, Mei-Po. 2000. Human extensibility and individual hybrid-accessibihty in space time: a multi-scale representation using GIS. In Information, Place, and Cyberspace: Issues in Accessibility, ed. Donald Janelle and David Hodge. Berhn: Springer-Verlag. Kwan, Mei-Po, and Hong, Xiao-Dong. 1998. Network-based constraints-oriented choice set formation using GIS. Geographical Systems 5: 139-162. Lake, Robert W. 1999. Postmodern urbanism? Urban Geography 20: 393-395. Lam, Nina Siu-Ngan. 1983. Spatial interpolation methods: a review. The American Cartographer 10: 129-149. Langran, G. 1992. Time in Geographic Information Systems. London: Taylor and Francis. Lee, Keumsook, and Hee-Yeon Lee. 1998. A new algorithm for graph-theoretic nodal accessibility measurement. Geographical Analysis 30: 1-14. Lenntorp, Bo. 1978. A time-geographic simulation model of individual activity programmes. In Human Activity and Time Geography, ed. Tommy Carlstein, Don Parkes, and Nigel Thrift. London: Edward Arnold. Lenntorp, Bo. 1976. Paths in Space-Time Environments: a Time-Geographic Study of the Movement Possibilities of Individuals. Lund Studies in Geography B: Human Geography. Lund: Gleerup. Lewis, Paul G. 1996. Shaping Suburbia: How Political Institutions Organize Urban Development. Pittsburgh: University of Pittsburgh Press. Levinson, DM. 1998. Accessibihty and the journey to work. Journal o f Transport Geography 6: 11-21. 186 Levinson, D M. and Kumar, A. 1994. The rational locator: why travel times have remained stable. Journal o f the American Planning Association 60: 319-332. Lloyd, William J. 1981. Understanding late nineteenth-century American cities. Geographical Review 71: 460-471. Lowe, John C. 1998. Patterns of spatial dispersion in metropolitan commuting. Urban Geography 19: 232-253. Loytonen, Markku. 1998. GIS, time geography and health. In. GIS and Health, eà. Anthony C. Gatrell and Markku Loytonen. London: Taylor and Francis. Mainguenaud, M. 1995. Modelling the network component of geographical information systems. International Journal o f Geographical Information 9:Systems 575-593. Marchand, Bernard. 1973. Deformation of a transportation surface. .4n«a/5 Association o f American Geographers 63: 507-521. Marshall, Alex. 2000. How Cities Work: Suburbs, Sprawl, and the Roads Not Taken. Austin: University of Texas Press. Martin, David. 1996. Depicting changing distributions through surface estimation. In Spatial Analysis: Modelling in a GIS Environment, ed. Paul Longley and Michael Battey. New York: John Wiley and Sons. Marx, Robert W. 1986. The TIGER system: automating the geographic structure of the United States Census. Government Publications Review 13: 181 -201. McDonald, John F. 1987. The identification of urban employment subcenters. Journal o f Urban Economics 21: 242-258. McDonald, John F., and McMillen, D P. 1990. Employment subcenters and land values in a polycentric urban area: the case of Chicago. Environment and Planning A 22: 1561-1574. McDonald, John F., and Prather, Paul J. 1994. Suburban employment centers: the case of Chicago. Urban Studies 31: 201-218. McLafferty, Sara. 1982. Urban structure and geographical access to public services. Annals o f the Association o f American Geographers 72: 347-354. McLafferty, Sara, and Preston, Valerie. 1992. Spatial mismatch and labor market segmentation for Afiican-American and Latina women. Economic Geography 68:406-431. 187 Metro (Metropolitan Service District). 2001. http://www.metro-region.org/ Metro (Metropolitan Service District). 1997. Regional Framework Plan. Portland: Metropolitan Service District. Miller, Harvey J. 1999. Measuring space-time accessibility benefits within transportation networks: basic theory and computational procedures. Geographical Analysis 3\: 1-26. Miller, Harvey. 1991. Modelling accessibility using space-time prism concepts within geographical information systems. International Journal o f Geographical Information Systems 5: 287-301. Miller, Harvey J., and Storm, John D. 1996. Geographic Information System Design for Network Equilibrium-Based Travel Demand Models. Transportation Research C 6: 373-389. Miller, Roger. 1982. Household activity patterns in nineteenth-century suburbs: a time- geographic exploration. Annals of the Association of American Geographers 72: 355-371. Mitchell, WilUam J. 1995. City o f Bits: Space, Place, and the Infobahn. Cambridge: The MIT Press. Mitchell, William J. 1999. E-topia. Cambridge: MTT Press. Mokhtarian, Patricia L.; Raney, Ehzabeth A.; and Salomon, Han. 1997. Behavioral response to congestion: identifying patterns and socio-economic differences in adoption. Transport Policy 147-160. Morris, J.M.; Dumble, P. L.; and Wigan, M.R. 1979. Accessibility indicators for transport planning. Transportation Research 13: A 91-109. Muller, Peter O. 1997. The suburban transformation o f the globalizing American city. The Annals o f the American Academy ofPolitical and Social Sciences 551: 44-58. Muller, Peter O. 1995. Transportation and urban form: stages in the spatial evolution of the American metropohs. In The Geography o f Urban Transportation, ed. Susan Hanson. New York: The Guilford Press. Muller, Peter O. 1989. The transformation of bedroom suburbia into the outer city: an overview of metropolitan structural change since 1947. In Suburbia Re examined, ed. Barbara M. Kelly. New York: Greenwood Press. Muraco, William A. 1972. Intraurban accessibility. Economic Geography 4?,: 188 Murayama, Yuji. 1994. The impact of railways on accessibility in the Japanese urban system. Journal o f Transport Geography 2: 87-100. Muth, R.F. 1985. Models o f land-use, housing, and rent: an evaluation. Journal o f Regional Science 25: 593-606. Muth, R.F. 1969. Cities and Housing: The Spatial Pattern o f Urban Residential Land Use. Chicago: University of Chicago Press. Nelson, Arthur C. 1994. Oregon's urban growth boundary policy as a landmark planning tool. In Planning the Oregon Way: A Twenty Year Evaluation, ed. Carl Abbott, Deborah Howe, and Sy Adler. Corvallis, OR: Oregon State University. Newsome, Tracy H.; Walcott, Wayne A.; and Smith, Paul D. 1998. Urban activity spaces: illustrations and application of a conceptual model for integrating the time and space dimensions. Transportation 25: 357-377. Nilles, J.M. 1991. Telecommuting and urban sprawl: mitigator or inciter? Transportation 18: 411-432. O’hUllachain, Breandon, and Reid, Neil. 1992. The intrametropolitan location of services in the United States. Urban Geography 13: 334-354. Openshaw, Stan. 1996. Developing GIS-relevant zone-based spatial analysis methods. In Spatial Analysis: Modelling in a GIS Environment, ed. Paul Longley and Michael Battey. New York: John Wiley and Sons. Paterson, Lindsay, and Goldstein, Harvey. 1991. New statistical models for analysing social structures: an introduction to multilevel models. British Educational Research Journal 17: 387-393. Peuquet, Donna J. 1999. Time in GIS and geographical databases. In Geographical Information Systems, Volume I: Principles and Technical Issues, Second Edition, ed. Paul A. Longley, Michael Goodchild, David J. Maguire, and David W. Rhind. New York: John Wiley & Sons. Peuquet, Dorma J. 1994. It’s about time: a conceptual framework for the representation of temporal dynamics in Geographic Information Systems. Annals o f the Association o f American Geographers 84: 441-461. Peuquet, Donna J., and Duan, Niu. 1995. An event-based spatiotemporal data model (ESTDM) for temporal analysis of geographical data. International Journal o f Geographical Information Systems 9: 7-24. 189 Pickus, J., and Gober, p. 1988. Urban villages and activity patterns in Phoenix. Urban Geography 9: 85-97. Pirie, Gordon H. 1979. Measuring accessibility: a review and a proposal. Environment and Planning A 11: 299-312. Pivo, Gary. 1993. A taxonomy of suburban office clusters: the case o f Toronto. Urban Studies 30: 31-49. Pivo, Gary. 1990. The net o f mixed beads: suburban office development in six metropolitan regions. Journal o f the American Planning Association 56: 457-469. Pisarki, Alan E. 1996. Commuting in America II: The Second National Report on Commuting Patterns and Trends. Lansdowne, VA: Eno Transportation Foundation. Plane, David A. 1986. The geography of urban commuting fields: some empirical evidence firom New England. Professional Geographer 32: 182-188. Pooler, James. 1987. Measuring geographical accessibility: a review o f current approaches and problems in the use of population potentials. Geoforum 18: 269- 289. Pred, Allan, and Palm, Risa. 1978. The status of women: a time-geographic view. In An Invitation to Geography, ed. David Lanegran and Risa Palm. New York: McGraw-Hill. Quiroga, Cesar A. 2000. Performance measures and data requirements for congestion management systems. Transportation Research 8: C 287-306. Raney, Elizabeth A.; Mokhtarian, Patricia L.; and Salomon, Han. 2000. Modeling individuals’ consideration of strategies to cope with congestion. Transportation Research F 3: 141-165. Recker, W.W.; Chen, C.; and McNally, M.G. 2001. Measuring the impact o f efficient household travel decisions on potential travel time savings and accessibility gains. Transportation Research 35:A 339-369. Reidl, Leop. 1998. Pycnophylactic interpolation script. http://gis.esri.com/arcscripts/scripts.cfin Rollinson, Paul A. 1998. The everyday geography of the homeless in Kansas City. Geografiska Annaler 80B: 101-115. 190 Rutheîser, Charles. 1996. Imagineering Atlanta: The Politics o f Place in the City o f Dreams. London: Verso. Schaerstrom, Anders. 1996. Pathogenic Paths? A Time Geographical Approach in Medical Geography. Lund: Lund University Press. Scott, Lauren M. 1998. Evaluating intra-metropolitan accessibility in the information age: operational issues, objectives, and implementation. Paper presented at Varenius conference, November 20-22, 1998, Pacific Grove, CA. Shaw, Shih-Lung. 1993. GIS for urban travel demand analysis: requirements and alternatives. Computers, Environment, and Urban Systems 17: 15-29. Shen, Qing. 1998. Spatial technologies, accessibility, and the social construction of urban space. Computers, Environment, and Urban Systems 22: 447-464. Small, Kenneth A., and Song, Shunfeng. 1994. Population and employment densities: structure and change. Journal o f Urban Economics 36: 292-313. Small, Kenneth A., and Song, Shunfeng. 1992. “Wasteful commuting”: a resolution. Journal o f Political Economy 100: 888-898. Song, Shunfeng. 1992. Spatial Structure and Urban Commuting. Berkeley: University of California Transportation Center Working Paper No. 117. Spence, Nigel, and Linneker, Brian. 1994. Evolution of the motorway network and changing levels o f accessibility in Great Britain. Journal of Transport Geography 2: 247-264. Sperling, Jonathan. 1995. Development and maintenance of the TIGER database: experiences in spatial data sharing at the U.S. Bureau of the Census. In Sharing Geographic Information, ed. Harlan J. Onrud and Gerard Rushton. New Brunswick: Center for Urban Policy Research. Steiner, Ruth L. 1994. Residential density and travel patterns: review of the literature. Transportation Research Record 1466: 37-43. Stewart, John Q. 1942. A measure of the influence of a population at a distance. Sociometry 5: 63-71. Sui, Daniel Z. 1999. Postmodern urbanism disrobed: or why postmodern urbanism is a dead end for urban geography. Urban Geography 20: 403-411. Tacken, Mart. 1998. Mobility of the elderly in time and space in the Netherlands: an analysis of the Dutch National Travel Survey. Transportation 25: 379-393. 191 Talen, Emily. 1999. Constructing neighborhoods from the bottom up: the case for resident-generated GIS. Environment and Planning B: 533-554. 26: Talen, Emily. 1997. The social equity of urban service distribution: an exploration of park access in Pueblo, Colorado, and Macon, Georgia. Urban Geography 18: 521-541. Taylor, Michael A.P.; Woolley, Jeremy E.; and Zito, Rocco. 2000. Integration of the global positioning system and geographical information systems for traffic congestion studies. Transportation Research 8: C 257-285. Thill, Jean-Claude. 2000. Geographic information systems for transportation in perspective. Transportation Research 8: C 3-12. Tri-Met. 2001. www.tri-met.org U.S. Census Bureau. 2001. www.census.gov U.S. Department o f Transportation. 1996. Scan o f Recent Travel Surveys. Washington, DC: USDOT. Vance, James E., Jr. 1991. Human mobility and the shaping of cities. In Our Changing Cities, ed. John Fraser Hart. Baltimore: Johns Hopkins University Press. Vance, James E., Jr. 1990. The Continuing City: Urban Morphology in Western Civilization. Baltimore: Johns Hopkins University Press. Vance, James E, Jr. 1966. Housing the worker: the employment linkage as a force in urban structure. Economic Geography 42: 294-325. Vance, James E, Jr. 1964. Geography and Urban Evolution in the San Francisco Bay Area. San Francisco: Institute of Governmental Studies. Villoria, O. G., Jr. 1989. An Operational Measure o f Individual Accessibility for Use in the Study o f Travel-Activity Patterns. Ph.D. Dissertation, Ohio State University. Wachs, Martin, and Kumagai, T. Gordon. 1973. Physical accessibility as a social indicator. Socio-Economic Planning Science 7: 437-456. Wachs, Martin; Taylor, B.D.; Levine, N.; and Ong, P. 1993. The changing commute: a case-study of the jobs-housing relationship over time. Urban Studies 30: 1711- 1729. 192 Waddell, Paul. 1993. Exogenous workplace choice in residential location models: is the assumption valid? Geographical Analysis 25: 65-82. Waddell, Paul; Berry, Brian J.L.; and Hoch, Irving. 1993. Housing price gradients: the intersection o f space and built form. Geographical Analysis 25: 5-19. Waddell, Paul, and Shukla, Vibhooti. 1993. Manufacturing location in a polycentric urban area: a study in the composition and attractiveness of employment subcenters. Urban Geography 14: 277-296. Wang, Fahui. 2000. Modeling commuting patterns in Chicago in a GIS environment: a job-accessibility perspective. Professional Geographer 120-133.52: Waters, N.M. 1999. Transportation GIS: GIS T. In Geographical Information Systems, Volume 1: Principles and Technical Issues, Second Edition, ed. Paul A. Longley, Michael Goodchild, David J. Maguire, and David W. Rhind. New York: John Wiley & Sons. Weber, Joseph, and Kwan, Mei-Po. 2001. Bringing Time Back In: A Study on the Influence of Travel Time Variations and Facility Opening Hours on Individual Accessibility. Manuscript. Wolch, Jennifer R.; Rahimian, Afsaneh; and Koegel, Paul. 1993. Daily and periodic mobility patterns of the urban homeless. Professional Geographer 159-169.AS : Wyly, ElvinK. 1999. Continuity and change in the restless urban landscape. Economic Geography 75: 309-338. Wyly, Elvin K. 1998. Containment and mismatch: gender differences in commuting in metropolitan labor markets. Urban Geography 19: 395-430. 193