INFORMATION TO USERS

This was produced from a copy of a document sent to us for microfilming. While the most advanced technological means to photograph and reproduce this document have been used, the quality is heavily dependent upon the quality of the material submitted.

The following explanation of techniques is provided to help you understand markings or notations which may appear on this reproduction.

1.The sign or "target” for pages apparently lacking from the document photographed is "Missing Page(s)”. If it was possible to obtain the missing page(s) or section, they are spliced into the film along with adjacent pages. This may have necessitated cutting through an image and duplicating adjacent pap * to assure you of complete continuity.

2. When an image on the film is obliterated with a round black mark it is an indication that the film inspector noticed either blurred copy because of movement during exposure, or duplicate copy. Unless we meant to delete copyrighted materials that should not have been filmed, you will find a good image of the page in the adjacent frame. If copyrighted materials were deleted you will find a target note listing the pages in the adjacent frame.

3. When a map, drawing or chart, etc., is part of the material being photo­ graphed the photographer has followed a definite method in "sectioning” the material. It is customary to begin filming at the upper left hand corner of a large sheet and to continue from left to right in equal sections with small overlaps. If necessary, sectioning is continued again—beginning below the first row and continuing on until complete.

4. For any illustrations that cannot be reproduced satisfactorily by xerography, photographic prints can be purchased at additional cost and tipped into your xerographic copy. Requests can be made to our Dissertations Customer Services Department.

5. Some pages in any document may have indistinct print. In all cases we have filmed the best available copy.

University Micrcxilms International 300 N. ZEEB RD.. ANN ARBOR, Ml 48106 8214077

Chamews, Mark Steven

DESTINATION CHOICE SET COMPOSITION

The Ohio State University Ph.D. 1982

University Microfilms International300 N. Zecb Road, Ann Arbor, MI 48106

Copyright 1982 by Charnews, Mark Steven All Rights Reserved DESTINATION CHOICE SET COMPOSITION

DISSERTATION

Presented in Partial Fulfillment of the Requirements for

the Degree Doctor of Philosophy in the Graduate

School of The Ohio State University

by

Mark Charnews, B.A., B.A., M.A.

*****

The Ohio State University

1982

Reading Committee: Approved by

Howard L. Gauthier Edward J. Taaffe S. Earl Brown

Howard L. Gauthier Department of Geography DEDICATED TO

DANIEL AND STEPHANIE CHARNEWS

ii ACKNOWLEDGMENTS

The writing of a dissertation requires a tremendous

amount of work. I would like to take this opportunity

to thank those who helped me complete this research.

First, I would like to thank Dr. Howard L. Gauthier.

His patience, guidance and friendship during the past four

years saw me through the initial idea to the finished form

of this dissertation. I would like to thank Dr. Edward

Taaffe for all the help he gave to me. I would like to

thank Dr. S. Earl Brown. His aid and sense of humor often

helped me maintain a sense of perspective. Also I would

like to thank my fellow graduate students. Such comradery

is rare.

Second, I would like to thank the Geography Department

for financial support during my years at The Ohio State

University.

Finally, I want to thank my parents. Through their

sacrifices and encouragement during my many years in college,

I was able to take advantage of opportunities they never had.

This dissertation is for them.

i i i VITA

January 28, 1953...... Born - Greenport, New York

1975...... B. A. , State University of New York at Buffalo

1975 - 1976 ...... Teaching Assistant, Depart­ ment of Geography, State University of New York at Buffalo

1976 - 1977 ...... Graduate Fellow, Interna­ tional Development and En­ vironmental Planning Group, Buffalo, New York

1977 - 1979 ...... Teaching Assistant, Depart­ ment of Geography, The Ohio State University, Columbus, Ohio

1979 - 1982 ...... Teaching Associate, Depart­ ment of Geography, The Ohio State University, Columbus, Ohio

Publications with Barry Lentnek and John V. Cotter, "Commercial Factors in the Development of Regional Urban Systems: A Mexican Case Study," Economic Geography, 54, 291-308. TABLE OF CONTENTS Page DEDICATION ...... ii

ACKNOWLEDGMENTS...... iii

VITA...... iv

LIST OF TABLES...... vii

LIST OF FIGURES...... viii

CHAPTER

1. INTRODUCTION...... 1

1.1 Background of Research ...... 3 1.2 Purpose of Research...... 5 1.3 Organization of Report ...... 7

2. LITERATURE REVIEW ...... 8

2.1 Traditional Covariance Models...... 12 2.2 Utility Maximization Models...... 19 2.3 Constraint Modeling...... 31 2.4 Demand Oriented Models ...... 33 2.5 Choice Set Models...... 38 2.6 Summary...... 43

3. METHODOLOGY...... 49

3.1 Sampling and Study Area Requirements . . 56 3.2 Survey Section I: Situational Variables...... 58 3.3 Survey Section II: Store Perceptions and Satisfaction Levels ...... 59 3.4 Multidimensional Scaling: Analysis Technique for Store Perception...... 65 3.5 MDS Analysis of Store Perception Individual and Neighborhood Level (from Survey Section II)...... 68 3.6 Survey Section III: Attitudes Towards Stores and Structure of Preference...... 77 3.7 Synopsis of Methodology...... 85

v Page

4. ANALYSIS...... 87

4.1 The Study Area and Sample...... 88 4.2 Journey to Work and Shopping Trip Frequency and Information Sources...... 99 4.3 Choice Set Composition...... 104 4.4 Respondent Attitude Profile ...... 145 4.5 Conclusions...... 163

5. CONCLUSIONS...... 168

BIBLIOGRAPHY...... 177

vi LIST OF TABLES Table Page

3.1 Stress Goodness of F i t ...... 67

4.1 Socio-Economic Characteristics of Neighborhoods ...... 95

4.2 Socio-Economic Characteristics of the Sample . . 97

4.3 Employment Travel Time (one way) for Respondents in Minutes...... 100

4.4 Number of Weekly Shopping Trips by Neighborhood...... 101

4.5 Use of Information Sources...... 102

4.6 Size and Number of Perceptual Groups Within the Individual's Choice Set...... 106

4.7 Stores Included in South Neighborhood...... 112

4.8 Attributes Used in Store Comparisons in South Neighborhood ...... 114

4.9 Stores Included in Central Neighborhood...... 124

4.10 Attributes Used in Store Comparisons in Central Neighborhood ...... 126

4.11 Stores Included in North Neighborhood...... 131

4.12 Attributes Used in Store Comparisons in North Neighborhood ...... 132

4.13 Satisfaction Levels for Selected Stores...... 140

4.14 List of Attributes and Levels...... 147

4.15 Tradeoff Analysis...... 152

4.16 Situational Variables of Perceptually Homogeneous Groups...... 157

vii LIST OF FIGURES Figure Page

2.1 Iso-Utility Curves for Destinations...... 21

2.2 Iso-Utility Curves with New Destinations .... 23

2.3 Modeling Domains...... 36

3.1 Flow Chart of Study...... 51

3.2 Frequency and Importance Scales...... 61

3.3 Respondent Task "Pick k out of n-1"...... 62

3.4 Similarities Matrix...... 63

3.5 Asymmetric Multidimensional Scaling...... 69

3.6 Similarities Matrix Example...... 72

3.7 Store Configuration...... 73

3.8 Weighted Asymmetric Multidimensional Scaling . . 76

3.9 Tradeoff Matrix Example...... 81

4.1 Study Area Within Franklin County...... 90

4.2 Store Locations...... 92

4.3 Location of Respondents...... 94

4.4 Stress Plots...... 117

4.5 Store Configuration South Neighborhood ...... 119

4.6 Store Configuration South Neighborhood ...... 120

4.7 Store Configuration South Neighborhood ...... 121

4.8 Store Configuration Central Neighborhood .... 127

4.9 Store Configuration Central Neighborhood .... 128

4.10 Store Configuration Central Neighborhood .... 129

viii Page 4.11 Store Configuration North Neighborhood...... 134

4.12 Store Configuration North Neighborhood...... 135

4.13 Store Configuration North Neighborhood...... 136

4.14 Attribute Pairing Diagram...... 149

4.15 Plot of Distances...... 150

ix CHAPTER ONE

INTRODUCTION

A fundamental concern in transportation geography is the

problem of spatial choice. In order to anticipate future

demands for transportation facilities as well as to correct

present shortcomings in the system, the process by which the

individual chooses those places he travels to is crucial. The process of spatial choice involves the gathering of informa­

tion on a number of possible destinations and then, through some type of evaluation process, the selection of one of these alternatives as the destination of a trip. The process of spatial choice over an interval of time for residents of an area results in the spatial demand for facilities by these people. Those people whose function it is to provide these facilities, be the facilities grocery stores, department stores, shopping malls, restaurants, parks or whatever, try to locate these facilities in response to spatial demands in order to maximize profits or best serve the public. The study of spatial choice for a given trip purpose provides informa­ tion on one of the processes which gives rise to the spatial structure of the facilities related to the specific trip pur­ pose. This is to say, the spatial choice is one process that gives rise to spatial organization. To understand the rela­ tionship between the process of spatial choice and spatial 1 2 structure results in an increased ability to provide faci­ lities to those who currently have little access to them, as well as to plan ahead for future demands upon the system.

The process by which the individual or groups of simi­ lar individuals choose from among a number of alternative destinations can be conceptualized as the following problem.

Given: 1) a set of m individuals at known locations, who have identical decision-making processes and spatial pre­ ferences I = an(^ ^ a set n *cnown spatial alternatives or destinations(D = d^,....d } comprising a spatial choice set of facilities for the conduct of a par­ ticular activity, (a), what is the spatial choice probability,

P. (d./cL ,....d ...... d ) of the m decision-makers choosing t y 1 2 n' 3 alternative dj for the conduct of activity a in some time period t.

The underlying assumptions of the problem are: 1) at least two or more alternatives in any choice set, 2) no systematic variations between individuals or the size of the choice set and 3) each individual chooses at most one alter­ native and chooses from the same choice set for a given pur­ pose (Burnett, 1973; Smith, 1975).

The spatial choice probability is expressed as a conditional choice decision for a given trip purpose at time t .

Pfc(dj/d1, ,dn) = Pfc(dj/aet) • P(aet)

The equation states that the unconditional probability of choosing an alternative dj is equal to the product of the conditional probability of dj being selected given activity a is selected and the probability of activity a being selected (Burnett, 1973).

1.1 Background of Research

The early attempts at modeling destination choice obscured the role of the individual by distributing trips from all origin zones to all possible destination zones by the use of zonal aggregate models in order to replicate actual trip patterns. Besides ignoring the role of individual beha­ vior in spatial choice, these descriptive "trip distribution" models did not include any policy related variables that planners could manipulate. Accordingly, an emphasis to modeling destination choice at the individual or micro level to overcome these shortcomings was suggested.

The early micro models usually assumed individuals use a utility maximization (optimal choice) strategy in choosing among alternatives. However, this approach received much criticism by later studies. Space-time budget studies

(Hagerstrand, 1969) suggest individuals' choice sets are far too small for optimal choice to occur. Such constraint approach studies of destination choice behavior demonstrate the existence of small choice sets and show institutional factors, corporate action, life cycle, etc., act to constrain the individual (Gray, 1975). A constraint approach to 4 to destination choice holds little hope for the establish­ ment of models which can function under the highly complex choice environment of space-time budget studies (Anderson,

1971). A demand oriented approach to destination choice which examines perceived supply rather than actual supply confirms the conclusions of the constraint approach. Thus, the conclusions of constraint and demand oriented studies suggest the inclusion of limited information and the stresses of time and other commitments in influencing spatial choice behavior.

Choice set modeling is concerned with establishing what portion of the actual set of destinations relevant to a given trip purpose are actually perceived by the individual. This approach has in a way redefined the basic destination choice problem in that the spatial choice probability is redefined as:

Pt = P(djeD) • P(dj/dj eD)

(adapted from Burnett and Hanson, 1979)

This equation states that the unconditional spatial choice probability is a product of the probability of an alternative being in the choice set and the conditional probability of choosing that alternative given that it is in the choice set.

This reformulation has the advantage of separating insti­ tutional constraints from personal preferences. Thus, if the former probability density function for inclusion of various destinations are highly peaked, that is, only a few stores are included in the individual's choice set, while the

latter choice strategy is flat, meaning choice from within

the choice set is even, consumer choice behavior would then

be most affected by those changes which influence the inclu­

sion or exclusion of stores from the individual's choice set.

It is this reformulation that provides the basis for this

study. The focus of this study is upon the composition of the

individual choice set. Composition refers to the number of

stores and the perceptual groupings of these stores as being

similar or dissimilar as viewed by the individual. By

examining the composition of the individual's choice set,

information as to the appropriate choice strategy may be

found by linking the degree of complexity of the choice set to

non-optimal choice theory. Work in this area has just begun

(Burnett, 1980; Heggie and Jones, 1978; Jones, 1979). This

study focuses upon this problem for grocery shopping, a

repetitive trip purpose often used in spatial choice studies.

1.2 Purpose of Research

The purpose of this research can be stated in four major objectives:

1. To establish the composition of the destination choice set of the individual for the purpose of grocery shopping. This entails the. determina­ tion of the number and perceptual groupings of grocery stores the individual patronizes.

2. To determine a set of grocery store attributes by which individuals discriminate between gro­ cery stores and to determine the manner by which these attributes are utilized by each individual 6

in discriminating between stores within their own choice set.

3. To identify groups of individuals who have similar attitudes with regards to the attri­ butes of grocery stores.

4. To establish the relationship between the situational variables of similar perceptual groups and the composition of their respec­ tive destination choice sets.

The purpose of these objectives is to expand our understand­

ing of the process by which destination choice occurs.

Specifically, this research attempts to define the complexity of individual choice sets in regards to its composition in order to identify appropriate choice rules by which the indi­ vidual chooses among alternatives.

The link between the complexity of the composition of

individual choice sets and the method of choice of a destina­ tion within the choice set would be extremely useful to urban transportation planners. If destination choice sets are small in areas of high mobility and income, factors other than insti- tional constraints such as government and corporate interests are at work. Under conditions of many available destinations, many of which are considered to be equally suitable, the existence of small choice sets would indicate that the indivi­ dual has simplified his choice problem by restricting the number of stores he actively considers. This scenario of choice would best be modeled with a non-compensatory, non- optimal choice theory. Such a theory differs dramatically from compensatory utility maximization choice models which have been suggested as appropriate in the literature. The

validity of using non-optimal choice theory or utility

maximization theory is currently a source of controversy

in spatial choice literature. The examination of the

composition of the individual's choice set may contribute

to the resolution of this controversy.

1.3 Organization of Report

This report consists of five chapters. Chapters One

and Two provide a definition of the problem of destination

choice and a select review of the literature which delineates

the trends of transportation research in the area of spatial

or destination choice which have influenced the approach

taken in this study. Chapter Three details the methodological

sequence which was developed for the purposes of examining

the objectives of this research. Chapter Four is a report

recounting the results of a personal survey conducted in a

small area of Northeast Columbus, Ohio from June to August of 1980. Chapter Five provides a summation of the conclu­

sions drawn from this research project and implications for policy. CHAPTER TWO

LITERATURE REVIEW

A large and diverse body of literature exists which

deals with the problem of destination selection or what

is currently referred to as spatial choice. Destination

choice modeling has attracted researchers from many disci­

plines. The earliest attempts were undertaken within the

area of urban transportation planning. Under the rubric of

trip distribution, mathematical models were developed to allo­

cate trips from zone to zone in such a manner as to duplicate

existing patterns. These models, while having received much

criticism at both the theoretical and empirical levels, are

among the mainstays of the urban transporation planning (UTP)

process today. Many new approaches to destination choice

modeling have been developed to improve upon these traditional

trip distribution models. Inspired by the rational choice

concept from economics, a spatial preference approach intro­

duced to destination choice the concepts of utility maximiza­

tion as the basis for choosing among alternatives and a method of evaluating an alternative by considering various

attributes of that alternative. Space-time budget studies

describe the role of institutional factors in constraining the

number of alternative destinations available to the individual. 8 Other studies have concentrated upon the perceived supply of

alternatives that the individual considers. Using the re­

sults of such perception studies and armed with the recent

advances in choice theory and perceptual scaling techniques

from the psychology literature, most recent efforts have been

in the development of spatial choice set theory. This study

is a part of that effort.

The purpose of this chapter is to review past approaches

to destination choice to show the evolution of modeling approaches to destination choice from the trip distribution

approach to current efforts in the development of spatial choice set theory.

Current research trends in destination choice empha­ size the role of the individual as a decision-maker (Stopher and Lisco, 1970; Ben-Akiva, 1973; Rassam et al., 1970). This emphasis is a fundamental shift away from the traditional macro, or zonal, forecasting models of the past which many transportation planners still use. An emphasis on the role of the individual as a decision-maker, or the micro approach, is the basic approach to destination choice taken in this study.

Many rationales for a shift from macro to micro models exist (Marble, 1967; Nystuen, 1967). One such rationale which states the basic advantages of the micro approach concisely is provided by M. Ben-Akiva. In essence, he argues that the specification of any type of model involves the postulation of relationships among variables by making cer­

tain assumptions about travel behavior. Accurate predic­

tion is contingent upon these behavioral assumptions. While

a model may be able to duplicate trends in the data with

accuracy, if its behavioral assumptions are erroneous, the

usefulness of the model for prediction may be limited. Con­

versely, if the behavioral assumptions are sound, then the

model should perform well in radically different conditions

which may result from a change in transportation planning policy. It is not sufficient that a model contain policy

relevant variables for planning purposes, but it must also

be valid in its behavioral assumptions by reflecting the actual decision-making process of the individual (Ben-Akiva,

1974). The result of this view is that destination choice modeling must ultimately be concerned with the development of models relevant to urban transportation planning purposes.

Such planning relevance requires that the objective of any travel demand model is to be able to predict travel behavior at some aggregate level (Ansah, 1977). Since travel deci­ sions are made at the individual level, rather than at the aggregate level, the micro approach holds the most promise for the inclusion of more valid behavioral assumptions in travel demand models.

Concentration upon the decision-making process of the individual is the basic tenet of the behavioral approach to destination choice. Much of the current destination choice 11 research easily falls within a three stage conceptual frame­ work which outlines the way this process is thought to occur.

Agreement upon this framework is evident by the frequent citing of articles by P. Burnett in destination choice stu­ dies (see Sheppard, 1980). This conceptual framework is:

Stage I. Activity Classification and Selection

A. The identification and timing of activities which require travel for the activity to occur.

B. Activity Types 1. Obligatory - necessary activities with requirements in time, frequency or location 2. Discretionary - activities that are option­ al or flexible in time and/or location

C. Activity Selection A choice point is reached upon selection of an activity

Stage II. Choice Set Formulation

A. An activity has been selected

B. A destination choice set is formed where all destinations known to the individual which are relevant to the activity and fit the indivi­ dual's space-time constraints are assembled.

Stage III. The Choice Rule

A destination is chosen by some selection criteria or rule and the trip is undertaken at the decided time. (adapted from Burnett, 1978)

Stages II and III constitute the spatial choice problem. 12

2.1 Traditional Covariance Models

The purpose of these models is to forecast the distri­

bution of trips originating from a set of zones to all

possible destination zones for a large urban area. Three

major classes of models have been developed to handle the

trip distribution problem in practice: 1) growth-factor;

2) gravity; and 3) intervening opportunity models.

Growth-factor models are the simplest type of the trip

distribution models. The Uniform Factor, Fratar and the

Detroit model are examples of this type of model which differ

only in the manner in which the growth factors are calculated.

These models distribute trips between origin and destination

zones based upon.the simple expansion of interzonal trips by

increasing the number of trips originating or demanded at

each zone. These increases are achieved by a growth factor

which reflects the planner's expectations of future growth

in each zone. This type of model has no provision for mea­

suring the effects of changes in the transportation system

in an area since there is no term related to any form of

travel cost, time or other travel impedance. This short­

coming severely limits the usefulness of the model. The

effects of changes in the transportation network, or people's

evaluation of travel time, cost, etc. cannot be modeled.

These models do have the advantage of requiring data on only

trip interchanges and total trip ends for each zone. So while these models have the advantage of minimal data 13

requirements and ease of computation, growth factor models

can only replicate existing patterns. Zonal growth factors

are estimated by means external to the model and the accuracy

of these estimates cannot be measured.

Gravity models represent an improvement over growth-

factor models in that a travel cost term, usually a measure

of interzonal distances is included. Pallin (see Astrom,

1973) applied an analogy of the gravity model to the trip

distribution problem in the form:

Tij = JLIili

where: = populations of zones i and j

= distance between zones i and j

K, n = constants (1 < n < 2 usually)

Tij = one way movements from i to j

A more sophisticated form of the gravity model developed specifically for trip distribution problems is of the following form:

where:

CK = total trips produced by zone i

Aj = total trips attracted by zone j

Rj. = a measure of the spatial separation of zones i and j

BfC. = constants associated with the production and attraction zones respectively 14

Through the term, changes in the transportation network and peoples' evaluation of travel cost, time, etc. can be modeled. Network changes can be included by an appropriate change in the inter-zonal distance values.

Evaluations of travel time, costs, etc. can be included in the model by either manipulating the distance exponent (n) or by changing (n) to a more complex function of distance or by expanding the factor into a more complex measure of distance by including travel time and costs as some sort of composite index. This improvement allows for planners to estimate the effects of changes in the transportation system or projected land use changes because an explicit expression of peoples' sensitivity to such changes is included in the model.

As the gravity model becomes more complex, so do its associated problems. In order to be of use to planners, the model must be calibrated so that all trips from origin zones are assigned to destination zones so that no destination zone is left either over or underfilled. It can be shown mathe­ matically that this is possible only under highly restricted conditions (Stopher and Meyburg, 1975, pp. 140-141). Another weakness arises from the model's tendency to consistently overestimate interzonal trips in densely populated areas and underestimate trips in sparsely populated zones. This sug­ gests that trip generation is not directly related to popula­ tion as implied in the formulation of the model. The behavioral assumptions by which the gravity model is formu­

lated therefore may be in error. A third problem area for

the gravity model arises from the near impossibility of

obtaining a single value for n. It appears that the value

should change for varying trip lengths and possibly other

trip characteristics. The gravity model offers no method

for the identification of the relevant trip characteristics

that may affect the n value.

A major conceptual oversimplification inherent in the

gravity model is that there is no provision for the influ­

ence of any third zone upon trips made between any given

pair of zones. The intervening-opportunities model

addresses this issue and represents the most sophisticated

form of macro-oriented models. Stemming from work in migra­

tion and intercity travel research, it was first proposed by

S. A. Stouffer (1940) for population migration and its use

in trip distribution is attributed to Morton Schneider as

part of the seminal Chicago Area Transportation Study

(CATS 1960).

The mathematical formulation of the intervening oppor­

tunities model is somewhat complex.

An example of the intervening-opportunities model is given in the following formula: 16

The underlying logic of the model is simpler than the formula suggests. The model assumes total travel time is minimized subject to the condition that there is a given probability of a destination being chosen if it is considered.

It also assumes that, given it is considered, the probability of a destination being chosen is constant and independent from the order in which the destinations are considered.

Thus the overall model assigns trips between zone pairs by the conditional probability of choosing a given destination zone, which is equal to the probability that the zone con­ tains an acceptable destination multiplied by the probability of a closer destination has not been chosen.

This model is intuitively more pleasing in that it models destination choice as a search process and at least provides an attempt of modeling individual behavior as a stochastic process. It does suffer from several shortcom­ ings. The model is only sensitive to the ordering of zones by distance or time, an ordinal level, and not to the quantity of distance or time, ratio scale level. Therefore trips involving natural barriers or restrictions are often over­ estimated. Problems of calibration exist in that interzonal trip movements replication is not involved in the calibration process. This means that zonal supply-demand requirements for trips can be met while specific estimated interzonal pair trip flows do not correspond with actual patterns. In this way the investigation of travel movement by small groups of 17

individuals is impossible with the use of this model

(Stopher and Meyburg, 1975).

Several criticisms of the macro-covariance type of destination choice models have been documented in the litera­ ture. First, the models are non-behavioral in nature. They only replicate the travel patterns of travelers at the time of the data collection. The influence of changes in traveler circumstances or the effects of an introduction of new travel alternatives can not be determined by these models (Brand,

1972). The serious inability of these models to predict changes is at least partially due to the calibration process of the gravity and intervening opportunities models. These models are calibrated using a fixed distribution of trip lengths which is usually measured in minutes. As pointed out in an article by Brand (1972), "Trip distribution is modeled as a function of a simple distribution of the trip lengths which prevailed at the equilibrium between supply and demand in the data base file." Thus the models are only descriptive and will not forecast a change in trip distribution with any proposed modification to the transportation system. The formulation of both the gravity and intervening opportunities models prohibits changing the total number of trips entering or leaving a zone (referred to as trip end density) with a corresponding change in the transportation system. Only trip allocations may be changed, so the models formulation makes it insensitive to traveler circumstances. In regards to the 18

problem of examining the effects of new transport alterna­

tives, travel impedance measures are usually measured for

auto only. Changes in the transit system will not effect

trip length or number of trips entering or leaving a zone.

Consequently the models are descriptive rather than causal

in that calibration takes place under old equilibrium con­

ditions of trip length and trip end density and not under

projected new equilibrium conditions with a change in the

transportation system.

The resultant insensitivity to changes in the transpor­

tation system or changes in traveler circumstances make these models inappropriate to policy oriented variables which in­

fluence these features. Therefore the performance of these models for planning purposes is unsatisfactory.

Another criticism arises from the fact that time of day

is seldom considered. It is common knowledge that the bulk of urban trips are made during morning and afternoon rush hours. Trip behavior must be affected by these conditions but have seldom been incorporated into destination choice.

Finally, these models are based upon zonal aggregate units of observation. This implies that it is the zones and not individuals that make travel decisions. The use of zones is inappropriate for the study of travel behavior since it has been shown that a large percentage of variation in socio-economic and behavioral characteristics exist within zones. This information is lost as the data is transformed 19 2 into zonal means. This results in misleadingly high R and

F values in aggregate models. Also, zone sampling distri­

butions are skewed and not normally distributed. This indi­

cates that zonal means are biased measures of zonal character­

istics (Stopher and Meyburg, 1975).

In summation, macro type models tend only to show co-

variance among variables at an aggregate or zonal level.

These models do not contribute much to the understanding of

destination selection behavior and their ability to forecast

travel demand is suspect (Stopher and Meyburg, 1975). For a model to forecast correctly, the model should be causal;

there must be a link between attributes of a transportation

system and the decisions of the individual (Domencich and

McFadden, 1975, p. 3).

2.2 Utility Maximization Models

In order to more fully understand travel behavior at an individual level, a utility maximization approach, from the economics concept of rational choice, provided a more sound conceptual basis for trip distribution problems (Beckmann and Golob, 1974).

One of the most interesting and well known of this type of destination choice model is the spatial preference model developed by Rushton (Rushton, 1969, 1971). This model borrows from revealed preference theory, which is part of the consumer economic literature (Samuelson, 1948). The purpose 20 of the approach is to try to incorporate actual consumer behavior in the form of individual preferences into desti­ nation choice modeling. These preferences are modeled as tradeoffs between destination attributes such as size of the facility and distance. By obtaining a series of con- sistantly paired comparisons of destinations, a unique preference ranking of the destinations can be revealed. By viewing the destinations as a combination of attribute levels,

(e.g., a grocery store can be viewed as being a certain size and distance from an individual's house), a tradeoff between size and distance by the individual is revealed in his pre­ ference ranking of the destinations he considered. Careful examination of the individual's rank ordered comparisons leads to the derivation of iso-utility or indifference curves for various levels of the attributes being studied, (see

Figure 2.1). The choice model used in revealed preference assumes that the consumer chooses the store which is thought of as a set of attribute levels that maximizes his utility, utility being comprised of those features of a destination which the individual finds useful to fulfill his demands.

The utility function is generalized over a set of similar groups of individuals. For this reason it is not entirely true to the micro approach. It is assumed that people's pre­ ferences for size versus distance, etc., may be similar for similar socio-economic groups while the destination character­ istics will vary as different bundles of attribute levels Distance

Note: utility level U1 is greater than U2 which is greater than U3 Figure 2.1

Iso-utility Curves for Destinations 22

from place to place. In theory, the model should be trans­

ferable since the indifference curves would remain stable,

and the individual would then choose the new destination

which has an attribute level bundle which maximizes his

utility. If, for example, the individual moved to a new

area, it is hypothesized that his iso-utility curves would

remain the same. The stores available to him would exist as

new points in two or more dimensional space and the location

that would be chosen would be that which is closest to the

highest level indifference curve (see Figure 2.2).

The view of destinations as a set of attribute levels

is an important feature of this theory. It allows for the

theoretical forecasting of new destination choice probabili­

ties by either the addition or deletion of points in the

individual's attribute space as new stores are built or old*

ones are closed. The view of a destination as a set of at­

tribute levels and the view of utility as arising from a com­

bination of several attribute levels are taken from a seminal

article in consumer economics by Lancaster (1966).

Traditional economics views consumer choice as the acqui­

sition of a "basket of goods." This basket of goods consists of those items which maximize the individual's utility sub­

ject to his budget. While this approach has served well in economic research, it suffers from one major drawback. The approach cannot predict in any way the effects of the intro­ duction of a new good or service upon consumer demand. This Size

Distance Distance Dote: X1 would be chosen since it maximizes utility Figure 2.2

Iso-utility Curves with New Destinations

N) LJ 24

shortcoming is the crux of Lancaster's contribution. Tradi­

tional economics views goods as the direct objects of util­

ity. This means that utility is derived from the consumption of the good and that the good forms the basic unit of con­

sumption. That is,the quantity of each good demanded is adjusted to maximize individual utility. Lancaster views consumption as an activity where goods, either singularly or

in combination are selected for the properties or character­

istics which these goods possess. It is these properties on which the individual bases his preferences. This view can be summarized in three basic tenets:

1. The good, per se, does not give utility to the consumer; it possesses characteristics, and these characteristics give rise to utility.

2. in general, a good will possess more than one characteristic, and many characteristics will be shared by more than one good.

3. Goods in combination may possess characteristics different from those pertaining to the goods separately. (Lancaster, 1966)

This approach allows for the introduction of a new good, or as in the spatial preference approach, a new destination, by the introduction of a new "set of attribute levels" e.g., a new store as a certain size and distance. This new point in the individual's attribute space is then considered as to whether it is capable of increasing the individual's utility.

Thus the individual selects goods on the basis of their shared characteristics rather than by the goods themselves as separate entities. While the spatial preference approach appeared to hold

great promise for spatial choice problems, several objections

to the approach soon cropped up in the literature (Pirie,

1976; Anderson, 1971; Golledge, 1970; Ewing, 1971; Sen, 1973;

Kraik, 1971). The spatial preference model, as used in

destination choice, reduced to its simplest terms, resembles

the gravity model in form. In the work of William Black

(1971) gravity type models are shown to be affected by spa­

tial structure. He found that distance exponents are af­

fected by the distance various alternatives are from the decision-maker. Thus, if it is found that people in rural areas travel further to shop at stores of a given size than their urban counterparts, this may simply be due to the stores in rural areas being located further apart than a true difference in utility functions. Therefore the transferabil­ ity of a spatial preference model to an area of different spatial structure may not produce reliable results.

Another criticism of this model involves the assumption that many alternatives exist and are being actively consi­ dered by the individual. Many alternatives must be compared by many individuals to generate the data required to accurately derive the iso-utility or indifference curves be­ tween the levels of attributes. This means that all the alternatives must exist in the choice sets for all indivi­ duals. Although many alternatives may exist in space for the individual, if there is not access to these alternatives 26

by the individual, he has no opportunity to choose among

these alternatives in a way that reveals his true preferences

(Eyles, 1971). Further, research shows that individuals have

the ability to discriminate between a limited number of

alternatives at one time (Ewing, 1971). The constraints of

time-space schedules, physiological capacity and other socio­

economic situational variables further limit the number of

alternatives to which the individual has access (Hagerstrand,

1969). For example, a study conducted in two Oxfordshire

towns (Dix, 1977) identified most housewives in one car

households conducted major food shopping trips as a binary

choice between a daytime trip to the center of town by bus

or evening or weekend trip by the family car. Also, the

effects of the actions of others and the fact that the indi­

vidual's actions are not made under a single purpose may

cause additional deviations of the individual's actions from

his actual preferences (Sen, 1973; Burnett, 1974). In

studying the effects of the introduction of flexible working hours, Shapcott and Steadman (1977) discovered most people made little adjustment to daily routines. Habit and inertia, as well as other environmental or household constraints, were suggested as factors inhibiting individuals from chang­

ing their course of action. The spatial preference approach of determining group preferences by examining the frequency of choosing an alternative is erroneous unless the alter­ natives are available to all (Ewing, 1971). In sum, it can 27

be concluded that examination of the number of alternatives

actively considered by the individual is of great signifi­

cance in modeling destination choice.

The spatial preference approach, which assumes many

alternatives (destinations) and a common utility function for

groups of individuals, makes it similar in structure to the

multinomial or probit models. These models are most fre­

quently applied to mode choice problems in transportation but

may be used in destination choice applications (Ben-Akiva,

1974; Burns and Golob, 1976). The basic premise of the model

is that the probability of choosing a certain alternative,

given certain conditions, e.g. time of day, etc., is equal

to the ratio of the utility derived from that alternative

to the sum of the utilities derived from all of the alter­ natives, utility being a measure of the usefulness of an alternative which is derived from the levels of various attributes that the alternative possesses. Mathematically, the model can be stated as the following:

P.x T J eV(x3,s)

where: = the probability of choosing alternative i

X1 = a vector of attribute levels associated with alternative i

s = a set of socio-economic characteristics of the individual 28

V = a utility function relating attribute levels to overall utility

(Domencich and McFadden, 1975)

The workings of the model can be illustrated by the following

example. If there exist three alternatives A, B, or C

each with an associated utility of 2, 1, 1, the probability

of choosing A is:

2 P a .5 2 + 1 + 1

In a similar fashion one can then calculate that the proba­

bility of choosing alternative B or C is one-quarter each.

Since the model assumes only one alternative can be chosen at a time, the P^ terms must sum to one for all alternatives possible.

Embedded with the model is the independence of irrele­ vant alternatives axiom (Luce, 1959). This axiom states that the relative odds of choosing one alternative over another

(Pj/P^) is independent of the existence of any third alter­ native. This feature is both the major strength and weak­ ness of the model. The strength of this feature is that new alternatives (viewed as a vector of attribute levels) can be introduced without re-estimation of the model when the

V(x,s) terms are calibrated. All that is needed is the addition of a new term to the denominator for each alterna­ tive and the expansion of the probability function to include 29

the new alternative. This feature is important in modeling

choice over time when certain destinations would not be

available at different times of the day.

However, the independence of irrelevant alternatives

axiom may prove too strong in some cases. Particularly,

problems arise when alternatives are not clearly distin­

guishable from one another (Stopher and Meyburg, 1974, p.

278). Due to the highly competitive nature of grocery

stores, store chains copy aspects of their successful counterparts to the point where one grocery store may be perceived as being very similar to another. As an example consider an individual who has an even preference for two stores, A and B, located in his town. These stores are members of different chains. According to the MNL model, the probability of choosing store A or B is one-half. If store chain B opened up another store in the town, say B', this could not affect the even probability of choosing between stores A and B. Assuming that stores B and B' are identical (and thus of equal preference) as is often the case with chain stores, the probability of choosing between stores

A, B, or B' would be one-third each. The result is that store chain B with its two stores now receives two-thirds of this person's business just by opening up another store.

But is this realistic?

It might be more logical to expect a one-half probabili­ ty of visiting store A with stores B and B1 sharing the other 30

one-half. The indifference for the individual may exist

between the two store chains and not between stores of the

same chain. In other words, the individual may not perceive

the second chain store, B', as a true third alternative. He

may perceive his choice as being between store A or one of

the chain B stores thus rendering one-half, one-quarter

choice probability set for stores A, B, B' respectively.

Compounding the problem even further, it is assumed in

the calibration process that the values of attribute levels

are linearly independent, i.e. uncorrelated (Domencich and

McFadden, 1975). In the case of grocery stores, successful

operations are often copied by competitors and stores within

a chain are operated similarly. Therefore correlations be­

tween stores are not only possible but expected. Again, the

reality of the situation violates the model's basic assump­

tions.

The objections to the assumptions of large choice sets,

definitions of alternatives, etc., go beyond the theoretical

unsuitability of the MNL and the spatial preference approaches

to destination choice but extend to a basic dissatisfaction

with the entire utility maximization approach (Sen, 1973;

Kraik, 1971; Mishan, 1961). In the case of grocery shopping,

in order to be truly utility maximizing, one must be know-

ledgable of the prices for foods in all reasonably accessible

stores, and of the cost in time and money to travel to these various stores to take advantage of any and all sales and 31

specials that these stores may offer, without making any

errors. Even if this is possible for the individual, the

extra savings realized by such perfect knowledge would likely

be negligible compared to shopping at a few nearby stores and

being relatively aware of a few major specials. Studies of

individual's choice sets find that the choice set is indeed

small, much smaller than that expected for true utility maxi­

mization. Recalling the premise that the purpose of destina­

tion choice modeling at the micro level is to model the

actual decision-making process of the individual, then

utility maximization or optimal choice theory frequency can

be expected to be at variance with actual behavior.

2.3 Constraint Modeling

Factors other than those related to individual preference are also considered in modeling destination choice. J.

Anderson states there is a danger that "observed behavior may be interpreted misleadingly as what people 'choose' to do rather than 'are forced to do'." (Anderson, 1971, p.

363). Institutional factors, government, corporate actions, life cycle and various interest groups present constraints upon the actions of individuals according to space-time budget geography literature.

Space-time budget studies do not represent a unified body of theory. The purposes and methodology varies from researcher to researcher. One such purpose of space-time 32

studies is to assess overall quality of life and urban system

efficiency (Meier, 1962). A second view of the purpose of

space-time studies is that the observation represents the

overt action resulting from choices reflecting people's

attitudes (Chapin, 1968). The work of Hagerstrand(1969)

focuses upon the influence of various aspects of the urban

environment in constraining peoples' action spaces.

The essential feature of such studies is the use of a

time budget or space-time budget which provides a catalogue

of where a person is at any time for a given period of time.

Thus a description of the sequence, timing, duration and

spatial coordinates of peoples' activities for a period of a day to a week is obtained (Anderson, 1971). Excellent reviews of space-time budget studies are available (see

Jones, 1977; Pred, 1977; and Thrift, 1977).

Empirical studies of constraint factors utilizing very simplifying assumptions, show space-time constraints pose considerable restrictions upon the alternatives available to the individual during the day (Hensher, 1975, 1976; Dix,

1977). Other constraint factors identified by studies using space-time budgets concern various aspects of the life cycle.

These constraints show up on the observation of very small choice sets, those destinations actively considered, for individual households (Westelius, 1973; Hensher, 1975;

Lenntorp, 1976; Ellegard, Hagerstrand and Lenntorp, 1977;

Fried et al., 1977; Heggie, 1977; Jones, 1977; Tivers, 1977). 33

The existence of small choice sets is corroborated by the

detailed studies in chronosophy (space-time structuring)

of the city (Parkes and Thrift, 1975; Thrift, 1977).

Space-time studies show conclusively that individuals

operate in space under the influence of many constraints.

One can expect that true preferences may not be evident in

the actions of consumers due to these constraints. This,

again, is made evident by the small choice set of destina­

tions that an individual maintains.

Since individuals operate under highly constrained

environments it would be of interest to examine the number

and types of stores that an individual visits under these

constraints. The examination of the composition of the

choice set of individuals under conditions of many possible

alternatives could shed light upon the degree to which this

person takes advantage of these spatial opportunities and the

degree to which he is constrained as compared to others in

the same surroundings.

2.4 Demand Oriented Models

In the formation of choice sets, it has been suggested

that it is not the actual supply of alternatives that is

significant but the perceived supply (Marble and Bowlby,

1968; Borg, Heuwinkel and Newman, 1977). This view is evi­ dent in several approaches in the study of spatial choice

such as the investigation of individual's mental maps 34

(Graham, 1977; MacKay et al., 1975) cognizied environments,

(Aldshogius, 1977; Golledge, 1977) and household action spaces (Horton and Reynolds, 1971; Higgs, 1975). While the effects of perceived supply are discussed in the litera­ ture (Ellegard, Hagerstrand and Lenntorp, 1977; Heggie and

Jones, 1978) detailed studies on the effects of supply of urban facilities in space and time on choice set formulation are generally lacking.

The effect of constraints upon demand by some elements of supply have been investigated. Some of the elements studied include money or time budgets (Anderson, 1971; Cullen

1971; Von Rosenbladt, 1972; Bahavi, 1974, 1976; Bain, 1975;

Robinson and Vickerman, 1976; Adler and Ben-Akiva, 1977).

Demand factors influencing the size of the choice set have also been shown to be significant. Complex measures of life cycle and life style are cited (Chapin, 1968; Ellegard et al, 1977; Dix, 1977; Jones, 1977; Heggie, 1977; Kutter, 1976;

Tivers, 1977).

Robinson and Vickerman (1976) focus only on shopping travel and use regression to model the number of trips made by the household as a function of various socio-economic variables and attractive forces that are patterned after an interaction type (gravity) model. Actual destinations are not considered. The interesting finding of the study is that time is not a crucial factor in explaining the number of trips made by the household. Restricting the study to a 35

single trip purpose seems to be extremely valuable in

understanding an individual's actions.

Most demand oriented studies as well as most space-time

studies, while useful in the examination of constraints upon

the individual, represent very detailed studies of travel

patterns and information levels resulting from what one would

expect to be many conflicting desires of the individual.

Derivation of useful summary measures of space-time behavior

remains a major problem (Anderson, 1971). Choice modeling

theory, as it exists at present, cannot explain the process

by which the individual juggles many often conflicting de­

sires against all time and money constraints which may be

present to produce the observed travel pattern (see Chombart

de Lauwe, 1963). The selection of a single trip purpose, as

in the Robinson and Vickerman article restricts the complex­

ity of the problem to a decision environment that is more manageable.

An article by Heggie and Jones (1978) well illustrates

the point. Four "domains" or model environments are des­ cribed in which spatial choice can occur (see Figure 2.3).

The four domains are: 1) independent; 2) spatio-temporal linkages; 3) interpersonal linkages; and 4) full inter­ dependent, a combination of domains 2 and 3. Most travel demand models are formulated to operate within an independent domain when making a decision on a destination for some activity. The consideration of other commitments in space and/or time and commitments to other individuals by the i i : 3 s p s : 3 e ::t h i i :"t e r -p s r s c :;a i LINEAGES Ilia Intra- Illb Inter- household household

II SPATIO-TEMPORAL IV FULL IKTETJJSPEirDEKCE • LIKEAG2S IVa Intra- IVb Inter­ household household

Figure 2.3 Modeling Domains 37 decision-maker are not significant influences. The assump­ tion of an "independent" domain in which a travel decision is being made may be quite reasonable in modeling certain instances. However if the actual domain in which the indivi­ dual makes a decision includes spatio-temporal constraints and interaction with other people then the domain of the problem lies outside the domain of the travel model. Thus the model forecasting results are unreliable because a major influence upon the individual are not included.

The "domain" of choice is not a trivial matter. Many conventional travel demand models perform poorly in empiri­ cal testing due to the models being used outside their appro­ priate domains (Murchland, 1977). In the case of complex spatio-temporal and interpersonal commitments, individuals can be expected to be less responsive to policy change or respond in unexpected ways. An example of so-called indirect effects of a policy change (in school hours) of mode choice for the school journey for children illustrates the effects of inter-personal and spatio-temporal linkages as individuals adjusted to the change (Heggie and Jones, 1978).

The conclusion drawn from the concept of modeling domain is that spatial choice decisions are often made in conjunc­ tion with time, location of the individual and the actions of other people. Care must be taken to select the models which incorporate the same domain as the problem being modeled. Given the current status of choice behavior 38

modeling, selection of a problem in the independent domain

is highly desirable. A portion of current research now

focuses upon recurrent single purpose travel in order to

achieve as independent a domain as possible. Grocery shop­

ping is a convenient and common trip purpose which may meet

these requirements.

2.5 Choice Set Models

The process by which a choice set develops through time

is the subject of a seminal work by P. Burnett (1973). A

single trip purpose of clothing shopping is selected for

investigation. A sample population is selected which is

homogeneous in character except for the individual's length of residence. It is found that those individuals with the

same information levels use basically the same attributes

to evaluate different places. Those living in the area a short time use general attributes, such as distance from

their house or from the center of the city, which are trans­ ferable from area to area. Those with longer residence time use more complex and area specific attributes. The emphasis of the investigation is on the development of individual's choice sets through time as related to learning behavior.

The choice set defined for the study is held constant for the entire sample population' even though some individuals have little or no knowledge of a particular place. This is not a flaw in the research design. There are only about thirty clothing stores in the area. Since the focus of the 39

study is upon learning behavior, the respondents were not

asked to give information on stores they did not patronize

or knew little about. However, the reasons for inclusion

or exclusion of these stores in each individual's choice set

are not examined. The most important finding in the paper

is that while the process of choice set formulation may

progress through one of many possible paths, the choice set

of the individual does stabilize with a firm set of clothing

stores. Therefore, one may conclude that it is the final,

near constant choice set of the individual that is of inter­

est since it is this choice set which represents the indivi­

dual's perceived supply.

Dash et al. (1976) investigates store search behavior

for the purchase of stereo equipment. While this type of

shopping does not usually involve recurrent behavior, the

study is relevant to choice set studies since the focus of

the study is on the number of different stores visited

before actual product selection is made.

Those who visited a greater number of stores before

selection must do so for some reason since time and money is

involved in such a more extensive search. Those who purchase

at a nearby department store save on both time and money

involved in the search process but may lose in quality and price of the stereo. The study suggests that shoppers can

be differentiated by store type visited, e.g. stereo store versus department store by the number of stores visited. 40

This could have great implications for geographic research.

It has often been noted that not all individuals travel to

the closest facility which fulfills their trip purpose. The

reasons why some individuals make this extra effort have not

been investigated satisfactorily and are modeled only as a

stochastic probability of going to a destination that is

further away rather than to a closer one. Focusing upon the difference between those individuals who frequent only a small number of stores and those who go to many stores may enable the postulation of destination choice models which use relevant socio-economic characteristics to predict likely destination choice changes with changes in the trans­ portation network or changes in neighborhood character. If the people of an area exhibit behavior in which they go to a few nearby stores, changes in the road network making the downtown more accessible will not be likely to have an im­ pact. Alternatively, if most individuals of a neighborhood do travel to stores that are further away, the higher acces­ sibility of a downtown shopping area may have the desired impact and these people may be expected to alter their shopping patterns to shop in the downtown area.

A framework conducive for evaluating choice sets of individuals in terms of the total number of stores and the number of perceptual groups contained within the set is pre­ sented in a paper by Wright (1975). The paper takes the view that decisions are made within a choice environment. This 41

environment consists of: 1) number of options; 2) queuing

time per option; 3) distractions (other commitments); and

4) time pressures. This environment will affect a satis­

fying (non-optimal) versus optimizing tradeoff in the

decision making process.

♦ The type of choice environment will have an effect upon

the type of choice strategy employed by the individual.

Accurate forecasting of an individual's actions will depend

upon modeling consumer choice with a strategy most like the one the consumer himself will use. Consumers often have to compromise between the benefits of an optimal decision among alternatives and the strains of making such a decision. If the strain of decision-making is perceived ; as too great for the individual, the problem can be simplified by using cut­ off or screening criteria to reduce the number of alterna­ tives or by using only a few attributes to make a decision.

One element of the consumer's choice environment which would influence using an optimal versus a simplifying choice strategy is the perceived cost of a suboptimal choice. The perceived costs of a suboptimal choice may lie with the degree of satisfaction with the alternatives an individual considers as a feasible alternative. If the number of alternatives is small and nearby, as suggested in time-space studies, but satisfaction levels are high, the perceived costs of a suboptimal choice may be low. In other words, going further away to a more optimal destination may not be 42 perceived as being worth it. If satisfaction levels for alternatives are low, then some other constraints are influencing the consumer to visit nearby destinations.

Another element of the choice environment which may influence the degree of optimality of choice strategy is the information load necessary to make an optimal solution. This information load increases with the number of attributes per alternative considered, the number of alternatives considered and/or the number of distractions present. Selection of a few overriding attributes or a small number of alternatives may also indicate the presence of a more non-optimal choice strategy on the part of the consumer. The use of simplifying strategies under time pressure and distractions has been observed in Wright (1975).

In the case of grocery shopping in an urban area, a vast number of stores exist as possible alternatives for the indi­ vidual. It is of importance therefore to examine the subset of stores that the individual actively patronizes to deter­ mine the number and variety of store types of a choice set for the purpose of food shopping. The examination of this choice set would provide information as to which of the many types of non-optimal choice rules may be applicable to destination choice for grocery shopping.

Recent work in destination choice theory calls for the incorporation of non-optimal choice models in future destina­ tion choice studies (Burnett, 1978; Sheppard, 1980). Several 43 models of non-optimal choice exist, primarily in the psycho­ logy literature such as Tversky's (1972) elimination by aspects model, conjunctive and disjunctive models

(Einhorn, 1970). These models differ from the more familiar utility maximizing models in that the individual does not make a choice by a complete series of tradeoff considerations to arrive at the best choice for his personal needs. Non- optimal choice models suggest the decision process is much simpler, using, possibly, time and distance as screening criteria to reduce the set of possible destinations to a select few. The actual decision process is then made from this small choice set.

Future research in destination choice in a behavioral context appears to be in the direction of studying the com­ position of individuals' choice sets to arrive at a frame­ work for evaluating alternative non-optimal choice models which best model the actual decision-making process of the individual as well as replicate his overt actions.

2.6 Summary

The problem of destination choice has received consider­ able attention in travel demand modeling. The first attempts at modeling destination choice behavior exist in conventional urban transportation planning in the form of aggregate co- variance models. These models are classified as being either growth-factor, gravity or intervening opportunity type models. 44

These models are termed trip distribution models. Dissatis­

faction with these models arises from the fact that they are based on zonal aggregates and that the models are descrip­ tive in nature and do not explain travel behavior or incor­ porate any policy related variables which planners could manipulate. Due to these shortcomings, modeling strategy was shifted to an individual or micro approach so that individual behavior could be included explicitly in model formulation.

Early attempts at micro type models assumed individuals use a utility maximization strategy in destination choice.

These attempts borrowed from economic consumer choice theory.

The two most well known models of this gendre are the spatial preference and the multinomial logit models. Each assume utility maximization under conditions of complete knowledge.

Objections to these models stem from the large number of alternatives that must be accessible to all individuals for the models to be valid and the assumption that individuals are capable of truly discriminating between destinations by utility maximization schemes. Empirical evidence suggests that individual choice sets, the set of destinations considered truly accessible, are too small for such a utility maximization strategy. This suggested the existence of several factors which act to constrain the individual from acting the way he would choose to do if these constraints 45

were not present. This observation was in part responsible

for the constraint modeling approach.

The constraint approach is most apparent in space-time

budget studies. These studies show institutional factors,

government, corporate actions, life cycle, and assorted

interest groups as factors which act to constrain individual

action. While such space-time budgets are useful for study­

ing individual action spaces, the choice environment under

which these choices are made have proven too complex to

model with the current state of choice theory. Due to

significant constraints upon individuals, the study of the

actual levels of knowledge maintained by individuals was

seen a possible approach to modeling choice behavior in a

constrained environment.

The view that actual supply of alternatives is not as

important as perceived supply is a basic tenet of demand oriented modeling. Mental maps, cognizied environments and household action spaces are examples of demand oriented

research. Demand oriented modeling studies confirm findings of space-time research by revealing limited information levels as distance from major activity centers for the

individual increase. This approach also shares a very com­ plex choice domain of travel behavior in that other space­ time considerations and interactions with other people influence the actions of people. Current choice models are incapable of operating in this modeling domain. Attempts to 46

structure choice strategies to include limited information

and the stress of time and other commitments are now being

studied in a choice set approach.

Choice set models are concerned with establishing what

portion of the actual environment relevant to the specified

trip purpose are actually considered by the individual.

Choice is viewed as taking place within a choice environment

of a number of options, queuing time per option, time pres­

sures and other distractions. Research in this approach is

being conducted currently. A major finding from this re­

search is that the choice sets of individuals are small and

relatively constant for those who have an established resi­

dence of a few years in an area. Thus, the individual

satisfies some criteria in gathering enough information

(destinations) to fulfill his trip purpose requirements.

This finding suggests that important information about the

behavior of the individual can be determined by close scru­

tiny of the choice sets of individuals.

Selection of a single relatively important trip purpose

of a repetitive nature which involves several possible

alternative destinations would be a sound approach to the

study of choice sets. A single trip purpose would likely

limit the behavior being modeled to an independent domain.

Grocery, clothing and other high order goods (e.g., stereos) are examples of such single trip purposes. This is not to

imply that these types of trips are made independently from 47

other trips, but that these trips may be the major purpose

of the observed travel behavior. To collect information on

the number of alternatives actively considered for a given

trip purpose a personal survey must be used. The techniques

involved in collecting choice set data is presented in the next chapter.

Many interesting questions arise in regard to choice sets. For example, for a given trip purpose, how many stores exist in an individual's choice set and how are these stores perceived in terms of similarities-dissimilarities. Number and perceptions of stores in the individual's choice set is termed the composition of the individual's choice set by the author. Another topic of interest is to determine the nature of the covariance of situational variables such as age, sex, income, family size, etc., with the composition of the choice sets of individuals. The relationship of size of the choice set and the geographic proximity and satisfaction levels of these destinations is also of interest. Recent choice set modeling work advocates the use of non-optimal choice theory for the modeling of destination choice. While empirical research in non-optimal choice behavior is just starting, relationships between increasing complexity of the choice environment and the use of non-optimal choice, usually involving the use of screening criteria to reduce the number of alternatives considered have appeared in the results of clinical testing. By examining the composition of the destination choice

sets for individuals with corresponding situational variables,

the complexity of destination choice for the individual may be defined. Identification of possible screening criteria used to simplify choice, such as distance, and satisfaction levels of alternatives would provide information for future studies which may consider how destinations become included or excluded from the choice set, how destination choice is actually made, etc., by the individual. CHAPTER THREE

METHODOLOGY

The investigation of the composition of the choice set

for individuals requires the collection of a large quantity of specialized information. Such information includes store perceptions and groupings, the attitudes of individuals

towards destination characteristics, information sources and situational variables of the individual. The purpose of this chapter is to describe in detail the methodology required for the collection and analysis of this data.

Data collection by personal survey is at best a compro­ mise between many conflicting goals. The need for a survey to gather data for this study is that the desired information is too specialized and detailed to be found in any existing data set. The advantage of a survey for this study is that the nature of the information collected can be exactly that which is desired. No substitute variables are necessary for the investigation of people's behavior. In designing a survey one must consider the costs in time and money to gather the information needed. It is necessary to incorpo­ rate the use of conjoint pairing and object similarities techniques so that the survey will not present the respondent with a task which is too difficult or take too long a time

49 50

to complete while also collecting the desired information.

Parts of the survey require direct contact between respon­ dent and interviewer and therefore requires a personal

survey as opposed to a form mailed through the mail. While

this feature increases the amount of time necessary to

administer the survey, the information gathered by the sur­ vey will be more consistent due to the presence of the

interviewer.

An interview time of thirty to forty-five minutes is chosen as a maximum length of the interview time. Beyond this maximum, it was felt that respondent resistance would become too great. Past studies on destination choice indi­ cate that choice behavior becomes stable after a person has established residence in an area for some time (Burnett,

1973). For this reason, a follow up survey is not deemed necessary. Therefore, the type of survey chosen for this study is a personal contact survey of thirty to forty-five minutes duration without a follow up.

A flowchart of the study, outlining the relationships between each section of the survey and the analysis tech­ niques is presented in Figure 3.1.

There are three major components to the survey. The first section is the usual personal profile of the respon­ dent's household. Age, size of household income levels, length of residence, etc., are obtained for each household to be used to examine any correlations between these 51

Section I Section II Section III Socio-economic Store Perceptions Attitudes Towards Profile Grocery Stores

Situational Store Variables Enumeration

Satisfaction Levels Tradeoff Matricies Information Sources

Similarities Matricies Data

Analysis Perceptual Grouping Grouping of of Stores by Individuals Individuals by CHOICE SET COMPOSITION Attitudes

Agreement of Structure of Store Evaluations Preference by by Neighborhood Group Conjoint Analysis

Choice Set Covariance

Figure 3.1

Flow Chart of Study 52

situational variables and provide a market segmentation of

the respondents. Included in this section is data on the

distance to work and number of cars owned. This information

is to be used as a surrogate measure of mobility. Also, the

number of individuals in the household who hold full or part-

time jobs which contributed to the household income is ascer­

tained to examine possible differences in grocery shopping

behavior between one and two or more paycheck households.

The second section of the survey is designed to investi­

gate the manner in which the individual perceives stores

within his choice set. The first part of this section con­

sists of an exhaustive list of all food stores the indivi­ dual has patronized for the past three months prior to the

time of the survey. For each store, the number of visits per month and average amount of time spent per visit is

recorded. A second part of this section of the survey asks

the individual to rate how satisfied the respondent is with each store on several store attributes by means of a seven point Likert scale. Also with the use of Likert scales, data on the importance and frequency of use of information sources such as television, newspapers, etc. is obtained. This infor­ mation is collected to insure that grocery store selection is a rational decision in that it is based upon information gathered by informed individuals. The relationships between satisfaction levels for alternatives and the number of alter­ natives in an individual's choice set are investigated. Also of interest is to examine the amount of variance among 53

individuals' responses to satisfaction levels for individual

stores. This is of interest to examine whether there is

close agreement among respondents as to the way they perceive

specific stores.

After obtaining a list of all food stores actively

patronized by the respondent, a similarities-dissimilarities

task similar to the "pick k out of n minus one" task is per­

formed. This task is designed to obtain data on the indivi­

dual's perceptual groupings of stores within his choice set.

A reference store is picked from the list of stores

obtained from the respondent and the person is then asked

to classify the remaining stores as being either similar or

not similar to the reference store. In this way, from the

N number of stores, one is chosen as a reference store. Out of the N minus one stores left, k are picked as being simi­

lar to the reference store. Thus the pick k out of N minus one task.

Since the task is new to destination choice studies and

is a key part of the survey design, the results are checked by analyzing the data for groups of individuals who live in close proximity within the survey study area. This is accomplished by making the choice set the same for each per­ son. For a given group of people, several stores will most likely be patronized by many of the group. These stores will become the choice set of stores for the group or neigh­ borhood. Individual responses as to the similarities or 54

dissimilarities between stores will be entered only for those

store pairs the individual patronizes. These store pairs

involving stores not in the individual's choice set will be

included in the analysis as missing data. The purpose of

this analysis is only to check to see if some agreement

exists in store perceptions across individuals and that the

pick k out of N minus 1 task produces reliable results.

The third section of the survey deals with peoples'

attitudes towards grocery stores. In developing a choice model, it is necessary to provide some basis by which to group individuals with similar behavioral patterns. The

identification of such groups becomes very difficult for destination choice. Past studies have shown that many indi­ viduals shop at just a few nearby stores. This suggests that individuals with similar objectives may shop at totally different stores simply because they live a few miles apart.

For example, both may wish to shop at the nearest store. But because the two people live a few miles apart, the nearest store is different for each. Clearly, individuals can not be grouped on the basis of their choice sets. However, peoples' attitudes do provide a strong indication of their behavior.

Those individuals who share similar attitudes as to the rela­ tive importance of various store characteristics may very likely share similar behavioral patterns. The attitudes of individuals toward grocery store shopping are identified by utilizing a conjoint pairing technique where pairs of 55

discrete levels of store attributes, for example, low,

average and high prices are traded off against another set

of attribute levels such as low, average and high levels of

quality of meats.

In this study responses are grouped by attitudes rather

than by grouping by similar socio-economic characteristics.

It is the individual's relative values placed on the levels

of various attributes that reveal his attitudes and these

values are referred to as the structure of preference for the

individual. While grouping by attitudes might not be as

convenient for the purposes of planning, this approach may

give a more revealing view on the range of variation in

behavior between groups.

Once the three sections of the survey are completed,

analysis can then begin. The attitudinal section (Section

III) allows for the grouping of individuals with similar

responses and assumedly similar behavior patterns. The

analysis of each identified group by conjoint pairing analy­

sis will identify the structure of preference, or relative

importance of various attribute levels, for each group. This

information, along with the individual's perceptions of

stores (Section III) and socio-economic characteristics

(obtained in Section I) are used to examine the relationships between peoples' attitudes, perceptions of stores in their choice set or choice set composition and the socio-economic characteristics of the respondent's household. 56

Having defined the overall structure of the survey, the

following sections of this chapter describe in detail the

sampling methods, survey techniques and analysis procedures

used to examine grocery store perceptions, attitudes and

choice behavior.

3.1 Sampling and Study Area Requirements

An acceptable study area for this survey must meet a

few very crucial criteria. First, the area must be compact

in size. Several studies (Pirie, 1976; Black, 1971) have

concluded that demand is not independent of supply. There­

fore it is desirable that the study area be small in size

so that all individuals within the area have reasonably the

same accessibility to a common group of stores. In this way

the supply of stores is perceived as being the same by all

individuals. Access to nearly stores is desirable to insure

that a true choice among several alternatives exists for the

individual (Eyles, 1971). On the other hand, the study area should be large enough to contain a large number of food stores so that the individual is not overly constrained by his environment to the point where almost no choice exists

(Hagerstrand, 1969). Another important consideration in choosing the study area for the purposes of this study is to find an area of at least lower middle to upper middle income households. Income levels should vary between lower to upper middle income and vary by life cycle in order to 57 investigate the changes in destination choice across these variables. Concentrating on middle income groups should insure a relatively high mobility factor for most of the sample population. High individual mobility should increase the number of possible stores to which the individual has access.

A stratified random sample procedure is utilized within the study area. A U. S. Census map is used to identify census blocks within each neighborhood of the study area.

To insure that the sample is spatially stratified, a list of randomly selected census blocks equal in number to the number of desired respondents for each neighborhood is made for the purposes of selecting a sample on the basis of census blocks.

A census block can be selected more than once from the list.

The respondent is then randomly selected from a list of addresses included within the block on the list. If the census block is selected twice, then two respondents are randomly chosen. This insures all residents within the study area have an equal chance of selection while also ensuring an even coverage of the study area so that any spatial biases in travel behavior within the study area will be evenly sampled.

In sum, the desired study area for the study should be small in size to insure individuals are familiar with most nearby stores. It must also encompass several viable alter­ natives to provide a large number of stores as feasible 58

alternatives. Lastly, the area must be of sufficient an

income level to insure high mobility for the sample house­

holds. These characteristics for the study area are deliber­

ately chosen to enhance the possibility of large destination

choice sets, i.e. high mobility, large number of nearby

stores. This strategy is to insure that no environmental constraints exist for the individual of such a magnitude so as to preclude choice. If small choice sets are found in the study area, then some other factor must be at work.

3.2 Survey Section I: Situational Variables

This section of the survey is designed to obtain a situational profile of the respondent's household. The usual characteristics of households are included. Age, sex, of respondent, marital status and number and age of off­ spring are determined for assessing the stage of life cycle present in the household. Total household income, auto ownership, education levels of husband and wife when appli­ cable are also included. The number of people employed within the household is obtained to examine possible shopping behavior differences between households where both spouses work and the more traditional one paycheck household.

A few household characteristics more specific to this study are also included. Commuting times by mode of trans­ portation for the head of the household and, when applicable, spouse are obtained as a partial measure of mobility for the 59 household. Also, the frequency of shopping trips by mode of transportation is obtained.

Finally, length of residence at current address as well as within the Franklin County area are obtained as a measure of familiarity of the surrounding area.

3.3 Survey Section II: Store Perceptions and Satisfaction Levels

The purpose of this section of the survey is to collect information as to the individual's perceptions of the stores in his choice set. This entails the identification of all stores the individual patronizes for the purchase of food, beer and wine to be consumed at home. The types of stores the individual should include are: supermarkets, convenience stores, butcher shops, wine and cheese shops, carry-outs and delicatessens. Restaurants, for example, are not considered in this study. Store name, address, frequency of visits as well as a "running average" of amount spent at each store per visit are elicited from the respondent.

Data on the frequency and importance of various food store information sources on the store selected or food pro­ ducts selected by the individual is collected to examine the degree to which the sample utilizes these sources as an aid in making grocery shopping decisions. The subject is asked to respond to a seven-point Likert scale to indicate the frequency or importance of each information source as an aid to grocery shopping. An example of the scales of frequency 60

and importance of information levels is presented in

Figure 3.2.

Having enumerated all stores actively considered by the

individual for grocery shopping, it is necessary to evaluate

perceptual groupings within the choice set and the under­

lying dimensions by which these stores are perceived. This

can be accomplished by the use of the "pick k out of n minus

one" task. This method is well documented in Coombs (1964).

Empirical applications of this technique are detailed in the

literature (Rao and Katz, 1971; Dobson and Kehoe, 1974). The

purpose of this task is to collect similarity judgments on a

given set of attributes. However, the method can easily be

modified to collect similarity judgments on destinations by

substituting destinations for the attributes of the subject

being investigated.

The task is performed after the stores patronized by the

individual have been elicited. The stores are recorded on a

similarities matrix survey device (see Figure 3.4). For the

purposes of facilitating the task for the respondent, the

store names are recorded on 3 x 5 index cards and are then

placed in front of the subject in a pile. A card is selected

by the interviewer and is placed in a spot separate from the

rest of the cards. This card bears the name of the "refer­

ence store."

The respondent is then asked to place the cards with the other store names in his choice set that are similar to the 61

Frequency ▼ery never often 1 2 3 4 5 6 7 where: 1 - very often 2 - often 3 - more often than others I know 4 - as often as others I know 5 - less often than others 1 know 6 - seldom 7 - never

Importance very not at Important all 1 2 7 where: 1 - very important 2 - important 3 - more important to me than others I know 4 - as important to me as others I know 5 - less Important to me than others I know 6 - slightly important 7 * not important at all

Figure 3.2 Frequency and Importance Scales 62

Stores similar to reference store Stores dissimilar

Reference Store

Stor e LStore A Store LStore

Store MStore B Store MStore

Store C Store N

Figure 3.3 Respondent Task "Pick k out of N-1" 63

Croups different Store is like: B 9 10 11 because:

1. ______A.

2. ______S.

3. ______C. 4. ______2).

5 . ______E. 6. ______?. 7. ______G. 8. ______E.

9. ______I.

10 . ______J. 11. ______Z.

A. B.

C. 2).

E.

P.

C. E.

I.

J. Z.

Figure 3.4 Similarities Matrix 64 reference store under the reference store card to form a column. Those cards bearing the names of stores that are dissimilar to the reference store are placed in another column, as in Figure 3.3. The left column of cards are those stores similar to the reference store and the right column are those stores dissimilar to the reference store.

The results are recorded by the interviewer across the row of the similarities matrix (Figure 3.4) corresponding to the reference store card. This task thereby divides the stores into two groups. The interviewee is then asked what characteristic or attribute differentiates the two groups. The response is recorded on the lines below the similarities matrix corresponding to the reference store row. The task is then repeated and results recorded until all stores in the individual's choice set have been used as the reference store.

The survey device used to record the data for this task is shown in Figure 3.4. The stores enumerated from the respondent are listed in the numbered blanks at the left hand side of the matrix. Corresponding numbers label the columns of the matrix. The store listed in the blank on the side of the matrix serves as the reference destination or store. For example, the third row represents the "pick k out of n minus one" task where the third store elicited from the respondent is the reference store. Thus the respondent is asked if store

1 is like store 3, store 2 is like store 3, store 4 is like 65

store 3 until store 3 is compared to all other stores. If

the store in question is judged similar to the reference

store (store 3) a one (1) is placed in the appropriate box

in the matrix. If the decision is not similar, a zero (0)

is placed in the appropriate box. Thus the individual is

asked to make a simple binary choice between store pairs.

The pick k out of n minus one task is then repeated for

each store serving as the reference, i.e. all n rows are

completed, n being the number of stores elicited from the

respondent. Due to the possibility of the respondent using different store attributes for discriminating between store pairs, the resultant matrix need not be symmetrical.

There are several advantages to this technique. The

task only requires a binary similar - dissimilar response so

the judgments can be made quickly and accurately. The speed at which the task can be accomplished is most useful when data is to be collected on a long list of items or a subset of a long list of items. Finally, the data collected by this method can be transformed for analysis by multi­ dimensional scaling techniques.

3.4 Multidimensional Scaling: Analysis Technique for Store Perception

Many versions of multidimensional scaling algorithms exist. The algorithm selected for analysis purposes for this study is the Alscal-4 multidimensional scaling package by

Forrest W. Young and Rostyslaw Lewyckyj (1979) . This program 66

represents the most advanced state of multidimensional scal­

ing (MDS) routines. The program accepts data of all forms

measured at nominal, binary, ordinal, interval or ratio

levels and has no limitations on the size of the number

of stimuli or subjects or total amount of data. Many models

of MDS are possible in this package including an asymmetric

model for multiprocess asymmetric data (Young, 1975). An excellent book which introduces the techniques of multi­ dimensional scaling is available by Kruskal and Wish (1978).

The basic problem the MDS routines address is to

"represent n objects geometrically by n points, so that the interpoint distances correspond in some sense to experimental dissimilarities between objects." (Kruskal, 1964). This means that a set of stimuli, say for example, a set of food stores can be represented in an n dimensional space so that those stores perceived as being very similar are very close together while those perceived as being quite different are spaced far apart.

Critical to MDS is the relationship between experimen­ tal dissimilarities (the data) and the distances in the stimuli configuration. An important advance in the MDS problem was made by Shepard (1962) when he suggested the utilization of a monotone relationship between dissimilarities and distance. The significance of this lies in the fact that dissimilarities may exist in ordinal or interval levels while distance is measured at the ratio scale level. That is, 67

dissimilarities may be rank ordered from most similar to

least similar. The difference between the objects that are

third and fourth is not the same as the objects that are

fourth and fifth. The difference between three and four

inches however, is the same as the difference between four and five inches, namely one inch.

Thus multidimensional scaling statistically fits object dissimilarities to the distances of the object configuration

in space. This is accomplished by performing "a monotone regression of distance upon dissimilarities, and use the residual variance, suitably normalized as our quantitative measure. We call this measure stress." (Kruskal, 1964, pp. 2-3)

Stress is, therefore, a residual sum of squares. The smaller the residual is the better. A verbal evaluation of stress from experimental and synthetic data from Kruskal

(1964) is summarized in Table 3.1.

Stress Goodness of Fit 20% poor

10% fair

5% good

2.5% excellent

0% IIperfect"

Table 3.1 68

3.5 MDS Analysis of Store Perception Individual and Neighborhood Level (from Survey Section II)

Two basic analyses of the dissimilarities data are

performed. The first involves the analysis of each indivi­

dual's matrix to identify the individual's perceptual group­

ings of the stores in his choice set. The second investiga­

tion is to analyze the dissimilarities between a collective

set of stores that are patronized by a number of respondents

in a given area. This is to determine if the perceptions of

stores by individuals is unique for each respondent or if

there is general agreement as to how food stores are per­

ceived as reported by the similarities matrix technique used

in this study.

The first analysis performed for each similarities ma­

trix involves an asymmetric multidimensional scaling model

(see Figure 3.5). The objects comprising the rows and columns of the square matrix are the stores in the indivi­ dual's choice set. The entries for the matrix are the "l's" and "0's" obtained from the respondent. The output is an object configuration of stores from which the number of perceptual groups in the choice set can be identified.

The pick k out of n minus one task is further modified for the purposes of this study. Completion of a row of the similarities matrix (see Figure 3.4) effectively sorts the set of stores in a choice set into a group of stores similar to the reference store and a group of stores dissimilar to it. 69

Input: Single nxn square ssyrraecric matrix 0 of dissimilarities

(oij * °ji>

Number of ways: 2 J Number of modes: 1 E C T I 6 .

OBJECTS

Output: Two rectangular matrices: 1 1 X Is an nxr matrix of coordinates 2 2 • (x^a Is coordinate of i'th object on O 0 • • b ; a'th dimension); and V Is an nxr matrix B J j of object weights (v^a Is weight of E *la via C 1 a. T * object 1 on dimension a). T • S • s •• Number of ways: 2 • Number of components: 2 n n 1 ••• i >•■ f Prom Alscal-4 User’s Guide DIMENSIONS DIMENSIONS

Model: Asysmetrlc Euclidian m *1*^1

"u • \/ Figure 3.5 Asymmetric Multidimensional Scaling 70

This allows for the opportunity to elicit from the respondent

an attribute by which the individual perceives a difference

between the two groups of stores. Over the course of the

survey task, a series of salient attributes by which the

respondent discriminates among the stores is obtained from

the respondent.

This modification is analogous to Kelly's triad sort

method of identifying attributes used in decision-making

(Kelly, 1955). The triad sort method identifies those

attributes which the individual uses to distinguish between

objects. The method involves ^resenting an individual with

three objects from a larger set of objects. The subject is

then asked to suggest a characteristic by which two of the

objects are the same but different from the third. The sort

is repeated until all triads possible are presented or until

no new characteristics can be elicited. In this way a set

of characteristics by which the individual discriminates

between objects in the set is obtained without the use of

characteristic lists or cues.

The modified task developed for this study is more

conducive to destination choice studies than Kelly's triad

sort. To use Kelly's triad sort in destination choice, three

stores would be selected at random and the individual asked why one is different from the other two stores. Using this

technique, the individual may be asked to consider the dif­

ferences between a group of three stores that he considers 71

to be the same. The differentiation between stores in such

a case would be an artificial contrivance on the part of the

respondent in that the task demands that he must differenti­

ate between three similar stores. The modified pick k out

of n minus one reference store. Thus the individual can

consider his entire choice set instead of just three stores

and considers differences between store groups that he

himself considers different.

This modification is of great importance in regards to

the interpretation of dimensions identified in MDS analysis.

Stores will be placed in a geometric configuration in

Euclidean space. By examining the manner in which the

stores are grouped, the MDS grouping can be compared to the

groupings in the similarities matrix. Those attributes

which have similar groupings as the MDS configuration can be

used to provide an interpretation of the dimensions of the

MDS solution.

An empirical example of the task can illustrate this point. The respondent's responses to a list of stores elicited from him as well as the salient attributes are

shown in Figure 3.6. The resultant MDS store configuration

is presented in Figure 3.7. Four perceptual groups of stores are discovered: A and D, E and C, F and G, and store B.

The interpretation given to the store configuration is that stores A and D, E and C, F and G are pairwise considered very similar in nature. Comparison between pairs show that 72

Groups different Store is like: 1 2 3 4 5 6 7 8 9 10 11 because:

(A) 1. Pi P *ear. .( G r a n d ™ p 1 1 1 p 1 A.

(B) 2. 5ic Sear (Keil) Q ’1 Q 0 0 0 B.

(C) 3. Korth Karket P p 0 1 p p C.

CD) 4. French loaf Baker- 1 0 0 0 0 0 D.

(2) 5. Rife's Deli 0 0 1 1 0 0 E.

(F) 6. Bia Bear (lane) 0 0 0 1 0 1 F.

(G) 7. Bia Bear (Einrdl. 0 0 0 0 0 1 G.

8. a.

9. i.

10. j .

11. — X.

A. Quality of goods______

B. Distance______

C. Quality______

D. Quality______

B. Specialty shops vs. general

F. Frequency of visits (distance)

g# Frequency of visits (distance) B. ______

I. ______

J. ______I. ______

Figure 3.6 Similarities Matrix Example ■ INCIWKM I ( MVRIZOKTALf VS DIFIM RIOS 2 (VCRTICAL1 MW. tttt.tttt.M tt.tttt. 2.20 2.23 2.17 0) 2. 17 2.011 -t4 2.011 2.00 V) 2.00 1.02 v* 1.42 l.»:i > 1.113 1.70 1.70 I. *7 O 1.67 I.Oil 1.30 1.00 t*. 1.30 1.42 o a 1.3.1 1.42 «>^ r 1.33 1.20 1.23 I. 17 n I.on t>a* I.1.011 17 1.00 U 1.00 0.02 0 .4 2 o.ir.i 0.113 0.70 0.73 0.67 0.67 o.on n.no o.oo o.so 0.42 0 .42 0.33 0.33 0.23 0.23 O. 17 O. 17 0.00 Quality o.on - 0.00 O • -o.oo -0.60 * -o.on -0.17 * -o . 17 - 0.20 -0.33 i -0.23 -0.42 -0.33 - 0.00 -0.42 -0.00 *» -0.30 -0.67 «t -0.8H -0.70 -0.67 -o.nn m -0.73 - 0.42 9 t-0.0 3 -1.00 0 • -0.42 - i.o n * -1.00 - I . 17 * -I.Oil - 1.20 0 - I . 17 -1.33 -1.23 -1.42 -1.33 -1.00 0 0 -1.42 -i.o n -1.00 -1.67 ft -1 .35 -1.70 -1.67 - 1 .03 * -1.78 - 1.02 * - 1 .03 - 2 .00 o • -1.42 - 2 .on - 2 . 0 0 -2. 17 - 2.011 - 2.20 -2 . 17 -2.23

•a 0000* -S'oOOO -?* 0600 2" ■ •*»*•. 1.7300 . 2.2000. 1.0000 - 1.0000 - 0.0000 - 0.0000 0.0000 1.0000 1.0000 2.0000 2.0000

- j Figure 3.7 w Store Configuration 74

stores do differ in some respect. Note that store B is considered as being different from all other stores except for North Market (C) in terms of distance (line 2 and attribute B on Figure 3.6). Because of this, the MDS con­ figuration places store B in an area all to itself. Dimen­ sion one is interpreted by the researcher as quality due to the fact that stores A, C, D, E, and G are similar in quality as compared to B and F, (line 1, Figure 3.6) and also stores C and E are similar in quality (line 3) and

A and D (line 4). Looking at store configuration for dimension one (quality, Figure 3.7) E and C and G and F are grouped together. In terms of frequency of visits, stores F, D, and G (line 5, figure 3.6) and stores G and

F (line 6, Figure 3.6) are considered similar. The config­ uration along the frequency of visits axis (Figure 3.7) groups the stores in a similar manner.

The purpose of the analysis of each similarities matrix individually is to describe the perceptual groups within each choice set. Due to the small size of the individual's choice set, this usually being less than six stores as found out in the results, MDS solutions of greater than two dimensions are unreliable. This is because six stores can be easily placed in two dimensions to represent the dissimi­ larities the respondent reports. Thus, this type of analysis should only be conducted with two dimensional MDS solutions 75 unless there are a great deal of stores in the individual's choice set.

Upon completion of the analysis of individual choice sets, investigation of the degree of agreement among indi­ viduals as to the perception of stores is necessary. This is the second type of MDS analysis performed in this study.

Since the choice set of individuals varies, only a small subset of all stores patronized by members of a neighborhood can be included in this aggregate choice set. Stores patronized by several respondents in a sub-area of neighbor­ hood of the study area will be included in a common choice set. Stores in the common choice set which are in each individual's choice set can be entered in an asymmetric individual differences model as illustrated in Figure 3.8.

Most of the data matrix for an individual is expected to be missing data. This is due to the fact that comparisons between stores not in the individual's choice set (unknown- unknown) and comparisons with stores where one is in the choice set and one store is not (known - unknown) are called for in the aggregate choice set. The unknown - unknown and known - unknown conditions are entered in the analysis as missing data for no comparison of stores in these situations are made by the respondents of the survey. Only known - known comparisons are valid data entries.

If too many stores are included in the analysis, direct comparisons by individuals may not exist for some stores. 76

Input. a (>2) u s aquara aayaaatrlc i»ti Mtrlcaa Ok , an* for aach of a oubjaeta (o0j^k la dlealailarlty 1 and J for aubj Number of wnyes Nuabar of aodea

OBJECTS S u B • 1 Output: Thrac rectangular aetrleao: X la J • 2 J’ an oxr matrix of the coordinate a of 1 O Sta object 1 on d loan a Ion a; V la as oxr ■ b aatrlx of the walghta of object 1 on J V- • X<1 via dlaenalon a; and V la an ago- aatrlx of C * T * the walghta of aubjeet k on dlaenalon a. g * S * Nuober of wojra: 3 *

Number of coaponanta: 3 •> a m From Alscal-4 l»a«*i i»,a •«' User's Guide DIMENSIONS DIMENSIONS DIMENSIONS

Hodel: Vclghtad aaynaetrlc Euclidian o ^ k •

• V |VkaCV* j / Figure 3.8 Weighted Asymmetric Multidimensional Scaling 77

For example, assume there are four stores, A, B, C, and D.

People on the east side of an area patronize stores A and B.

Those in the middle of the area patronize stores B and C.

Those of the west side patronize stores C and C. Stores A,

B, C, and D are situated on an east-west road from east to west respectively. Each individual thus compares A to B or B to C or C to D. No individual directly compares A to

D since the stores are located so far apart. The Alscal multidimensional scaling routine will place store A and D in a store configuration far apart because they are not directly compared. This result may occur even though in reality people would have considered stores A and D as similar. By dividing the study area into subsections, this problem may be minimized.

Results from this analysis will allow for the possibi­ lity of reliable store configurations beyond two dimensions due to the greater information in the data. An important consideration in this analysis is the stress value obtained.

A low stress value will indicate that individuals are in relative agreement in their perceptions of stores and that the perceptions of individuals are not unique as analyzed by this technique.

3.6 Survey Section III: Attitudes Toward Stores and Struc­ ture of Preference

The purpose of travel demand modeling is at some point in time to be able to make aggregate predictions of travel 78 behavior (Ansah, 1977). In order to accomplish this with a micro-approach model, the sample is usually grouped by some market segmentation scheme so that group behavior can be estimated. Two approaches of segmenting a sample population are most commonly used (Golob and Dobson, 1974). The first approach is to separately analyze subgroups of the popula­ tion that are of interest to the researcher. Examples of such segmentation schemes can be found in Golob et al. (1972) and Gustafson and Navin (1972).

The second approach involves segmenting the sample on the basis of their perceptual judgments. As stated in Dobson and Kehoe (1974, p. 1) "...personal behavior in the selection of one course of action over another can often be determined in advance by an understanding of the perceptions and pre­ ferences that individuals have of the alternatives in question." The advantage of using this approach is that the various groups are guaranteed to be different in terms of their judgments. However, the groups may not fall into categories of direct interest to policy-makers but signifi­ cant insights may be gained from such an approach.

It is the second approach to market segmentation that is used in Section III of the survey. That is to say, indivi­ duals- are grouped by their perceptual judgments of food store selection. This presents a practical problem in regard to assessing preferences in destination choice sets. The stores in an individual's choice set are often a small 79

number and not common to all individuals in a sample. A

common set of perceptual judgments must be presented to all

individuals in order to determine their preferences. The

method for determining preferences is the Conjoint Prefer­

ence Estimation scheme presented in Knight and Mencich

(1974). By using this technique, called conjoint analysis,

a "structure of preference" for food store attributes can be established. What is meant by structure of preference is the specific form of the utility function by which choice is made.

The structure of preference concept is well established

in the literature, '(Fishburn, 1964; Goldberger, 1967;

Henderson and Quandt, 1971) running the gambit from general utility functions to more simple variants where utility functions are constructed for ease of calibration (Equations

3.1 and 3.2). A more restrictive model makes the problem feasible.

The general utility or preference function can be expressed as:

U = U.(qi...qn) (3.1) where: IK is the utility level (ordinal) and (q^...qn) are variables indicating the level of each of n attri­ butes.

A multiplicative - separable form (Fishburn, 1964) can be expressed as: N U = 7r U. (q. ) (3.2) i=l 1 1 Thus total utility is the product of N partial utilities

which are defined for only one attribute. This allows for

tradeoff estimates to be estimated independently from the

levels of all other attributes. This assumption is referred

to as "want independence" (Goldberger, 1967). "Want inde­

pendence" means that the tradeoff between two attributes is

not affected by the level of any third attribute being con­

sidered.

If a logrithmic transformation of Equation 3.2 is taken,

an equivalent additive - separable form is obtained. Trade­

offs will be obtained by comparing discrete levels of each

attribute in a tradeoff matrix as shown in Figure 3.9. The

individual numbers from 1 to K (K cells in the matrix) are

assigned to each cell of the matrix to show the individual's preference for varying levels of the two attributes corres­ ponding to the row and column headings. Thus on an abstract level, the individual is making decisions on what he would most prefer in a grocery store's characteristic makeup. This is done by considering tradeoffs between attributes two at a time.

Since attribute levels are discrete, the function U^(q^) can be set to the constant P.. for the preference for level j l j c J of attribute i. Equation 3.2 then becomes: Distance to 3tore (one way) C-1 2-5 4-5

Low

Price Medium

High

Figure 3.9 Tradeoff Matrix Example 82

For a set of N attributes, N (N-l)/2 possible pairings are possible. However, not all of these pairings need to be made. Direct pairings should be made between attributes that are of interest to the researcher. Comparisons between any two attributes should not have more than one intermediate attribute in the pairing of the various attributes.

Several programs exist for the analysis of the tradeoff matrices. MONANOVA by Kruscal, POLYCON by Young, PREMAP by

Carroll and MONORE by Johnson are examples. While each of these programs are significantly different in mathematical formulation, empirical work shows no major differences in results (Green and Tull, 1975). Due to availability,

Johnson's technique is used for analysis purposes in this study.

Input data for MONORE is the respondent's rankings from

1 to K for the K cells of the tradeoff matrix. The cell marked with a 1 is the most preferred cell of the matrix.

The algorithm calculates a P matrix of preference or part utility parameters P^j. In the multiplicative model, the parameters are standardized so as to sum to unity when summed over all levels of each attribute with no loss in generality.

At each iteration, the previous P^j's are compared to actual rankings by a badness of fit measure phi (Johnson,

1973). 83

m Z Z 2 k ij 6ijk (Zik " Zjk} (3.4) m Z X. 2 K 13

where: 1 if sign (ZiR - Zjk) = sign 6 i j k =

0 otherwise

k - attributes j - levels for a single attribute i - individual Yik - data observations

Phi thus resembles a rank order correlation statistic and

ranges from 1.0 (perfect fit to 0.0). Improvements are made

on the P matrix by a gradient search method which produces

a partial gradient matrix G, each element a partial deriva-

tive of phi with respect to P^j* These partials are used to

identify minimum or maximum value of phi, as in differential

calculus, and adjustments are made to the P matrix. The process iterates until specified stopping criteria are met.

Another test of fit involves multiplying the two appro­ priate calculated utility values to derive a predicted util­

ity value which is compared with actual individual rankings.

The comparison is made using Kendall's tau, a rank-order correlation coefficient. Tau values of .9 or better are considered reasonable for prediction while tau values from 84

.7 to .9 are useful for analysis (Knight and Mencich,

1974).

The magnitude of the weightings (p^^'s) will not only

identify the important attributes individuals use in destination choice but also the relative importance of each

level of a given attribute. These relative weightings are called the structure of preference.

Before actual analysis of the tradeoff matrices is undertaken, the individual responses are grouped on the basis of the similarity of their responses to the tradeoff matrices of Section III. Many algorithms for grouping are available. The technique used for this study is described by Johnson (1967). The technique, called Cluster, starts by treating each individual response as a separate group or cluster. Groups are then merged together one pair at a time by selecting the two groups whose responses are the most similar. The two groups are merged into one group and the process continues until only one group is left consisting of all individuals. Cluster uses a Euclidean distance matrix to determine group clustering. The distance matrix is computed by the formula:

dfx^j) = (x^ - x j) (xi - x j) (3.5) where: x^ is the ith observation vector.

As the clustering procedure continues, the amount of distance within groups increases. This information is used by the investigator to determine the appropriate number of groups 85

to use in further analysis. Upon clustering, the MONORE

conjoint program is used to investigate the structure of

preference for the groups identified.

3.7 Synopsis of Methodology

In order to study the composition of the destination

choice set for the individual, a personal survey is con­

structed to obtain the desired information. This survey

obtains information not available in other data sources

and collects data in a form that is analyzed by statistical

techniques that have only become available in recent years.

The survey is divided into three parts. Section I collects

the various socio-economic variables of interest in most

consumer behavior studies. This is where its resemblance

to most other surveys ends. Section II, through the use of

a similarities matrix, based upon the pick k out of n minus one task, provides information as to the perceptual group­

ings of stores as well as the underlying dimensions, or

store characteristics, by which the individual differentiates

between stores. The data from this task is analyzed by

Multidimensional Scaling Techniques to discover the actual perceptual groupings of stores. The third section of the survey consists of a series of tradeoff matrices to which the individual responds so that his attitudes toward dis­ crete levels of various store characteristics can be ascer­ tained. This section is of importance since it can be used to group individuals into homogenous groups based solely

upon their attitudes rather than by their overt behavior.

This feature is very important to a destination choice

study. Previous research has shown that most individuals

patronize stores very close to their residences. Two

behaviorally similar individuals placed in different environ­

ments, may exhibit different behavior patterns solely due to

the supply of alternatives. Using the tradeoff matrix data,

analyzed by conjoint pairing techniques, allows for the

grouping of individuals based upon their reactions to the

tradeoff matrices section which is independent of actual

store locations. In this way, using the data from all three sections, a more revealing study of destination choice beha­ vior can be undertaken. CHAPTER FOUR

ANALYSIS

This chapter presents the results of a survey conducted in Columbus, Ohio from June to August 1980 based upon the survey design described in Chapter Three. The main purpose of this survey is to examine the destination choice set of individual residents for a small area of suburban Columbus.

The destinations considered are only those concerned with the trip purpose of food shopping. The area selected for the study features several large shopping centers as well as a large supply of smaller food stores. While income levels are fairly constant, the area contains lower middle to upper- middle income levels. Given these conditions in the study area, this chapter describes the relationships between the composition of the choice sets of the respondents of the area, the attributes by which individuals discriminate between stores in their choice set and the common attitudes different groups possess regarding how they feel about several important features of grocery stores.

To report the results of the survey, this chapter is organized into five sections. The first section describes the study area and the socio-economic make-up of its

87 88 residents. Section two reports respondent's commuting times for the journey to work, frequency of shopping trips and information levels on food stores. This data provides information on the degree of mobility of the sample popula­ tion as well as checking that grocery store choice is a rational decision based upon information the individual has acquired. Section three reports the choice set composition of the individual. The number of stores the individual actively patronizes, the manner by which the individual groups these stores as being similar in nature or not, and the attributes of grocery stores by which the individual discriminates between these stores are described in detail.

The fourth section describes the contrast in attitudes between different groups of individuals regarding their attitudes towards the importance of various grocery store characteristics. The fifth and final section provides a summary of the major findings of the study.

4.1 The Study Area and Sample

Ideally, for purposes of this research, the study area should provide an environment where many alternatives exist and also contain households of sufficient levels of income that the opportunity to utilize all the alternatives are present. To achieve this goal, the study area must meet a few select criteria. First, it must be compact in size to insure that most individuals surveyed are familiar with and 89

have access to the same set of stores. Second, the study

area must encompass several viable and competing alternatives

for grocery shopping. Third, and lastly, the area must

consist of households of a sufficiently high income level

to insure a high degree of mobility. The area should also

encompass a range of income levels to examine variations

in choice sets with varying income and other situational

variables such as size of household, age, number employed, etc.

The study area selected for study consists of Census

Tracts 69.40, 69.22, and 77.30 which for a relatively

small block of land in the northern part of Columbus (see

Figure 4.1). The study area is about 3.5 miles long from north to south and averages 1.5 miles in width except for

Census Tract 60.40 which is 2.5 miles in width. While each of the Census Tracts are within the same area of the city, they do differ in socio-economic characteristics. For clarity in presenting the results of the analysis of the data, reference to these Census Tracts will often be made.

Census Tracts 69.40, 69.22 and 77.30 are referred to as the

North, Central and South neighborhoods respectively.

A large number of food stores exist in the study area

(see Figure 4.2). Twenty-three stores in the immediate area are identified as actively visited by the sample population.

Several other stores outside the study area are also patron­ ized by some of the respondents. This insures a large supply Study Area

Morse Road Cooke Road _/■

Downtown Cclucbus

Figure 4.1 Study Area Within Franklin County 91 of potential destinations for any individual in the area.

Sample size for this study, as in any study, is a tradeoff between time, money, accuracy and depth of informa­ tion required, as well as methodological limitations. The computational constraints of MDS studies tend to favor small data sets. For example, a paper by Burnett (1973) uses thirty respondents to investigate the dimensions for the choice of clothing stores. Given the nature of this study, requiring a 45 minute person to person survey, it was determined a priori that a sample size of about thirty respondents per neighborhood would be sufficient for the purposes of the study. Further sampling would be undertaken only if the nature of the responses indicated the extra effort to increase the size of the sample would prove useful.

Even though the study area is relatively small, vari­ ance in the perceived supply of grocery stores may exist for individuals residing in various parts of the study area and even within a given neighborhood. Depending upon the exact location of an individual, the relative location of area stores will change thereby changing the relative accessibility to all existing stores, in the view of the shopper. To overcome this possible variance, it is necessary to insure that the sample population is selected uniformly over the study area. For this reason the study area is spatially stratified by Census Block, a random list of blocks generated and then a respondent is randomly selected Schrock Road

I 270 49

1M

02 1£1

M •H 25 Ctf, 70*

■P

M o r s e Road

South Neighborhood Store Numbers on Tables: •51 4.7,4.9,4.11

Cooke Road

Figure 4.2 Store locations 93

from the Census Block address list. This strategy assures

that all households within the study area have an equal

chance of being picked while also insuring that all areas

within each neighborhood are sampled. A map showing the

locations of the respondents is presented in Figure 4.3.

In order to obtain some understanding of the character

of the three neighborhoods in the' study area, a few selected

socio-economic characteristics of the neighborhoods are

presented in Table 4.1. The three neighborhoods show, as

of 1970, a large diversity in each characteristic. Annual

mean income varies significantly between neighborhoods. The

South neighborhood has about a $13,500 mean income while

the Central neighborhood has a mean income of $1,000 higher

and the North neighborhood, $1,000 lower than the South.

The percentage of high school graduates varies from 77% to

85% to 90% for areas South, Central and North respectively.

Length of residence, while nearly the same for South and

Central, about 50%, is considerably lower for the North area which is 25%. This is because most of the North area housing, particularly in the eastern half is of recent construction.

This may in part account for the housing value of the North neighborhood surpassing the value of the South simply because

the housing is newer. Finally, there is a diversity in the size of the household for the three areas. The South area having the smallest number (3.29) of people per household, yet the highest percentage of long term residents. This Schrock Road

I 270

161

l§1

rH,

M o r s e Road

Cooke Road

Figure 4.3 Location of Respondents 95

Socio-Economic Characteristics of Neighborhoods

(Census Data 1970)

Table 4.1

South Central North

Census Tract 77.30 69.22 69.40

Population 4506 9589 7330

Mean Income $13,404 $14,383 $12,382

% High School Grad 77.4$ 84. 9% 90.1%

Same House for 5 Years 55.3% 45. 7% 25. 2%

Median Value House 22,100 27,100 23,200

Size of Household 3.29 3.59 3.83 96

suggests an older population in the South area since an

advanced stage of the life cycle usually means few

children still at home, an established residence, and a

moderate income. The North area has the largest average

household size (3.83) which suggests that many of the

residents are at a point where they have several children

at home. Based on these few figures, the study area shows promise as containing a population which varies in life cycle as well as a moderate variation in income levels which

is required to meet the goals of the study.

A selected list of socio-economic characteristics for the sample population generated by this study is presented

in Table 4.2. While some demographic changes have occurred

in the area, the needed diversity of age and life cycle have remained along with the desired variety of income levels and high mobility of the households. The percentage of high school graduates has increased for all three areas since 1970. However, the percentage of people who have lived in the same house for five years has shifted. Only

30 percent of the South area residents have lived in the same house for five years as compared to 55 percent in 1970.

Size of household has also decreased for the area. This information plus personal canvassing of the area suggests that the South area is in a state of transition. The neigh­ borhood has a large percentage of elderly and retirees whose presence, in part, causes the low household size, income 97

Table 4.2

Socio-Economic Characteristics of the Sample

Variable South Central North Total Sample

Age % under 40 37.0% 32.3% 59.3% 42.4%

Income

Under 3,000 0. 00% 0.00% 0.00% 0.00%

3,000 - 4,999 2.50% 1.25% 0. 00% 3.75%

5,000 - 6,999 3.75% 1.25% 0. 00% 5.00%

7,000 - 9,999 1.25% 0. 00% 2.50% 3.75%

10,000 - 14,999 7.50% 2.50% 1.25% 11.25%

15,000 - 24,999 15.00% 8.75% 10.00% 33.75%

25,000 - 34,999 1.25% 12.50% 16.25% 30.00%

35,000 - 49,999 2.50% 7.50% 1.25% 11.25%

50,000 - more 0. 00% 1.25% 0.00% 1.25%

Size of Household 3.19 3.61 3.63 3.48

# Employed in Household 1.22 1.58 1.33 1. 39

# Autos in Household 1.85 2.03 2.00 1.96

% High School Education 83.3% 93.4% 92.2% 90.1%

Same House for Five Years 30.0% 42.0% 55.6% 43.5% 98 and employment levels. The newer residents of the area are younger married couples with one or two children and have only lived in the area for a few years.

The Central neighborhood has also changed in the past ten years. While the size of the household has remained relatively constant other changes are evident. This neigh­ borhood has the highest number of people employed per house­ hold of the three areas. This is because many of the families in the area are two paycheck households. In many cases, both husband and wife are career oriented (93 percent high school graduate) as opposed to the wife having a low paying part-time job. Hence the existence of several house­ holds in the $25,000 - $50,000 income range. The average number of cars owned in this area is very slightly in excess of two cars per household. This suggests a high degree of mobility in this area.

The North section is more difficult to describe as a single entity. The western half is older and more inexpen­ sive housing predominates as compared to the eastern half.

While the area has the youngest population composition,

59 percent of household heads under forty, over 55 percent have lived in the same house for five years or more. Size of the household and car ownership are comparable to the

Central area but there are fewer two paycheck households and household incomes are lower than the Central neighborhood.

The larger percent of heads of households under forty 99

(compared to Central) suggests this neighborhood is in an early period of the life cycle with small children, an observation which corresponds with personal canvassing of the area.

In sum, the study area taken in total meets the desired conditions of the survey design of being compact, highly mobile, yet varying in income and life cycle measurements.

4.2 Journey to Work and Shopping Trip Frequency and Information Sources

Information on work travel times and number of shopping trips made per week are obtained to examine the degree of mobility of the sample population. As can be seen in

Tables 4.3 and 4.4, the sample population as a whole is highly mobile. Considerable time is spent by many of the sample households in the journey to work. Further, the typical respondent makes two, three, even four or more shopping trips per week in the course of an average week.

Thus, the respondents are most likely very capable of patron­ izing as many stores as they wish and are at least capable of expressing their effective demands for food stores.

Responses to importance and frequency of use of a list of information sources by the individual are presented in

Table 4.5. The respondents indicate the importance or frequency of use of various information sources on a seven point Likert scale. The scales range from one (1) meaning Table 4.3

Employment Travel Times (One Way) For •Respondents in Minutes

Time South Central "Nor

Not Employed 17 15 14 0 - 4 2 4 3 5 - 9 2 4 3 10 - 14 2 5 2 15 - 19 2 2 1 20 - 24 2 0 0 25 - 29 0 0 0 30 - 39 0 1 1 40 - 49 0 0 0 50 - 59 0 0 0 1 hour+ 0 0 0 Not Applicable * 0 0 3

Respondent's Spouse Travel Time Not Employed 11 8 2 0 - 4 0 2 3 5 - 9 2 1 5 10 - 14 6 5 3 15 - 19 6 7 3 20 - 24 1 4 4 25 - 29 0 0 1 30 - 39 1 3 2 40 - 49 0 0 2 50 - 59 0 0 0 1 Hour + 0 0 1 Not Applicable* 0 1 1

* Works at home or salesman 101

Table 4.4

Number of Weekly Shopping Trips by Neighborhood

Number of Weekly Shopping Trips South Central North

1 3 5 6 2 11 10 9 3 9 5 5 4 2 3 3 5 0 4 1 6 1 1 1 7 1 1 0 10 0 1 0 14 0 1 0 Table 4.5 a. Frequency of Use of Information Sources for Food Specials and Sales ( 1 - very often, 7 - never) Average Response/Standard Deviation South Central North Total Sample Newspapers 2.62/2.06 2.84/2.25 2.48/1.97 2.66/2.07 Television 5.59/1.78 5.65/1.74 5.74/1.65 5.66/1.71 Radio 6.52/0.51 6.10/1.37 6.00/1/57 6.20/1.25 Flyers (in mailbox) 1.89/1.22 2.13/1/86 1/56/1/50 1.87/1.56 Store Window Displays 4.74/2.26 5.06/1/98 5.30/1.77 5.04/2.00 Advice from Friends 4.78/1/97 5.80/1.50 5.22/1/83 5.29/1.79 b. Importance of Information Sources Upon Which Store Patronized (1 - very important, 7 - not important at all)

Newspapers 3.48/2.50 4.29/2.41 3.30/2.27 3.72/2.41 Television 5.78/1.76 5.90/1/62 6.07/1/66 5.92/1/66 Radio 6.52/0.75 6.19/1.42 6.19/1.42 6.19/1/80 Flyers (in mailbox) 2.81/2.22 2.48/1/98 1.11/1.99 2.47/2.06 Store Window Display? 4.74/2.18 5.16/1.86 5.22/1.91 5.05/1.97 Advice from Friends 4.93/2.11 5.65/1.91 4.70/2.20 5.12/2.08

c. Importance of Information Sources upon What Foods Purchased Newspapers 2.96/2.33 4.06/2.31 3.56/2/53 3.55/2.40 Television 5.85/1.68 5.68/1.80 6.00/1.78 5.84/1.74 Radio 6.56/0.58 6.13/1.50 6.22/1.53 6.29/1.29 Flyers (in mailbox) 2.37/2.02 2.61/2.04 1.70/1.66 2.25/1.94 Store Window Displays 4.41/2.10 5.10/1.83 5.26/2.09 4.93/2.01 Advice from Friends 4.89/2.08 5.81/1.60 5.19/2.22 5.32/1.98 103 very important or very frequent to seven (7) meaning not important at all or never. Flyers, placed in mailboxes or hung from doorknobs in plastic bags are most important and used most frequently. Flyers consistantly received the lowest scores for all three areas as a source of information for specials, which stores to patronize as well as being an influence upon which foods are purchased. Newspapers also serve as a major source of information for grocery shopping.

Television, radio, store window displays and advice from friends are all considered unimportant and not frequently used by respondents. In fact, the variance of response is of such magnitude that all four sources are utilized to the same low degree. This indicates that individuals definitely make an effort to gather information on grocery items. How­ ever, only those sources which give maximum information at lowest required effort are utilized, these sources being flyers and newspapers. This result is important since it shows the process of selecting a grocery store is based upon a conscious gathering of information and as such, it is a rational process, but the information is limited in quantity.

This result supports the selection of grocery shopping as a subject for a destination choice study. The use of grocery shopping is valid since it is a repeated action based upon information the individual has gathered and it is an activity that most people have to undertake. Also, since the res­ ponses to the questions were reasonably similar across the 104

individuals, variations in peoples' choice sets can be

expected to be attributed to differences in behavior as opposed to differences in information levels.

4.3 Choice Set Composition

The study area contains a large supply of possible destinations for the trip purpose of food shopping. The sample population exhibits a high degree of mobility. Given

these favorable conditions, it is of interest to examine how the residents of this area take advantage of such surround­ ings. This section details the findings of the survey con­ cerning the "composition" of the choice set of the individual for the purpose of food shopping. "Composition" of the individual's choice set is defined as the number and percep­ tual groupings of the stores that the individual actively patronizes.

The composition of the individual's choice set is described in Table 4.6. The average number of stores in a choice set for the entire sample is 3.56 with a standard deviation of 1.51 stores. Unfortunately, no other studies have examined the size of the choice set for food stores, at least none that reported the number of stores for just food shopping, so it is not known if this can be considered typical for a grocery store choice set or not. However, at an intuitive level, and in light of the numerous weekly food 105

shopping trips made by individuals, the size of the choice

set does seem small.

This observation could have a great impact on destina­

tion choice behavior at a theoretical level. Research on destination choice from space-time studies has shown the existence of small choice sets for individuals. It has been suggested that these small choice sets are due to the con­ straints of time and mobility under which the people in these studies exist. In this study, however, small choice sets exist despite the favorable conditions of mobility and high income levels. It is quite likely that another explanation is necessary to account for the existence of small choice sets in this study.

Examination of Table 4.6 sheds some light upon the com­ plexity of the individual's choice set composition. As the size of the individual's choice set increases from one to three stores, the vast majority of the individuals consider each of the stores of their choice set as being distinct.

That is to say, the number of perceptual groups within the choice set is almost equal or is equal to the number of stores. In the case of a two store choice set, 78 percent of the people considered each store as being distinct (i.e. two stores; two perceptual groupings of one store each). The remaining 22 percent consider their two stores as similar

(i.e. two stores, one perceptual group including both stores).

For a three store choice set, the composition of the set is Table 4.6

Size and Number of Perceptual Groups

Within the Individual's Choice Set

Number of Perceptual Groupings 1 2 3 4 5 Row Percent 1 3 - - - - 3.5%

2 3 11 - - - 16..1

3 3 4 15 - - 37. 7 Number of 4 0 9 8 3 - 23.5 Stores in Choice Set 5 0 2 6 1 0 10. 6

6 0 1 0 0 1 2.3

7 0 0 0 1 1 2.3

8 0 0 2 0 0 2.3

9 0 0 0 0 1 1.3 100.0% 107

is similar to the two store pattern. Almost 70 percent of the people with a three store choice set consider each store as being distinct. Only 18 percent have a pair of stores which they consider similar while 14 percent consider all their stores the same.

This tendency to perceive all the stores as distinct breaks down as the individual's choice set contains four or more stores. In the case of a four store choice set, a full 85 percent of these individuals have stores they per­ ceive as similar in their choice sets. The remaining 15 percent perceive each of the four stores as being distinct while no one considered all the stores in their choice set as being the same.

It seems apparent that the purpose of increasing the size of the choice set for the individual from one to three stores is mainly one of achieving a certain level of satis­ faction. One or two stores are not sufficient for the indi­ vidual to satisfy his demand so he goes to other stores that he considers different from the first two to meet his demand.

In choice sets of four or more, increasing the size of the choice sets allows for comparison shopping. This conclusion is drawn from the observation that only in the larger choice sets do similar, i.e. competitive stores, appear in the majority of those choice sets.

This observation has serious consequences for spatial choice theory. Even with a choice set of three stores, the 108

majority of the people are in a "no choice" situation. If

the three stores are considered as distinct by the indivi­

dual, then the destination choice problem for the individual

reduces to choosing that store which fulfills his needs at

the time he is deciding to go, or to not go at all. Thus the choice problem is not to go to that store or stores that

best meet his demands but to go to the stores that are satis­

factory. The significance of this situation is that choice

is made without trading off store attitudes. The choice set

for these individuals has evolved so that the choice problem

is very simple. Given the specific needs of the individual at a given time, he selects that unique store in his choice set that meets those needs. No other store will do.

While the size of the individual’s choice set is impor­ tant, the individual's perceptual grouping of the stores in his choice set which provides information as to the way people evaluate the stores within their choice set is also important. Most spatial choice theories assume the indivi­ dual is capable of evaluating several stores on a set of attributes in order to arrive at some sort of preference ranking of the stores which they wish to patronize (Lancaster,

1966; Rushton, 1969, 1971). This evaluation process is assumed to be similar across similar individuals to make possible the postulation of a generalized choice model that will replicate individual choice behavior. Thus a serious issue in choice theory is whether or not there is agreement 109

among the individuals as to how the stores available to them

are perceived and if the method of discriminating between

stores is the same.

In order to investigate this issue of store perception,

an Alscal multidimensional analysis is performed to obtain

a measure of overall fit of the perceptions of stores common

to the choice sets of a group of individuals. This agreement

will be measured by observing the stress value for a MDS

store configuration solution for a common group of stores

for each of the three neighborhoods in the study area. A

low stress value suggests that store perceptual grouping by many individuals are similar in that the similarities - dissimilarities can be represented by one store configuration

in Euclidean space. A high stress value would indicate little agreement or that a different method of discriminating between stores was employed by the sample population.

Due to the nature of the similarities matrix data, it was found that only a small group of stores are present in a majority of the respondents' choice sets. It is therefore necessary to check for agreement among members of the indi­ vidual neighborhoods. "Neighborhood choice sets" are constructed by including any store which shows up three or more times in neighborhood responses.

Much of the data in the neighborhood analysis consists of comparisons between stores not in an individual's choice set with those which are (unknown vs. known), and between 110

stores where both are not in the individual's choice set

(unknown vs. unknown). This missing data is not usual for

Alscal analysis. Alscal is, however, capable of accepting missing entries. It is important to emphasize that the analysis is being performed for the purposes of the investi­ gation of group agreement as opposed to careful scrutiny of destination configurations. Thus the analysis is valid for

the study, but the problems of the missing data, as detailed

in Chapter Three, may cause problems in the interpretation of the store configurations based upon group perceptions.

Briefly, to repeat the anticipated problems of the analysis, Alscal tends to place stores not directly compared in the data far apart in configurations. This normally means that the stores are considered dissimilar in nature.

However, if several stores are not directly compared, the routine will place them far apart in configuration regardless of what actual perceptions would have been. This problem will show up with high stress values and extreme difficulty in interpreting the dimensions of the Euclidean space con­ figuration of the stores.

Having noted the possible problems, a brief discourse of the findings of the MDS analysis for each separate neigh­ borhood is presented.

The South Neighborhood

The South neighborhood is a very compact sub-area of the study area. It is generally of a lower level of income Ill

compared to the other two areas and consists mainly of young

married couples with small children and older individuals

and couples of or near retirement age. From the total number

of stores in the area, only eight are found to be patronized

by enough individuals in the sample to be included in the

analysis. The criteria for selection of stores is that the

store must be included in three or more of the residents’ choice sets.

The stores included in the analysis for the South neighborhood are listed in Table 4.7. The list includes the

three major chain stores in the area as well as one large store in the Graceland Shopping Center, two or three miles west of the study area. Also included are two convenience stores and a small independent food store, all of which are in the South neighborhood area. The last store on the list,

Carfagnas, is a speciality butcher shop which features im­ ported Italian food products and enjoys quite a reputation for quality in the entire study area.

The data on which this analysis is based is a transfor­ mation of the similarities matrix data. Those stores in the neighborhood choice set (see Table 4.7) are checked with those in the individual's choice set. Those store pairs in the individual's choice set will have an evaluation as to whether the pair are similar or dissimilar. These evaluations are then placed in an aggregate similarities matrix for the neighborhood. For the South neighborhood the aggregate Table 4.7

Stores Included in South Neighborhood

Alscal Survey Character Number Store Address Description

A 01 Kroger Morse, Karl Major chain store B 11 Big Bear Karl, 161 Major chain store C 12 Big Bear Morse Center Major chain store D 13 Big Bear Graceland Major chain store E 31 Food Center Morse, Maize Small independent grocery store F 42 Carfagnas Karl, 161 Butcher shop and Italian imports G 51 Lawson's Karl, Shanley Convenience store H 54 Save Mart Morse Convenience store 113 similarities matrix is an eight by eight matrix, the rows

and columns of which correspond to the list of stores in

Table 4.7.

An eight by eight similarities matrix exists for all

but one of the 27 respondents within the South neighborhood.

The respondent not included had only one store in his choice

set and did not compare his store to any others.

Included in the original similarities matrices is a list

of the characters in the individual uses to discriminate

between stores in his choice set. A comprehensive list of

the use of these characteristics for the South neighborhood

is presented in Table 4.8. The first point of interest is

that only seven characteristics are used by the entire neigh­

borhood sample as well as the entire sample to discriminate

between stores. Of even more interest is that a majority of

individuals use only two or three store characteristics to

discriminate between stores in their own choice set. The

small number of characteristics used is not surprising due to

the small number of stores in an individual's choice set.

The surprising aspect is that the specific attributes used by

individuals vary. This is to say that while several indivi­

duals may use two characteristics of grocery stores to dis­

criminate between stores in their choice set, the specific pair of characteristics may be any two of the seven listed

in Table 4.8. Table 4.6

Attributes used in Store Comparisons in South Neighborhood Respondent Varietyy .pricePrice yQuality u a n t y unecicoutCheckout time Service Hours Distance (Size2______(Convenience) 1 X X X 2 X X X 3 X 4 XX 5 XX 6 X X 7 X X B XX 9 X X 10 X X 11 X 12 X 13 X X 14 X 15 X 16 XX 17 XX 18 X 19 X 20 X X 21 X X 22 X 23 X X 24 X 25* 26 X X 27 X X Total 14 11 8 7 5 « only one store in choice set Number of Attributes i 3 2 1 ( Number of respondents 3 14 9 115

This phenomenon may cause severe problems vis-a-vis the MDS analysis at the neighborhood level. The purpose of the analysis is to test the degree of agreement among indi­ viduals in regards to their perceptions of stores. If individuals discriminate between stores using different cri­ teria, then a given store pair may be judged similar or dissimilar depending upon which characteristic is used by the individual. This may cause problems in the interpreta­ tion of the dimensions of the store configurations.

Referring to Table 4.8, it can be seen that size and perhaps prices are the most used characteristics, but quality, convenience and service are also frequently used. Thus only a small subset of the seven store characteristics listed are used by any one individual for the purposes of discriminating between options in his choice set.

This observation also has serious consequences for choice theory. The two or three characteristics by which individuals discriminate between stores in their choice set are very important in regards to decision-making. These salient attributes are not only viewed as important by the individual, but one can assume that the individual per­ ceives varying levels of these attributes for the stores in his choice set or the attributes could not be used to dis­ criminate between stores. For example, cleanliness of the store is often listed as an important attribute for a store to possess. However, if all stores are perceived as clean, 116

cleanliness can not be used to discriminate between stores.

On the other hand, if quality is perceived as varying from

store to store, then quality can be used as a salient

attribute. If the salient attributes used vary from indivi­

dual to individual, then the problem of constructing an

aggregate choice model becomes difficult. On the one hand,

all seven characteristics must be included in the model or

the model will exclude important variables. On the other

hand, including all seven variables makes the models more

complex than the actual decision-making process of the

individual which is the raison d'etre of the micro approach.

Having outlined possible problems with the analysis, the

results of the MDS analysis of neighborhood perceptions of

stores is presented.

The first task in interpreting the results of the MDS

routine is to select the proper dimensionality of the store configuration solution. A plot of the stress values for the

South neighborhood (as well as the other two neighborhoods)

is found in Figure 4.4. Ideally, the existence of a kink

in the stress curve would point out the proper dimensional solution. In the case of the South neighborhood, this kink does indeed exist at the three dimension solution.

The second task in interpreting the results of the MDS routine is to identify a meaning for each dimension of the solution. The first dimension corresponds directly with the size of the establishment. The size of the store, or .300 Central Neighborhood

North Neighborhood .200 South Neighborhood Stress

.100

3 4 5 Dimension of Solution

# Notice kink in plot

Figure 4.4 Stress Plots 118 variety of selection, is the attribute most used in store differentiation by individuals and therefore provides in­ tuitive evidence as to the first dimension being properly identified. Further, stores A, B, and C are large stores while stores G and H are small stores. These two groups of stores correspond to the left and right extremes, respectively of dimension one. Stores E and F, in the middle of the axis are medium size stores. Finally, store

D represents a large chain store slightly smaller in size than A, B, or C.

The second and third dimensions did not at first seem to have any interpretation except for, possibly, conven­ ience where convenience is taken to mean distance for the area from the store to the respondent's house. Upon further examination however, dimension two is found to place the stores A and H from west to east as labeled on Figure 4.5,

4.6, 4.7 in the store's actual spatial orientation in the study area. Dimension three does the same but on a north to south orientation. The phenomena of the spatial inter­ pretation of dimensions two and three arise from the missing entries problem discussed previously, in Chapter Three.

This anticipated problem seems, therefore, to present a large problem in the analysis and thus no strong conclusions can be made from the analysis pertaining to the agreement of individuals in their perceptions of stores based on this data. This problem could be corrected in future H

D

E

Dimension One Tinensio.'. Two Tinensio.'.

Figure 4.5 Store Configuration South Neighborhood B

F

Dimension One

G

a> a> M .£ r_i £ •H0

£to 01 n

Figure 4.6 Store Configuration South Neighborhood 120 B

F

Dimension Two G

A

C Dimension Dimension Three

Figure 4.7 Store Configuration South Neighborhood 122

destination choice studies by providing the respondent

with a common set of stores in the neighborhood to evaluate

by a modified “pick k out of n minus one" task. However,

the reason why this was not done in the survey conducted

for this study remains in that the respondent may have to

evaluate stores that are outside his choice set, i.e.

stores which he or she knows little or nothing about.

The reasonably low stress value of .180 for the three

dimensional solution, despite the previously mentioned

problems, does lend some support to the premise that people

do agree as to the perception of various stores as derived

by this technique. However, differences in the specific

attributes used to differentiate between stores by indivi­

duals may also lead to some disagreements in perceptions

which contribute to a high stress value.

The Central Neighborhood

The Central neighborhood, so named because it is in

the center of the study area between Route 161 and Morse

Road, has thirty-one respondents to the survey. Of these

respondents, three had to be removed from the analysis

because they had only one store in their choice set and could not by definition provide any information as to how

they discriminate between stores in their choice set. 123

Seven stores are found to be included in three or more of the survey respondents in the Central neighborhood. Of the seven stores, three are large major chain stores loca­ ted within the Central neighborhood study area. The second trio of stores are small independent stores located in or on the border of the Central neighborhood. The last store,

Carfagna's, a butcher shop and imported Italian foods store, is also included within the Central area. While all residents of the area are capable of traveling greater distances, they show a distinct preference for nearby stores.

The incidence of usage of the seven identified store characteristics used by individuals to discriminate be­ tween stores in their choice set is presented in Table

4.10. As in the South neighborhood, a large percentage

(61 percent) uses either two or three attributes to dis­ tinguish between stores in their choice set. The reason for the use of so few salient attributes by the individual is due to the small size of their choice sets. However, the specific attributes used by individuals does vary between individuals as in the South neighborhood. As shown in

Table 4.10, the vast majority of respondents use size, or variety, to distinguish between stores. This is not surprising since store size definitely does vary from store to store. However, the attributes of price, convenience and service are also frequently used. Distance, quality and Table 4.9

Stores Included in Central Neighborhood

Alscal Survey Character Number Store Address Description

A 01 Krogers Morse, Karl Major chain store

B 11 Big Bear Karl, 161 Major chain store

C 12 Big Bear Morse Center Major chain store

D 31 Food Center Morse, Maize Small independent grocery store

E 32 IGA Forest Park Yellowwood Small independent grocery store

F 38 Gateway 161, Karl Small independent grocery store

G 42 Carfagna1s 161, Karl Butcher shop and Italian imports 125

store hours are also used by a few respondents. The same

problem for the MDS analysis of the Central neighborhood

exists as in the South neighborhood. Only two or three

attributes are used by individuals, those few attributes

varying for each individual. Because of this lack of con­

sistency of judging criteria, it can be expected that store

configurations produced by MDS analysis will exhibit a high

stress value and dimensions which will not have clear inter­

pretations.

Referring to Figure 4.4, a kink exists at the three

dimensional solution for the Central neighborhood. The

store configurations are depicted in Figures 4.8, 4.9, and

4.10. Dimension one, once again has a very obvious inter­

pretation as store size. Stores A, B, and C, being large

chain stores, are grouped together on one end of dimension

one. Stores D, E, F, and G, small stores, are grouped

together at the other end. As for dimensions two and three,

the most important feature of the store configurations is to

notice that stores D, E, F, and G are spread out in Euclidean

three dimensional space. The reason for this spread is not due to the usual case of the four stores being considered dissimilar. Rather, it is because the four stores, which are

small in size, are not directly compared by many individuals

in the data. Thus dimensions two and three are meaningless

in terms of this analysis. The stress value is high (.267)

indicating, quite correctly, that the MDS model does not Table 4.10 Attributes used in Store Comparisons in Central Neighborhood

Respondent Varietv Price Quality Checkout time Service Hours Distance (^i2 e; (Convenience) ______28 X X 29 X X 30 X X 31 X 32 X X 33 X X 34* 35 X X 36 X 37 X 3£ X 39 X 40 X 41 X 42 X 43 XX 44 X X 45 X 46 X X 47* 48 X 49 XX 50 X X 51 X X 52 X 53 X X X 54 X X 55 X X 56 X 57 X X X Sfi* Total 23 8 2 7 6

Number of Attributes used # Number of respondents 3 2 1 0 one store in 4 15 9 3 choice set G

D

Diraension One “B-

o

c o oCO£ a •HQ

Figure 4.8 Store Configuration Central Neighborhood i^iiienslon One

© U© fTi CO cCO © E

Figure 4.9

Store Configuration Central Neighborhood 128 Dimension Two

®a) f-i x:CH D a o

01 £J n Figure 4.10

Store Configuration Central Neighborhood 130

provide a good representation of store perceptions. The main

reasons for the bad fit are the missing data problem and the

use of a few but varying attributes by individuals in the

evaluation of the similarity or dissimilarity of stores.

However, it is worth noting that despite the data problems,

in both the South and Central store configurations, size,

the most commonly used attribute, is easily identified as

the interpretation for dimension one.

The North Neighborhood

The northernmost subsection of the study area consists of twenty-seven survey respondents. Eight stores are found to be included in the choice sets of three or more indivi­ duals. Included in the eight are three major chain stores.

All three stores are very close to the North neighborhood.

Two small independent grocery stores are also included in the list as well as two small convenience stores and

Carfagna's, the butchershop and imported Italian foodstuffs store. A list of these stores is found in Table 4.11.

Although the people in the area are capable of going further distances, they tend to patronize nearby stores.

The occurrence of use of store characteristics by individuals is detailed in Table 4.12. The same pattern of usage occurs as in the two other neighborhoods. The majority of the respondents (85 percent) use only two or Table 4.11

Stores Included in North Neighborhood

Alscal Survey Character Number Store Address Description

A 02 Krogers 161, Cleveland Major chain store

B 05 Krogers-Savon Schrock, Rt. 3 Major chain with department store C 11 Big Bear Karl, 161 Major chain store

D 34 IGA Renzetti's Sharon Woods, 161 Small independent grocery store E 38 Gateway 161, Karl Small independent grocery store F 42 Carfagna's 161, Karl Butcher Shop and Italian Imports G 53 Lawson's Schrock Convenience store

H 73 Convenient Suderland Convenience Store Food Mart Table 4.12

Attributes used In Store Comparisons In North Neighborhood

Respondent Variety Price Quality Checkout time Service Hours Distance (Size; (Convenience)

59 X X 60 XX 61 X X 62 X 63 X X 64 X 65 X X X 66 X X ♦ 67 66 X X X 69 X 70 X X 71 X X 72 X X 73 X X 74 X X 75 X X X 76 X X 77 X X 76 X X 79 X X X 60 X X 61 X X 82 X X X 83 X X 84 X X 85 X XX Total 21 9 5 7 7 2 4 « one store choice set, no discriminating attributes Humber attributes uBed in store discrimination Humber of 3 2 1 0 Respondents 5 18 3 1 133 three dimensions to discriminate between stores in their choice set. Twenty-one of the twenty-six individuals in­ cluded in the analysis use size as a discriminating charac­ teristic for stores. This stands to reason since there is a large variance in size among the stores in the store list for this neighborhood. The store attributes of price, quality, convenience and service are used less often than size but by different individuals. Distance and store hours also occur as discriminating factors for a few of the res­ pondents. The lack of a consistent overall set of dis­ criminating attributes for each individual in the neigh­ borhood and the small size of their choice sets suggest MDS analysis results should be similar to the other two neigh­ borhoods previously discussed.

Examination of Figure 4.4 shows no true kink exists for any dimensional solution for the North neighborhood.

For reasons of consistency with the analysis of the other two neighborhoods, the three dimensional solution is chosen to match the other two neighborhoods.

The store configurations are presented in Figures

4.11, 4.12, and 4.13. The stress value for the solution is .260, which is very high in terms of the model's inabil­ ity to replicate the patterns of store dissimilarity present in the data. In spite of this, dimension one clearly shows up as a size dimension. Stores A, B, and C are at one end H

A C

MmpnRlnn flrtp

E G

D o £ B o _ *H F m a> e Q

Figure 4.11

Store Configuration North Neighborhood Dimension One

H

B

Figure 4.12

Store Configuration North Neighborhood 135 E D

JE ^ Dimension Two

Figure 4.13 Store Configuration North Neighborhood 136 137 of dimension one and are very large stores. Stores D,

E, and G are of moderate size and are situated in the middle of the dimension one axis and stores F and H are small stores, situated at the other end of the axis. Again, dimensions two and three elude interpretation. The MDS routine, Alscal, tends to spread stores not directly com­ pared out in three dimension Euclidean space.

Summary on Neighborhood Analysis

Three main observations can be made from the results of the analysis of the individual's choice sets by neigh­ borhoods. First, most individuals discriminate between stores within their choice sets on the basis of only two or three attributes. The specific attributes used, however, vary for each individual over seven identified attributes of food stores. These attributes are: selection (size of store), price (expensive versus inexpensive), quality, speed of checkout (convenience), service, store hours and distance. Selection (size of store) is most often used for store discrimination. Price, quality, convenience and service are, taken together, used roughly half as often as selection. Hours and distance are used to dis­ criminate between stores by less than 10 percent of the sample. This last observation is not surprising since the stores in most individual's choice sets are usually located 138

close to the respondent's home. Thus for most people, all

their stores are close and henc6 distance is not a dis­

criminating factor. Similarly, most people consider most

store hours to be the same. Only 3.5 percent of the sample

find 24 hour stores to be a worthwhile feature to consider

in their perceptions of stores within their choice set.

This observation has important consequences for destin­

ation choice modeling. First, it suggests that destination

choice is indeed a simple process. Individuals discriminate

between stores they actively consider using only two or

three attributes. For a choice model to represent a larger

group of individuals, all seven identified attributes must

be included. Since only two or three attributes are con­

sidered in discriminating between stores in the individual's

choice set, the concept of tradeoffs between the few attri­

butes used by each individual and the total list of attri­

butes used by the entire sample can not be considered valid.

The second observation is that variety (size) shows up

in the Alscal stimuli configurations of the stores consid­ ered in each neighborhood. Except for major chain stores,

knowledge of smaller stores is limited to only a few res­ pondents within each neighborhood. Distance, while not considered an important factor, does appear as being a con­ straint upon the individual in terms of the information

levels of smaller stores. However, satisfaction levels for 139 stores patronized are high. This observation is based upon a series of questions dealing with the respondent’s evalua­ tion of stores in their choice sets for each of the seven attributes found important. Stores in the individual's choice set are rated by the individual on a scale of one to seven for each of the seven attributes. A rating of one consistantly means the best rating, seven the worst, and four means average. Since over 80 stores are identified by the subjects, satisfaction levels for only those stores used in the neighborhood analysis are presented. Table 4.13 shows that each store receives favorable scores in many of the attributes. Even more surprising than the high satis­ faction levels is the amount of agreement among individuals with the ratings. The standard deviations of the responses are all near or below 1.0. This indicates that the percep­ tion of stores by individuals again is not unique to the individual but agreement is possible across subjects.

The only exception to a high degree of satisfaction with stores in the area is the perceived high prices and low quality of the convenience stores (the bottom four rows in Table 4.13).

This second observation has important consequences for destination choice theory. For an area such as the study area, with many stores and a high satisfaction level for these stores, there is little incentive for the individual Table 4.13 Satisfaction Levels for Selected Stores Variety Cleanliness Hours Checkout time Distance Quality Prices i 01 Krogers Morse and Karl 1. 51 1. 92 1.17 2.63 2.07 3.29 3. 37 02 Krogers 161 & Cleveland 1.22 2.22 1.11 1.56 1.33 3.11 2.56 05 Krogers/Savon Schrock & Rte. 3 1.00 2.50 1. 00 2.00 1. 25 3.00 3. 50 11 Big Bear Karl & 161 1.10 1. 62 1. 58 1.93 1.60 3. 44 2.44 12 Big Bear Morse Center 1.26 2.84 1.89 2. 32 1.68 3.89 2.84 13 Big Bear Graceland 1.20 2.20 . 140 2.20 1. 20 3.60 3.00 31 Food Center Morse & Maize 2.84 1. 25 2.16 1.70 1. 23 3.06 1.45 32 IGA Forrest Park Yellowwood 3.47 1.12 2. 53 2.81 1.65 3.82 2.12 34 IGA Renzetti's Sharonwoods & 161 3.80 1. 07 1. 67 1. 73 1. 64 3. 80 2.20 38 Gateway 161 & Karl 2. 91 1. 50 1. 50 2. 09 2. 09 2. 73 1.73 42 Carfagna1s Karl & 161 3.10 1. 90 1. 90 1.05 1.90 3.50 1.55 51 Lawson1s Karl & Shanley 6.66 1.67 1. 00 5. 00 4.00 5.00 1. 33 53 Lawson's Schrock Road 5.50 1. 25 1.50 3.67 1.50 6.00 4.00 54 SaveMart Morse 5. 33 1.00 1.00 5. 00 3.67 5.67 3. 00 73 Convenient Food Sunderland 5.75 1. 00 1.00 6.50 3.00 5.25 1. 00 Mart 141 to gather information on stores further away from his resi­ dence than those he currently patronizes. This means the perceived benefits from going to a store located further away is most likely to be small. Therefore, changes in behavioral patterns, particularly in switching to a new store will most likely be resisted except by those residing close to a new store location. Tradeoff analysis to exa­ mine how to draw more people to a new store may be mislead­ ing. Even if the new store is "better" the amount of improvement will not be perceived as being worthwhile to change one's purchasing habits. On the other hand, if sat­ isfaction levels were low, a drastic change in shopping behavior could be expected with the introduction of a new store that most people would find acceptable.

Further, if satisfaction levels were low and choice sets size small, this would indicate that individuals may indeed be constrained by outside factors. In this case, the effects of mobility, time and other factors may prove more of an influence of individual behavior. Studies relating choice set size and satisfaction levels could provide an excellent measure of an area's ability to provide the ser­ vices that the residents desire.

The existence of small choice sets, two or three attributes being used to discriminate between stores in the choice set and high satisfaction levels for those stores taken together have important consequences for choice 142

theory. In essense, the consumer in the study area is

confronted with a large number of alternatives, all of

which prove highly satisfactory to their customers. Indi­

viduals then select only a few stores from this large number

to form his choice set. The stores within the small choice

set are evaluated by the individual on the basis of two or

three alternatives. This scenario is in direct opposition

to a compensatory, utility maximization type of choice model.

The use of such a model would require the gathering of a

large amount of information for each of a great number of

stores. In order to arrive at a "best" choice of a store,

tradeoffs between all relevant attributes for all stores must be considered. The evidence of the survey suggests

that the individual does not use all seven attributes. The

task of processing all this information would be beyond the

limits of almost anyone. Since many of the stores have equally high satisfaction levels, the utility derived from patronizing these stores would be about equal. Thus util­

ity maximization theory would suggest that the probability of choosing any of these stores would be about the same.

This would mean that over the course of time the consumer would patronize many of these stores and not patronize any one store exclusively since that could occur only if the utility of patronizing that store is much greater than any of the other stores. This feature runs directly against observed grocery store choice behavior. It can be 143 strongly concluded then, that compensatory (tradeoff between attributes) utility maximization models are clearly not applicable in this case.

On the other hand, a different class of choice models seems ideally suited for this scenario. This class, in short, being non-compensatory, non-optimal choice theory.

The consumer in this area is confronted by a large number of stores that are nearly indistinguishable from each other.

This is due to the high level of competition between stores; each store copies the other's operation and offers nearly identical goods. Thus the penalty of selecting a store that is not the "best" store for fulfilling the food demands of the consumer is not significant. That is, almost any store will be satisfactory. Therefore, to simplify the choice problem, the individual may utilize a series of screening criteria, for example, store size and distance to find a few stores in the area that are satisfactory to him. Stores either meet the screening criteria or do not. There is no tradeoff between those attributes that are considered important. The screening criteria are adjusted, e.g. distance is decreased, until a small number of possible destinations is formed. From this small group of stores, a choice set is then developed.

Couching the problem of destination choice in non- optimal choice theory changes the questions of interest in research. Instead of extensive tradeoff analysis, the 144 investigation of which attributes are used to screen the possible choices and the necessary levels of satisfaction for these attributes a store must possess to be included in the choice set or not becomes of interest.

Another topic of interest for research is to determine what relationships exist between the number of alternatives and the type of choice rule that may be applied. To accept the applicability of non-compensatory, non-optimal choice theory over compensatory utility maximization theory for grocery store selection redirects the entire direction of research in this area.

The third major observation is that despite information level problems in the data for smaller stores, the stress values obtained for each neighborhood for the stimuli configurations for the stores are low enough to suggest a general consensus among the sample as to their perceptions of stores. However, the missing data problem makes any strong conclusion impossible. In a future study, it would be of interest to do a perceptual grouping analysis when a common set of stores is presented to each individual. On the other hand, the observed small choice sets indicate that information levels on other stores outside the choice set will most likely be guesses by the individual and the re­ sults may therefore be misleading. 145

The overall performance of Alscal on the data given

the data set's problems, is highly satisfactory. In all

cases the size dimension clearly existed in each of the

store configurations. The addition of the recording of

discriminating factors used by respondents in discrimi­

nating between stores in their choice set proved very

helpful, not only in identifying interpretations of the

MDS dimensions but also pointing out possible problems in

the data set in that different characteristics are used by

different individuals to discriminate between stores.

4.4 Respondent Attitude Profile

Essential to any modeling scheme is the provision of

some criteria by which to aggregate individual behavior for modeling purposes. There seem to be two major approaches to aggregation schemes. The first is to perform a market

"segmentation" of the sample into various socio-economic classes deemed appropriate a priori by the investigator.

More recently, a second approach has come into existence where the individuals are clustered into several groups based upon characteristic responses to the survey device used or other behavioral criteria. This second approach reveals groups which have similar response patterns. These groups can then be examined for correlations between groups and various situational variables. The second approach is 146

intuitively more appealing since it groups individuals by

common perceptions and then examines changes in situational

variables to describe each group rather than artificially

dividing the sample into groups and then looking for

behavioral differences.

Several clustering programs exist for separating a

sample into groups. The algorithm used in this study is

the "Cluster" routine from the Statistical Analysis System

(SAS) computer package.

Finding a uniform criterion to group individuals by choice set characteristics is nearly impossible due to the very few stores that would be included in all the respon­ dents' choice sets. Thus, to identify groups of similarly disposed individuals, a conjoint analysis section is

included in the survey.

Seven attributes of grocery stores were found to be the attributes of grocery stores that the respondents used to discriminate between stores in their choice sets. To determine the relative importance of these attributes as perceived by the individual, a conjoint analysis of discrete levels of the attributes is undertaken. The attributes with their corresponding levels are presented in Table 4.14.

There are 21 (n x (n-l)/2) possible pairings of seven at­ tributes for conjoint analysis. However, it is not necessary for all twenty-one pairings to be made. For the Table 4.14

List of Attributes and Levels

Distance to store (one way)

0 - 2 miles 3 - 4 miles 5 - 6 miles 7-10 miles

Variety of goods

Major grocery store, with deli and bakery Large grocery store, with deli Small grocery stores

Pr ices

Above average Average Below average

Quality of meats and vegetables

Always fresh Usually fresh Fresh only on certain days

Cleanliness

Clean wide aisles Some cartons on aisles (some dust) Narrow aisles (some dirt)

Checkout time

Lines 0 to 1 person long Lines 2 to 4 people long, express lane Lines 5 to 7 people long, no express lane

Store hours

24 hours 7am - 12 am 8am - 8 pm (not open Sundays) 148

survey, a pairing scheme is designed so that the selected

attribute tradeoff pairings are reasonable for real life

situations but also allow for comparisons between almost

any pair of attributes to be Accomplished either directly

or through one intermediate attribute. A visual represen­

tation of the pairings of attributes is diagramed in

Figure 4.14. Eleven out of the twenty-one possible pair­

ings are to be considered by the respondent. The size of

the tradeoff matrices are either 3 x 3 or 3x4. Thus the

individual expresses a rank order preference for a pair of

attribute levels. Depending upon the pattern of the re­

sponses, the relative importance of each level of the

seven attributes can be analyzed by the use of Johnson's

monotone regression routine (MONORE).

The responses to the tradeoff section of the survey

by the individual are then used to identify similar be­

havioral groups based upon the responses to all seven

attributes. A plot of the number of distances (a measure

of dissimilar responses within a cluster) can be found in

Figure 4.15. A large distance value is undesirable since

it indicates lost information while a small number of

groups is desirable for segmentation purposes. Thus a

tradeoff between accuracy and number of groups exists. The most favorable condition exists if an identifiable kink

exists in the error plot. This condition would suggest that

a large amoung of information is lost at a clustering Distance

Figure 4.14 Attribute Pairing Diagram fflOSPctOH'b 2000 1800 1900 1400 1600 1700 1000 1200 1300 1500 1100 800 200 300 400 500 900 600 700 100 0 3 5 7 91 1 2 3 4 15 14 13 121110 9 8 7 6 5 4 3 2 ITotegroups seven curveat in kink Plot ofDistances Plot Humber of Clustersof Humber Figure 4.15 Figure 150 151

solution of fewer groups than the number of groups

where the kink exists, the condition being analogous to the

MDS stress method of selecting the proper dimensional

solution of the stimuli configuration. As can be seen in

Figure 4.11, an identifiable kink is clearly visible at the

seven group solution.

Once the appropriate groups are identified from the

Cluster analysis, Johnson MONORE routine is employed to analyze the response patterns for each group. The results are summarized in Table 4.15.

Each group contains several respondents except for group seven which consists of one respondent. This single response was not incorporated into another group until the three group solution. The response is therefore at vari­ ance with almost the entire sample. For this reason this response is considered an outlier and is not consi­ dered in this analysis.

Two more of the eighty-five respondents are eliminated from the analysis since they could not complete the tradeoff section of the survey.

A brief explanation of the meaning of Table 4.15 is in order. An entry in the matrix describes the relative desirability of the attribute level i for the group j in the matrix. A high positive value indicates a very high desir­ ability for that level of the attribute. A low negative 152

Table 4.15 Tradeoff Analysis

Groups Phi .236 .190 .199 .193 .162 .123 Tau .567 .621 .610 .602 .662 .620 N = 8 16 6 9 20 23

Prices High -.181 -.204 -.283 -.222 -.127 -.062 Avg. .049 .082 .134 .121 .131 .067 Low .262 .333 .327 .258 . 372 .225

Quality High .398 .409 .328 .369 .295 .174 Avg* .035 .064 .085 .092 . 085 .080 Low -.346 -.347 -.264 -.340 -.167 -.027

Cleanliness High .328 .349 . 392 . 346 .317 .177 Avg • .072 .089 .076 .120 .145 .092 Low -.305 -.306 -.319 -.367 -.255 -.040

Distance 0- 2 . 312 .011 .092 .196 .245 .211 3- 4 -.101 .084 .020 -.068 -.017 .266 5- 6 -.217 -.099 -.108 -.089 -.041 -.112 7-10 -.380 -.429 -.446 -.607 -.544 -.880

Variety Large .256 .286 -.207 .035 . 231 .094 Medium .088 .077 .123 .030 .137 . 061 Small -.281 -.184 .271 .177 -.147 .021

Hours 24 hr .069 .120 -.058 -.013 .207 .083 7-12 .121 .123 .150 .109 .079 .103 No Sunday 8-8 -.035 -.084 .064 .085 -.080 .028

Checkout 0-1 . 315 .298 .285 .210 .273 .197 Lines 2-4 .079 .096 .127 .060 .109 .090 5-7 -.272 -.242 -.268 -.126 -.178 -.059 153 value, e.g. -0.5, indicates that the specific level of the attribute which represents the row is highly undesirable.

Values near zero (0.0) indicate an intermediate preference.

The MONORE model is calibrated for each group. Comparisons of values up and down individual columns are entirely valid.

Interpretation of values across rows however, may be mis­ leading. Such attempts are analogous to comparisons of regression coefficients between two separate models. Since monotone regression assumes ordinal level data, true error measurements of the coefficients are not possible. However across the row comparisons are possible since they indicate changes in the model as different behavioral groups are fitted to the monotone regression equation.

Notice should be given to the Phi and Tau values for each group. The Phi values, ranging from .123 to .236, are the "badness of fit" criteria by which the algorithm calibrates the monotone regression model. A second test of the model's overall accuracy can be accomplished by multiplying the two appropriate calculated utility values.

For each respondent, the calculated and observed utility values are then compared via Kendall's Tau (Siegel, 1956).

A Tau value of 1.0 indicates perfect fit. Lesser values can be from a variety of sources. Misspecification of the preference function is one such source, intransitive or illogical rankings are another or there may be a weakness 154

in the estimated algorithm such as a local optimal solution

being found by the program (Knight and Menchick, 1974). All

responses were checked for internal consistencies as well as

careful observation of the responses as they were being

made by the respondent. It appeared that the individuals

were indeed trading off between attributes.

The Tau values ranged from .567 to .662. Two factors,

however, may account for the low but by no means insigni- 2 fleant Tau values. Being somewhat analogous to R values

(from ordinary least squares regression) some 60 percent of the variance in individual's responses is accounted for

by each model for the six groups. The first reason for the

low Tau values is related to the observed phenomena that

individuals discriminate between stores on the basis of two or three specific attributes per individual from a set of seven identified attributes. While the small number of attributes used may be due to the small size of the observed choice sets, the variety of attributes used in the aggregate

indicate that individuals place varied degrees of impor­ tance to each attribute.

A second problem further complicating the importance weighting problem, is that the importance of a particular store attribute may depend upon the satisfaction levels of that attribute for the actual stores patronized. For example, cleanliness may be an important feature of a store for the individual. However, if the individual is satisfied with the level of cleanliness, i.e. all stores in the area are clean,, then cleanliness is not a suitable attribute to distinguish between stores. If the individual does not use cleanliness as a means of evaluating stores, then it may be merely an artificial question for the individual to compare varying levels of cleanliness to other variable levels. As the tradeoff section of the survey was con­ ducted in the presence of the interviewer, often the phrase,

"if the store is dirty I just won't go in," was heard.

This indicates that there is indeed no actual tradeoff be­ tween cleanliness and other attributes. Thus actual beha­ vior may be at variance with the model, i.e. model mis- specification. Distance is another attribute which posed problems. Often a particular distance, usually the 7 to 10 mile level was not considered a viable alternative for the individual. The respondent would often state, "I would not go that far." Considering the large selection of stores within the relatively small study area, such a statement makes perfect sense.

As stated, these problems may collectively be a poten­ tial source of the observed Tau values for each group.

Still, the Tau values are of such magnitude that interpre­ tation of the values in Table 4.13 is valuable.

In order to identify trends in shopping behavior, a socio-economic profile of each group with related food shopping behavior is constructed. These profiles are 156 summarized in Table 4.16. These profiles represent people who place similar values on the various levels of store attributes.

Group One

This group is generally younger than sample average,

25 percent single, with less formal education and a smaller family size. The average income of this group is the lowest of all six groups. This group also has the smallest number of stores in their choice sets and are tied with group three for making the fewest number of shopping trips per week.

In terms of this group's structure of preference the first item of interest is that in spite of the low level.of income, price levels are a low priority. This may be due to the fact that the low family size means the individual household's total food bill is low. Cleanliness and store hours are similar in importance to several of the other groups. There is a desire to avoid small stores. Perhaps the most interest concerning this group is their attitude towards distance. They place the highest premium of any group on traveling short distances. There is a significant desire on the part of this group to not travel 3 - 4 or

5 - 6 miles to a store, again the highest resistance of any group. However, the longest distance interval, 7-10 miles, is not considered as big of an obstacle to this group 157

Table 4.16 Situational Variables of Perceptually Homogeneous Groups Group 1______2______3______4______5______6 Total A g e 1 6 - 2 4 1 1 " ”5 - 0 2 1 5 25 - 39. 3 5 3 2 12 5 31 40 - 65 4 10 3 7 6 17 47 Income 0- 9,999 2 1 1 1 0 4 9 10,000-24,999 5 6 1 3 11 8 34 25,000-50,000+ 1 8 4 4 6 11 34 Sex Male 2 1 1 1 4 4 13 Female 6 15 3 8 16 19 69 Marital Single 2 1 0 2 2 2 9 Status Married 6 15 6 7 18 21 73 Ave. Household # 3.125 3.625 4.000 2.889 4.300 3.000 3.482 H Employed 0 1 2 1 2 1 6 13 1 4 7 2 2 10 8 33 2 or more 3 7 3 5 9 9 37 Years at 1-3 2 8 2 1 6 6 27 address 4-9 3 3 0 1 6 2 15 10+ 3 5 4 7 6 15 40 Ave. # drivers 2.625 2.499 2.167 2.444 2.800 2.651 2.583 A v e ff autos 2.250 1.937 2.000 1.778 2.050 1.957 2.000 Years respondant 11.000 13.062 14.177 14.565 13.905 13.178 13.331 schooling Years spouse 11.333 14.533 13.667 14.125 14.389 14.198 14.000 schooling # stores in choice 2.625 4.250 3.000 3.444 3.450 3.783 3.565 set # groups in choice 1.875 3.000 2.667 2.444 2.250 2.522 2.482 set # trips per week 2.500 3.188 2.500 2.718 2.800 2.869 2.894 158

compared to any of the others. This group also places the

highest priority upon short checkout lines and places the

highest resistance to stores with long lines.

In sum, this young, small family, low income group is

the most concerned with convenience, that is, nearby stores

of a large size, with short lines which may not have the

cheapest prices. This group minimizes the number of food

shopping trips they make, also for convenience. This group

appears to exhibit a satisfying mode of behavior.

Group Two

This group is a slightly older group than group one.

It represents the second highest average income level of

the entire sample. Since all but one respondent is male,

the husband or traditional "head of household" has the

highest educational level of any group (spouse's schooling).

There is also the greatest disparity between husband and

wife's education levels. This group also has the shortest

length of residence of any of the groups and owns about

an average number of automobiles per household.

The most outstanding feature of this group is that

they, by far, have the largest choice sets and number of

perceptual groupings of stores of any of the other groups.

They also make the greatest number of shopping trips per

week. 159

Looking at the structure of preference, quality, variety and low prices seem to be of importance to the individual. Distance does not seem to offer any great barrier if under seven miles. This group expressed inter­ est in short checkout lines and showed the greatest concern for a 24 hour store of any group.

In sum, this group has a concern for quality, variety and convenience, in terms of short checkout lines and low prices. To achieve this, frequent shopping trips are made and distances traveled are greater. They visit more stores and consider many of the stores in their choice set as dis­ tinct. Thus, there appears to be a high degree of optima­ lity in their behavior.

Group Three

This group is somewhat younger than the sample average, high income, married group and has the second largest family size of the identified groups. Half are two pay­ check households and most are long time residents of their neighborhoods. The number of stores in their choice set is quite small but the number of perceptual groups of these stores is relatively high indicating these individuals shop at food stores that they consider to be quite distinct from 160 from one another. Further, this group ties the first group for the fewest average number of shopping trips per week.

Examination of this group's structure of preference shows a concern for low prices and cleanliness. Distance is not considered as severely restrictive as other groups.

An interesting note, however, is that this group expresses a strong preference for smaller stores. This observation is somewhat collaborated by an actual dislike for 24 hour stores( . 058 entry) . Twenty-four hour stores are almost always large chain stores in this area. This result is not unique

(it is also evident in group four), and carries important consequences for spatial choice theory. Models of spatial choice, such as Rushton's Spatial Preference Theory

(Rushton, 1969, 1971) view a tradeoff between attributes such as size and distance as the mechanism by which indi­ viduals choose between alternatives. They select what combination of attributes (a store) which maximizes their utility. Usually this tradeoff is conceived of as trading off increasing distance (increasing time and cost of shopping) for a larger store of greater variety or size.

Evidence of preference for smaller stores makes such a tradeoff concept invalid. Will one travel even further to go to an even smaller store? What is probably more in line with actual behavior is that the criteria of a small, usually friendly atmosphere store, is used as a screening 161

criteria. The nearest small store is selected and these

individuals are simply not interested in size - distance

tradeoffs.

In sum, this group is somewhat younger than sample

average, married, have children and have two sources of

income. They make relatively few shopping trips per week

to select few but distinct stores. Time may operate as

a constraint upon food shopping behavior.

Group Four

This group represents the oldest of the groups. Most are married with very few children still living at home.

Household size is the smallest of all groups. Income is about average with the entire sample. Most of the group, however, have one or two paychecks for income (not retirees).

Average length of residence is the longest of all groups.

The average education levels for the household are the highest of any of the groups'. Choice set size and number of weekly shopping trips is about average.

The structure of preference for this group can be best described as average. The characteristics that do stand out are a strong negative reaction to long travel distances and the preference for small stores coupled with a slight dis­ like for 24 hour stores which has been observed before. 162

Also this group does not seem to mind long checkout lines as much as the other groups.

In sum, this is an older, married, established in the neighborhood group which expresses an interest in smaller

strores. The average choice set size for this group is

3.00, slightly under average, and they resist truly long shopping trip distances.

Group Five

This group is the youngest of all groups, predominant­ ly married with the largest family size. Income levels are average for the entire sample, one half of the households are two paycheck, and length of residence varying from one to more than ten years uniformly. Shopping behavior, i.e. number of shopping trips, choice set size and perceptual groups are only slightly below sample average.

In terms of structure of preference, low prices, a nearby store and the availability of 24 hour shopping are important to the household. Long distances for shopping are quite undesirable. Quality seems to be a low priority for these households. There is a strong preference for large and'medium sized stores.

In sum, this is a young married with family group of average, but often two paycheck income, who favor larger

24 hour stores with low prices that are nearby. Quality is not of true concern to the household. 163 Group Six

This is an older group of relatively low income. They

are long time residents of the area and have a smaller

household size than sample average. The size of the

individual's choice set is large, 3.78 stores. This group

also contains a large number of retirees.

The structure of preference shows both a high desire

for nearby stores as well as a strong aversion to long

distances. Variety, store hours, cleanliness and quality

are of little concern. While this group is least bothered

by high store prices, the group shows some preference for

low price stores but not at the sacrifice of distance.

There seems to be no preference for the size of store patronized except that it is nearby. Despite all of this,

the choice set for the sample is large and an average number of shopping trips are made each week.

In sum, an older, lower income group desiring nearby stores with lower prices, if nearby. They tend to travel to a relatively large number of stores two or three times a week. There may possibly be a recreational motive for these people in their shopping behavior.

4.5 Conclusions

The survey conducted in an area of many possible destinations for grocery shopping has provided a great deal 164 of information as to the choice behavior of people in regard to food shopping.

The respondents use several sources of informat?*on to aid them in food shopping decisions. Store flyers and television seem to be the most influential sources of food store information. Most of the respondents use the same information sources at the same levels indicating variations in shopping behavior are more likely due to differences in the individual's demands for goods and personal attitudes towards stores rather than differences in their informa­ tion levels.

The size of the destination choice sets for individuals is small, an average of about 3.5 stores per individual.

The stores in the smaller choice sets, three or less stores, are usually considered as distinct from one another. This means that, given the specific desires for a particular gro­ cery shopping trip, only one of the stores may be suitable to fulfill these demands. This suggests no tradeoffs exist among alternatives in the individual's choice set. He either goes to the store which fulfills his demand or no trip is made. Hence a sort of "no choice" situation may exist for the individual.

Analysis of the neighborhood choice sets indicates three major findings. First, most individuals discriminate between stores in their choice set on the basis of two or three attributes. Size is the most commonly used attribute. However, from individual to individual, the specific attri­

butes used do vary. This indicates that the destination

choice problem for the individual is simple, discrimination

is based on a few attributes. However, a comprehensive model for a group of individuals will need to incorporate

several attributes. Thus aggregate modeling requires the

use of several attributes while the individual uses only

two or three. This aggregate scheme could cause misleading

tradeoff analysis since not all of the attributes are con­

sidered at the individual level. While at the aggregate level tradeoffs between attributes of destinations may appear to exist, individual behavior suggests tradeoffs only exist between two or three attributes by any single individual. This difference between aggregate versus indi­ vidual behavior illustrates the power of a micro approach to choice behavior. Since the use of tradeoffs by indivi­ duals does seem limited and that the choice sets of indivi­ duals are small, the use of non-compensatory choice models

(non-optimal) seems in order. The second observation is that there is some degree of agreement as to store percep­ tions by using the "pick k out of n minus one" task. How­ ever the problem of missing data makes the analysis results presented in the study suspect. In future studies it may be adventageous to include a common set of stores for indi­ viduals in an area to consider. However, if the choice sets 166 of these individuals are of a similarly small size, such

analysis may itself be misleading due to the inclusion of many stores unfamiliar to the individual. The third obser­ vation is that for this area, satisfaction levels for stores is high. Since satisfaction levels are so high, the perceived penalty for a sub-optimal destination choice is low. Under such conditions, sub-optimal choice behavior would suggest store selection would be based on screening criteria. Size and distance seem likely characteristics for screening criteria but the criteria may be used differently by various individuals. Some prefer small stores over large ones and vice-versa.

Age seems to play an important part in shopping beha­ vior. Younger households with fewer children have small choice sets, make few shopping trips per week and desire large nearby stores. Younger households seem most concerned with convenience and not with quality or even price. Older households also with few children living at home also de­ sire nearby stores but prefer the smaller stores and patron­ ize a greater number of stores perhaps for something to do in a recreational vein.

Those with large families can be separated into two groups. The first group features slightly younger families with two working adults and favor few shopping trips to stores with lower prices and are willing to travel some 167

distance to arrive at that store. There is also a prefer­

ence for smaller stores. The second group is slightly

older with slightly higher incomes and education. They

show a strong preference for quality, variety and low prices and are willing to travel greater distances. They also patronize the greatest number of stores and show pre­ ference for large stores.

Two paycheck households tend to show a behavioral adjustment to time constraints. They tend to shop at a few large stores and are more concerned with convenience than quality, price or cleanliness. They also tend to make fewer shopping trips per week.

A longer length of residence has the effect of reducing the number of stores patronized. While this may in part be due to familiarity with the area, it is most likely due to the factors of an older individual and a smaller family size which reduced demand for food.

It is interesting that certain patterns do not show up in the analysis. Except for a slightly more frequent num­ ber of shopping trips made per week, male shopping patterns and preferences are the same a.s female patterns for grocery shopping. CHAPTER FIVE

CONCLUSIONS

The objective of this study is to advance the under­

standing of the spatial choice problem by examining the

composition of the destination choice sets of individuals.

The study uses grocery shopping as the trip purpose for

examination due to the repetitive nature and frequency of

which such trips are made. The examination of the number

of stores actively considered by the individual and the way

they are perceived as being similar or dissimilar can pro­

vide information as to whether the individual merely satis­

fies his food purchase demands or patronizes several simi­

lar stores in order to comparison shop. This chapter pro­

vides a summary of the major findings of this study along with some brief comments as to the significance of these

findings with regard to choice modeling research. Also, a

series of speculations are provided as to the implications of the findings of this study for future research.

The size of the individual's destination choice set for grocery shopping is small, averaging about 3.5 stores. This result concurs with space-time budget studies conducted in

Europe. These studies, however, suggest that the existence

168 169 of small choice sets are the result of various constraints placed upon the individual by government and other institu­ tional forces, time pressures, lack of mobility, few available alternatives, etc. In the area where this study was conducted, these constraints are very unlikely to exert such a strong influence. Income levels for the study area range from middle to upper middle income. Car ownership is near a two car per household level, indicating a high level of mobility. Finally, the number of food trips undertaken per week indicates a substantial amount of time is directed toward this activity. All of this evidence taken together suggests that for this area, factors other than the con­ straints suggested by space-time budget studies must be at work in reducing choice set size. Thus a modeling approach based solely on institutional constraints would not explain the small choice sets of this area.

The occurrence of small choice sets in conditions of high income and mobility levels runs counter in utility maximization approaches to destination choice studies.

Many utility maximization models exist. However, they all adhere to the basic tenet that the probability that a given destination will be chosen is based on the utility derived from- that choice compared to the total utility of all available alternatives. For this approach to be correct in this area, the few stores actively patronized by the indivi­ dual must be of very high utility while the many stores not 170 patronized must render very little utility. These other stores, due to competition, etc., usually offer the same goods at nearly the same prices. While these stores do differ in distance from the consumer, the differences in the study area are small in both time and money given the mobility and income levels of its residents. Since dis­ tance was shown not to be the most important consideration in evaluating stores and most area stores were rated equally satisfactory, choice by utility maximization, i.e. choosing the best store available, does not seem appropri­ ate. Further, many utility maximization models require the individual to evaluate many stores before the best store can be chosen. The results of the study indicate that most individuals do not evaluate many stores before arriving at their choice set.

The perceptual groupings of stores within individual's choice sets also has serious implications for modeling.

Most choice modeling assumes that choice occurs as a selection from a group of viable alternatives. However, in the case of individuals with small choice sets, one to three stores, each of the stores in their choice sets is considered to be separate and distinct from one another.

Thus, for almost 50 percent of the sample of this study, a type of "no choice" situation exists. The individuals with small choice sets go to that store which meets their 171

requirements. No other store will do. This means that no

tradeoffs (non-compensatory) behavior occurs. In larger choice sets, four or more stores, two or three stores within the choice set are seen as similar. In this case an active choice between alternatives within the choice set for a given trip is possible, but only for a subset of the stores within the individual's choice set. This indicates that even in the larger choice sets, most choice is of a non-compensatory nature.

Another important consideration in destination choice modeling is the manner in which the alternatives are per­ ceived in terms of various store attributes. In this study, a total of seven attributes are found as being used by the entire sample population. These attributes are: variety

(size), quality, prices, distance, cleanliness, checkout time (convenience) and store hours. Size is most commonly used by the sample in discriminating between stores. How­ ever, the manner in which individuals display a preference for size varies across individuals. Some respondents show a stronger preference for larger stores as might be expected.

Others show a strong preference for the smaller, typically friendlier stores. In terms of any tradeoff analysis, this type.of preference would be difficult to model. If a small store is friendlier, is it more preferable? While this feature is deemed important by individuals, not one respon­ dent mentioned it as an important or salient attribute in 172

discriminating between stores in his choice set. Friend­

liness, if a desired attribute, either exists in a store or

not. Such individuals are not likely to give up this fea­

ture for other considerations. That is to say, no tradeoffs

are likely to exist between friendliness and other attri­

butes of stores.

The validity of tradeoff analysis in the case of

grocery shopping is further eroded by the following obser­

vations. While a total of seven attributes are found to be

used to discriminate between stores by the sample popula­

tion, each individual uses only two or three attributes to

distinguish between stores in his own choice set. This is

true even in cases of large choice sets. More importantly,

however, is that the specific attributes used vary from person to person. This observation points out the strength of the micro approach. At an aggregate level, seven attri­ butes are significant in store choice considerations. At

the individual level, however, only two or three attributes are important for any one case. The second conclusion is in regard to tradeoff theory. If only two or three attributes are used to discriminate between stores at the individual level, then only tradeoffs between these few attributes are valid. Tradeoffs between attributes used and not used are of questionable value. At the aggregate level, such trade­ offs appear valid. But if a tradeoff analysis is 173

performed for the seven identified attributes, it goes

beyond the simple manner by which the individual perceives

alternatives. While it may give some very useful informa­

tion as to individual preferences at an abstract level, to

use this information to predict behavior at an individual

level may be misleading since behavior is non-compensatory.

As an example, a tradeoff analysis can be conducted

between cleanliness and another store attribute, say price.

However, in reality, the situation may be that the indivi­

dual would never consider going into a "dirty" store. No

tradeoff thereby exists although the device by which trade­

off analysis is conducted can be completed, with some dif­

ficulty, by an individual respondent to the survey.

Satisfaction levels for stores in the study area are

very high and equally high for stores of comparable size.

The observation of high store satisfaction levels and small

choice sets lend themselves to an important conclusion.

Since satisfaction levels are so high, perhaps there is

little reason for the individual to extend his spatial

search for a more satisfactory store. The perceived cost of a sub-optimal choice is very low since nearby stores are highly regarded and stores further away are not considered

significantly better to the consumer. This phenomena

strongly suggests the use of non-optimal choice theory as an appropriate means of modeling behavior in this framework. 174

In a choice environment of limited time and the availability of numerous options of nearly equal levels of satisfaction, choice may be one of merely finding an alternative which proves satisfactory as opposed to finding the best alterna­ tive.

Several major future avenues of study are immediately evident from the results of this study. One is to study the relationship between the choice environment, number of alternatives, degree of variety among alternatives, time constraints, etc., and the type of non-optimal choice rule which may be used by the individual. Several non-optimal rules, such as the elimination by aspects model by

Tversky (1972) and the conjuctive and disjunctive models described by Einhorn (1970) seem well suited to modeling food store choice behavior. Studies examining the relation­ ships between the complexity of choice and the choice rule used are beginning in psychology and marketing but only in a clinical environment. Geographic contributions to this area could be in the establishment of what screening cri­ teria are used to simplify the choice problem. That is, what attributes are used to eliminate obviously unacceptable alternatives without extensive information about these alternatives being required. Further, the levels at which an alternative would be considered acceptable for these screening criteria and whether these levels are j 175

similar across individuals would be crucial for the

eventual establishment of non-optimal choice theory for destination choice modeling. This could be accomplished

by the critical examination of the process by which indivi­ duals include or exclude a possible destination from his choice set.

Another major avenue of study is to replicate this

study in areas of even higher income levels and much lower

income levels. How the size and complexity of the choice set changes at various levels of income would be of interest.

Similar studies can be conducted for other trip pur­ poses. Retail commercial store trips immediately come to mind. Department stores can be considered a higher order center than grocery stores. The size of the choice sets for retail establishments and the methods used by indivi­ duals to decide which stores to visit would be of immense commercial value.

The usefulness of this study goes beyond the original problem of destination choice. This study provides a method to relate overt behavior modified by the composition of the individual's choice set to the satisfaction levels for these stores. If choice set size is small while satisfaction levels are low, this suggests individuals are either severely restricted in movement so that these desires are not met or the available alternatives do not meet local 176

demands. While on the other hand, if choice sets are small

and satisfaction levels high, the demands of the residents

of the area are indeed met by their surroundings. In such

a way, a city-side survey of destination choice set

composition with the structure of preference section could

be used to identify areas of high dissatisfaction and re­

stricted movement within the city and identify the features of these areas which are not in accordance with what the population wants. This information could help urban planners target future areas of development to best meet

the needs of the local residents by providing new destina­

tions which would be most likely considered as being

included in the choice sets of the residents of these areas. BIBLIOGRAPHY

Adler, T., and M. Ben-Akiva, A Theoretical and Empirical Model of Trip-Chaining Behavior, Draft Manuscript, Department of Civil Engineering, MIT, 1977.

Aldskagius, H., "A Conceptual Framework and a Swedish Case Study of Recreational Response and Environmental Cognition," Economic Geography, Vol. 53 (1977), 126- 152.

Anderson, J., "Space-Time Budgets and Activity Studies in Urban Geography and Planning," Environment and Plan- ning,A,Vol. 3 (1971), 353-368.

Ansah, J. A., "Destination Choice Set Definition in Travel Behavior Modeling," Transportation Research, Vol. 11 (1977), 127-140.

Astrom, T. R., Laws of Traffic and Their Applications to Traffic Forecasts with Special Reference to the Second Bridge Project, Stockholm, Royal Institute of Techno­ logy, 1973.

Bain, J. H., Activity Choice Analysis Time Allocation and Disaggregate Travel Demand Modeling, Ph. D. Disserta­ tion, Department of Civil Engineering, MIT, 1975.

Beckmann, M. J. and T. F. Golob, "Traveler Decisions and Traffic Flows," in D. J. Backly ed., Transportation and Traffic Theory, Sydney: A. H. and A. W. Reed, 1974.

Ben-Akiva, M., Structure of Passenger Travel Demand Models, Ph. D. Thesis, MIT, 1973.

______. "Alternative Travel Behavior Structures: Structure of Passenger Travel Demand Models," Trans­ portation Research Record No. 526 (1974), 26-42.

Black, W. R., "The Utility of the Gravity Model and Esti­ mates of its Parameters in Commodity Flow Studies," Proceedings of the Association of American Geographers Vol. 3 (1971) 5*8-357“

177 178

Brand, D. "The State of the Art of Travel Demand Forecast­ ing: A Critical Review," Graduate School of Design, Harvard University Cambridge, 1972.

Brog, W. , D. Heuwinket and K. Neumann, "Psychological Determinants of Use Behavior," Round Table, Vol. 34, (1977) OECD, Paris.

Burnett, K. P., "The Dimensions of Alternatives in Spatial Choice Processes," Geographical Analysis, Vol. 5 (1973) 187-204.

______. "A Three State Markov Model of Choice Behavior within Spatial Structures," Geographical Analysis, Vol. 6 (1974), 653-668.

______. Choice and Constraints Oriented Modeling: Alternative Approaches to Travel Behavior, Federal Highway Administration DOT-FH-11-9389, Final Draft, 1978.

______. "Behavioral Geography and Modeling Urban Spatial Behavior," in R. G. Golledge and K. R. Cox eds., Behavioral Problems in Geography Revisited: A Perspective, London: Methuen, 1980.

______and S. Hanson, "A Rationale for an Alternative Mathematical Approach to Movement as Complex Human Behavior," Forthcoming, Proceedings, 7th Annual P.T.R.C. Meetings, Warwick, England, July 1979.

Burns, L. D. and T. F. Golob, "The Role of Accessibility in Basic Transportation Choice Behavior," Transporta­ tion, Vol. 5 (1976) 175-198.

CATS, Chicago Area Transportation Study, Final Report, Vol. II (1960) 81-82 and 111.

Carrothers, G., "An Historical Review of the Gravity and Potential Concepts of Human Interaction," Journal of the American Institute of Planners, Vol. 22 (1956) 94-102.

Cesario, F. J. and T. E. Smith, "Directions for Future Research in Spatial Interaction Modeling," Papers of the Regional'Science Association, Vol. 35 (1975) 57-72. ' 179

Chapin, F. S., "Activity Systems and Urban Structure: A Working Schema," Journal of the American Institute Of Planners, Vol. 34 (1968) 11-18.

Chombart de Lauwe, P. H., "The Social Sciences: Urbanism and Planning," International Journal of Comparative Sociology, Vol. 4 (1963) 19-30.

Coombs, C. H., A Theory of Data, Wiley: New York, 1964.

Cullen, I. G., "Space-Time Budgets and Activity Studies in Geography and Planning," Environment and Planning A, Vol. 3 (1971) 353-368.

Dash, J. F., L. G. Schiffman and C. Berenson, "Information Search and Store Choice," Journal of Advertising Re­ search , Vol. 16 (1976) 35-40.

Dix, M. C., Report on Investigations of Household Travel Decision Making Behavior, Working Paper 27, Transport Studies Unit, University of Oxford, Oxford, England, 1977.

Dobson, R., Market Segmentation: A Tool for Transportation Decision-Making Resource Paper, Third International Conference on Behavioral Travel Modeling, Tanunda, Australia, 1977.

______., J. F. Kehoe, "Disaggregated Behavioral Views of Transportation Attributes," Transportation Research Record No. 527 (1974) 1-15.

______., and M. L. Tischer, Perceptual Market-Segmenta- tion Technique for Transportation Analysis, Paper presented at the Transportation Research Board Meetings, Washington, D.C. , 1978.

Domencich, T. A., and D. McFadden, Urban Travel Demand: A Behavioral Analysis, North-Holland, Amsterdam, 1975.

Einhorn, H. J., "The Use of Non-Linear, Noncompensatory Models in Decision-making," Psychological Bulletin, Vol. 73 (1970) 221-230. “

Ellegard, A., T. Hagerstrand and B. Lentorp, "Activity Organization and the Generation of Daily Travel: Two Future Alternatives," Economic Geography, Vol. 53 (1977) 121-152. 180

Ewing, G. O., An Analysis of Consumer Space Preferences Using the Methocf of~~Paired Comparisons, Ph. D. Thesis, McMaster University, Hamilton, Ontario, 1971.

Eyles, "Pouring New Sentiments into New Theories: How Else Can We Look at Behavioral Patterns?" Area Vol. 3 (1971) 242-250.

Fishburn, P. C., Decision and Value Theory, Wiley: New York, 1964.

Fried, M. et al., Travel Behavior: A Synthesized Theory, Final Report Phase I, Project B8-14, Washington, D.C., N.C.H.R.P. 1977.

Goldberger, A. S., Functional Form and Utility: A Review of Consumer Demand Theory, Workshop Paper 6703, Madison: Center for Systems Formation Methodology and Policy, Social Systems Research Institute, University of Wisconsin, Mimeo 1967.

Golledge, R. G., "Some Equilibrium Models of Consumer Behavior," Economic Geography, Vol. 46 (1970), 417-424.

______. Representing, Interpreting and Using Cog­ nized Environments, Paper presented at the Regional Science Association Meetings, Philadelphia, 1977.

Golob, T. F. and R. Dobson, "Assessment of Preferences and Perceptions Toward Attributes of Transportation Alternatives," TRB Spec. Rept, 149 (1974) 58-81.

______, E. T. Canty, R. L. Gustafson and J. E. Vitt, "An Analysis of Consumer Preferences for a Public Transportation System," Transportation Research, Vol. 6, (1972) 81-102.

Graham, E., "What is a Mental Map," Area Vol. 8 (1975) 258-62.

Gray, P., "Non Exploration in Urban Geography," Area Vol. 7 (1975) 228-235.

Green, P. and D. Tull, Research for Marketing Decisions, .Englewood Cliffs: Prentice Hall, 1975.

Gustafson, R. L. and F.P.D. Navin, "User Preferences for Dial a Bus, A Comparison of Two Cities," General Motors Research Laboratories, Warren, Michigan, Publ. GMR-1217, 7-21. 181

Hagerstrand, T., "What About People in Regional Science?" Papers of the Regional'Science Association, Vol. 24, (1969) 7-21.

Heggie, I. G., Putting Behavior into Behavioral Models of Travel CHoice, Working-Paper 22, Transport Studies Unit, University of Oxford, Oxford, England, 1977.

______, and P. M. Jones, "Defining Domains and Models of Travel Demand," Transportation, Vol. 7(1978) 119-126.

Henderson, J. M. and R. E. Quandt, Microeconomic Theory: A Mathematical Approach, McGraw Hill: New York, second edition, 1971.

Hensher, D. A., Multitrip and Multipurpose Journeys, Some Suggestions, Working Paper 15, Transport Studies Unit, Oxford, England, 1975.

______, "The Structure of Journeys and the Nature of Travel Patterns," Environment and Planning A, Vol 8 (1976) 655-672.

Higgs, G., "An Assessment of the Action Component of Action Space," Geographical Analysis, Vol. 7 (1975) 35-54.

Horton, F. E. and D. R. Reynolds, "Effects of Urban Spatial Structure on Individual Behavior," Economic Geography Vol. 47 (1971) 36-48.

Johnson, R. M., "Pairwise Nonmetric Multidimensional Scal­ ing," Psychometrika, Vol. 38 (1973) 11-18.

______, "A Simple Method for Pairwise Monotone Re­ gression," Psychometrika, Vol. 40 (1975) 163-168.

Johnson, S. C., "Hierarchical Clustering Schemes," Psychometrika, 32 (1967) 241-254.

Jones, P. M., New Approaches to Understanding Travel Beha­ vior: The Human Activity Approach, Resource Paper, Workshop A. Third International Conference on Behavioral Travel Modeling, Tanumda, South Australia, 1977.

______, "A Methodology for Assessing Transportation Policy Impacts," Transportation Research Record, No. 723 (1979) 52-58. Kelly, G. A., The Psychology of Personal Constructs, Norton Press, 1955. 182

Knight, R. L. and M. D. Menchik, Conjoint Preference Estimation for Residential Land Use Policy Evaluation, Rand Publication P-5172, 1974.

Kraik, K. H., "Discussion of Communal Behavior and the Environment," in Behavior and Environment: The Use of Space by Animals and Men, ed. A. H. Esser, Plenum, N.Y. 1971.

Kruskal, J. B., "Multidimensional Scaling by Optimizing Goodness of Fit to a Non-metric Hypothesis," Psycho­ metrika , Vol. 29 (1964) 1-27.

______, and M. Wish, Multidimensional Scaling, Beverly Hills, CA: Sage Publications, Inc. 1978.

Kutter, E., Individual Needs and Urban structure Determin­ ants of~Behavior-Oriented Model Construction in Travel Planning, Mimeo, Techinisch en Universitat, Berlin, Germany, 1976.

Lancaster, K. J., "A New Approach to Consumer Theory," Journal of Political Economy, 74 (1967) 132-157.

Lenntorp, B., "Paths in Space-Time Environments: A Time Geographic Study of Movement Possibilities of Indivi­ duals," Lund Studies in Geography, Series B. Gleerup, Lund, 1976.

Luce, R. D., Individual Choice Behavior, Wiley: New York, 1959.

MacKay, D. B. et al., "Cognitive Maps and Spatial Behavior of Consumers," Geographical Analysis, Vol. 7 (1975) 19-34.

Marble, D. F., Three Papers on Individual Travel Behavior in the City, Transportation Research Center, North­ western University, Evanston, IL, 1967.

______, and S. Bowlby, "Shopping Alternatives and Recurrent Travel Patterns," in F. Horton ed. Geographic Studies of Urban Transportation and Network Analysis, Northwestern University Studies m Geography, No. 16, Evanston: Northwestern University Press, 1968.

Meier, R. L., A Communications Theory of Urban Growth, MIT Press, Cambridge, Mass., 1962.

Mishan, E. J. , "Theories of Consumer Behavior: A Cynical View," Economica, Vol. 28 (1961) 1-11. 183

Murchland, J. D., Annotated Bibliography on the Validation of Transport Models (unpublished) Transport Studies Group, University College, London, 1977.

Nicholaidis, G. C. and R. Dobson, "Disaggregated Perceptions and Preferences in Transportation Planning," Transpor- tation Research, Vol. 9 (1975) 279-295.

Nystuen, J. D., "A Theory and Simulation of Intraurban Travel," in Quantitative Geography: Part I - Economic and Cultural Topics (Garrison, W. L. and D. F. Marble, eds) Department of Geography, Northwestern University, Evanston, IL Studies in Geography, Vol. 13 (1967) 54-83.

Olsson, G., "Central Place Systems, Spatial Interaction and Stocastic Processes," Regional Science Association: Papers, Vol. 18 (1966) Vienna Congress.

Oppenheim, H., "A Typological Approach to Individual Urban Travel Behavior Prediction," Environment and Planning A, Vol. 7 (1975) 141-152.

Parks, D. and N. Thrift, "Timing Space and Spacing Time," Environment and Planning A , Vol. 7 (1975) 651-670.

Pirie, G. H. , "Thoughts on Revealed Preference and Spatial Behavior," Environment and Planning A, Vol. 8 (1976) 947-956.

Pred, A., "Planning Related Swedish Geographic Research," Economic Geography, Vol. 3 (2) (1977) Special Issue.

Rao, V. R. and R. Katz, "Alternative Multidimensional Scaling Methods for Large Data Sets," Journal of Marketing Research, Vol. 8 (1971) 488-494.

Rassom, P. R., R. H. Ellis, and J. C. Bennett, The N-Dimen- sional Logit Model: Development and Application, Peat, Marwick, Mitchell & Co., Washington, D.C., 1970.

Robinson, W. S., "Ecological Correlations and the Behavior of Individuals," American Sociological Review, Vol. 15, (1950) 351-357.

Robinson, R. V. F. and R. W. Vickerman, "The Demand for Shopping Travel: A Theoretical and Empirical Study," Applied Economics, Vol. 8 (1976) 267-281. 184

Rushton, G., "Analysis of Spatial Behavior by Revealed Space Preference," Annals Association of American Geographers, Vol. 59 (1969) 391-400.

______, "Behavioral Correlates of Urban Spatial Structure," Economic Geography, Vol. 47 (1971) 49-58.

Samuelson, P. A., "Consumption Theory in Terms of Revealed Preference," Econometrica, Vol. 15 (1948) 243-253.

Sen, A., "Behavior and the Concept of Preference," Econo- metrica Vol. 40 (1973) 241-249.

Shapcott, M. and P. Steadman, A Study of the Introduction of Flexible Working Hours in Reading,(unpublished) The Martin Centre for Architectural and Urban Studies, Cambridge University, 1977.

Shepard, R. N., "The Analysis of Proximities: Multidimen­ sional Scaling with an Unknown Distance Function," Psychometrika, Vol. 27 (1962) 125-139, 219-246.

Sheppard, E. S., "The Ideology of Spatial Choice," Papers of the Regional Science Association, Vol. 45 (1980) 197-213.

Shively, W., "Ecological Inference: The Use of Aggregate Data to Study Individuals," American Political Science Review, Vol. 63 (1969) 1183-1196.

Siegel, S., Nonparametric Statistics for the Behavioral Sciences, New York: McGraw-Hill 1956.

Smith, T. E., "A Choice Theory of Spatial Interaction," Regional Science and Urban Economics, Vol. 5 (1975) 137-176.

Stopher, P. R. and T. E. Lisco, "Modeling Travel Demand: A Disaggregate Behavioral Approach - Issues and Applications," Transportation Research Forum Proc., 1970.

Stopher, P. R. and A. H. Meyburg, Urban Transportation Modeling and Planning, Lexington Books, D. C. Heath and Company, Lexington, Mass. 1975.

Stouffer, S. A., "Intervening Opportunities: A Theory Relating Mobility and Distance," American Sociological Review, Vol. 5, No. 6 (1940) 845-867. 185

Thrift, N. J., An Introduction to Time Geography, Concepts and Techniques in Modern Geography No. 13 (1977) Geo Abstracts Ltd., University of East Anglia: Norwich.

______, "Time and Theory in Human Geography: Parts 1 and 2," Progress in Human Geography, (1977) London: Edward Arnold.

Tivers, J., Constraints on Spatial Activity Patterns: Women with Young Children, Occasional Paper, 6 (1977) Geography Department, Kings College, London.

Tversky, A., "Choice by Elimination," Journal of Mathemati­ cal Psychology, Vol. 9 (1972) 341-367.

Von Rosenbladt, B., "The Outdoor Activity System in an Urban Environment," in A. Szalai, ed. The Use of Time, The Hague; Mouton Press, 1972.

Westelius, 0., The Individual's Way of Choosing Between Alternative Outlets, Stockholm: Svensk Bygotjunsh, 1973

Wilson, A. G., "A Family of Spatial Interaction Models and Associated Developments," Environment and Planning A, Vol. 3 (1971) 1-32.

Wright, P., "Consumer Choice Strategies: Simplifying vs. Optimizing," Journal of Marketing Research, Vol. 12 (1975) 60-67.

______, "The Harassed Decision Maker: Time Pressures, Distractions and the Use of Evidence," Journal of Applied Psychology, (no date).

Young, F. W., An Asymmetric Euclidean Model for Multiprocess Asymmetric Data, U.S.-Japan Seminaron Multidimensional Scaling, 1975.

______, and R. Lewyckyj, Alscal-4 User's Guide, Data Analysis and Theory Associates, University of North Carolina, Chapel Hill, 1979.

Zahavi, Y., Travel Time Budgets and Mobility in Urban Areas, Washington, D.C., FHWA, 1974.

______, Travel Characteristics in Cities of Developing and Developed Countries, Staff Working Paper, 230 (1976) World Bank, Washington, D.C.