Urban and Regional Planning Review Vol. 4, 2017 | 251

Regional Differences in the Socio-economic and Built-environment Factors

of Vacant House Ratio as a Key Indicator for Spatial Urban Shrinkage

Hiroki Baba* and Yasushi Asami**

Abstract The phenomenon of urban shrinkage is surging through cities in . To understand the current situation of urban shrinkage in Japan, the number of vacant housing would be one of the key indicators. The purpose of the study is to clarify the regional differences in the socio-economic and built-environment factors of vacant house ratio as a key indicator for spatial urban shrinkage. The characteristics of cities are identified by means of socio-economic and built-environment factors, which enable us to describe the current situation of spatial urban shrinkage more precisely. In this analysis, 771 cities in Japan are original units in order to articulate the geographical differences in urban shrinkage. To differentiate regions, we classified cities according to urban employment area (UEA) into 4 categories. Using 7 socio-economic and built-environment variables, a general linear mixed model was employed to clarify the statistical significances in each UEA category. The key findings were as follows. First, strong trends of vacant house ratio and its factors among categories were found. Second, socio-economic factors affecting vacant house ratio in each category were statistically significant in all the categories. Third, built- environment factors also had a clear relationship with vacant house ratio, but the influences varied among the four categories. Finally, in order to prepare for future spatial urban shrinkage, areas of anticipated spatial urban shrinkage were identified from the results of the study.

Keywords shrinking cities, vacant house, socio-economic factor, built-environment factor, urban employment area, generalized linear mixed model

* Doctoral Student, Department of Urban Engineering, Graduate School of Engineering, The University of Tokyo ** Professor, Department of Urban Engineering, Graduate School of Engineering, The University of Tokyo E-Mail: [email protected]

(C) 2017 City Planning Institute of Japan http://dx.doi.org/10.14398/urpr.4.251 Urban and Regional Planning Review Vol. 4, 2017 | 252

1. Introduction The phenomenon of urban shrinkage is surging through cities in Japan. Urban shrinkage is normally defined by the decline of population in a city (Hollander et al. 2009) [1]. The population decline in Japan started from 2008, and the total population will decrease by approximately 30% by 2055 (NIPSSR 2012) [2]. Therefore, nearly all cities in Japan are expected to experience urban shrinkage. Population decrease reduces the efficiency of infrastructure, and consequently infrastructure will require more budget to maintain (Asami 2014) [3]. However, it could also be an opportunity to restructure the cities (Haase 2008) [4]. One of the main impacts that urban shrinkage entails is an increase in vacant housing (Rieniets 2009) [5]. A large amount of vacant housing will lead to anxiety about crime and the deterioration of the city environment (Spelman 1993) [6]. Considering the issues above, the quantity of vacant housing is associated with the spatial impacts of urban shrinkage, so the quantity of vacant housing would be one of the key indicators for urban shrinkage. In this study, we define the spatial situation of urban shrinkage as “spatial urban shrinkage.” Employing the vacant house ratio in order to explain spatial urban shrinkage enables us to clarify the multiple factors relating to shrinking cities because vacant housing is associated with the built-environment and dwelling intentions of local residents. Along with vacant housing, there are various types of socio-economic and built-environment factors. Therefore, it is impossible to explain the dynamics of cities using a simple factor. However, we could explain spatial urban shrinkage with the use of vacant housing to some extent, by using several important factors so that more detailed discussion is possible regarding the dynamics of vacant housing. According to Figure 1, a change in social and political conditions transforms the dynamics of cities as an external stressor. This stress affects the socio-economic status of local residents which is associated with population decline. This dynamic change has a spatial impact on current land use, and finally vacant properties such as vacant housing would increase. Within the conceptual image of spatial urban shrinkage, some of the socio-economic and built- environment factors can be observed, and they are closely related to the vacant house ratio, which could be an indicator of spatial urban shrinkage. The purpose of the study is to clarify the regional differences in socio-economic and built-environment factors of vacant house ratio as a key indicator for spatial urban shrinkage. An increase in vacant housing is the result of a shortage of people to maintain the property. Therefore, vacant house ratio could be one of the indices indicating the extent to which urban shrinkage is or will be advancing. Hence, analyzing cities according to the ratio of vacant housing would provide us with suggestive results as an important indicator of spatial urban shrinkage. Then, we need to categorize cities into several types of regions. We define a region as a group of Figure 1. Conceptual image of cities which are considered to have similar spatial urban shrinkage characteristics. Cities located in the Tokyo Urban and Regional Planning Review Vol. 4, 2017 | 253 metropolitan area and cities located in rural areas have totally different socio-economic and built- environment characteristics, so identifying the characteristics of cities by region enables us to describe spatial urban shrinkage more precisely. After clarifying what kind of factors positively or negatively affect the vacant house ratio, this paper offers a possible way of predicting spatial urban shrinkage. Although there is no panacea to improve the quality of areas where spatial urban shrinkage is occurring, we are able to anticipate where shrinking cities will occur. In this study, all of the cities in Japan will be analyzed. Since the motivation of this study is to clarify the factors of spatial urban shrinkage in Japan, we should focus on the analysis unit, from which it is possible to acquire the data for the entire country and need not to gather all the detailed information such as age, structure, and floor area of each building. Therefore, we focused on a regional scale and adopted municipality as the analysis unit. However, due to the limited data available, we excluded the data from towns and villages. We also excluded the cities with a population of 500,000 or more granted special rights by government ordinance for the following reason. We would like to focus on cities which are expected to suffer from urban shrinkage. Most of the cities with a population of 500,000 or more granted special rights by government ordinance are expected to experience a population growth, and a significant number of local residents commute to such cities. If we include the cities with a population of 500,000 or more granted special rights by government ordinance in the three major UEAs, different characteristics of cities are mixed together, and we cannot clarify the characteristics of the cities, where urban shrinkage is happening. We also would like to compare several types of regions under the condition of a similar city scale, so that we can consider the regional differences in the factors affecting the vacant house ratio. Therefore, the cities with a population of 500,000 are regarded as outliers compared with the other cities. Finally, 771 cities were obtained as analysis units. Using such cities, we are capable of conducting trans-regional analyses in order to articulate the geographical differences of spatial urban shrinkage.

2. Literature review 2.1. Shrinking city studies In several countries in Europe, urban shrinkage has started ahead of Japan. Therefore, researchers have been trying to clarify drivers pushing urban shrinkage in multiple study areas. In Germany, whose cities are suffering from urban shrinkage, economic disparity and declining birth rate are significant drivers (Wiechmann 2012) [7]. Several cities in the United States have shrunken because of long-term industrial transformation (Wiechmann 2012; Schilling and Logan 2008) [8]. Japan also encounters similar kinds of drivers such as declining birth rate and increased concentration of population in major metropolitan areas (Martinez-Fernandez et al. 2016) [9]. Generally, most of the causes and effects of urban shrinkage have been clarified, but the causes and effects differ depending on the countries. Recently, a multitude of field investigations on urban shrinkage have been carried out, and such single case studies have brought about many meaningful causes and effects (Rink et al 2012; Haase et al. 2012) [10, 11]. For instance, Bontje (2005) [12] investigated in detail the relationship between urban shrinkage and the city government responses in Leipzig, and suggested development strategies for post-socialist cities. Although such case studies describe detailed but specific features of urban shrinkage, the context and dynamics of the phenomenon on a large scale should be focused on as well (Grossmann et al. 2013) [13]. Haase et al. (2014) [14] conceptualized urban shrinkage including such context and dynamics. It is a theoretical framework which Urban and Regional Planning Review Vol. 4, 2017 | 254 includes causes, impacts, responses, and feedback loops. The author argues that the framework does not explain all the types of shrinkage, but it helps researchers discover the time-space context of urban shrinkage in study areas. Consequently, drivers of urban shrinkage vary depending on the cities. There are many studies discovering drivers in single study areas, but the dynamic relationships between drivers in cities need to be researched further.

2.2. Current situation of vacant housing and lots as an emerging phenomenon of spatial urban shrinkage in Japan As we mentioned above, shrinking cities are increasing in Japan. Therefore a large amount of research has been conducted explaining the current situation of vacant housing and lots as an emerging phenomenon of urban shrinkage. On a city scale, vacant housing and lots are generated in a mosaic pattern. Several research studies have focused on the relationship between such mosaic patterns and types of residential neighborhoods (Ujihara et al. 2016a; Ujihara et al. 2016b) [15, 16]. The research concluded that older and highly sprawling residential neighborhoods are inclined to have a large amount of vacant housing and lots. Sakamoto and Yokohari (2016) [17] examined the transformation of vacant housing and lots in residential neighborhoods, and clarified the differences between the features of vacant housing and those of vacant lots. Such quantitative research has been conducted recently, but the number of quantitative research articles is not enough to understand the current situation of spatial urban shrinkage in Japan. In the case of comparative research on a country scale, not so much research has been done yet. Abe et al. (2012) [18] focused on vacant lots, and investigated their transition in the central business districts of 37 cities across Japan. Kanamori et al. (2015) [19] analyzed the factors of vacant housing and estimated the number of vacant housing from the perspective of the equilibrium between construction and demolition. Both studies explained the difference of spatial urban shrinkage on a regional scale, but cities are not categorized in a meaningful manner, and further comparative analyses need to be conducted.

2.3. Characterization of cities by socio-economic and built-environment factors Researchers focusing on single case studies clarified the relationship between spatial urban shrinkage and socio-economic and built-environment factors. Takeshima et al. (2004) [20] estimated the change of population in suburban neighborhoods, and concluded that decreasing birth rate and aging population in neighborhoods are attributed to the deterioration of neighborhoods and the increase of vacant housing. Economic factors are also reported as drivers of spatial urban shrinkage. Abe et al. (2012) [18] found that land value has a negative effect on the ratio of underused land. Looking at the subjective satisfaction of local residents, it is clarified that a better dwelling environment is related to a low ratio of vacant housing and lots (Katayama et al. 2006) [21]. Recently, a dynamic transition of socio-economic conditions occurred among cities in Japan, and such socio-economic factors are important to reflect spatial urban shrinkage. Built-environment factors are associated with the spatial differences of vacant housing and lot. Nakai et al. (2012) [22] surveyed the densely built-up area, and found that vacant housing tends to be built out of wood, relatively small, and adjacent to a narrow road, which is less than 4 meter, although the results of the survey were restricted to a densely built-up area. Sakamoto and Yokohari (2016) [17] broadened the target area to the city scale, and clarified the relationship between vacant housing and lots and adjacent roads. Although socio-economic and built-environment factors, which affect the ratio of vacant Urban and Regional Planning Review Vol. 4, 2017 | 255 housing and lots, have been clarified, comprehensive studies of cities with the use of such factors need to be conducted further. Some pieces of previous research have added multiple variables into the model explaining the vacant house ratio (Sakamoto and Yokohari 2016) [17], but they usually utilized only built-environment variables due to the problem of data collection. Since the analysis unit of our study is the municipality, we could have access to various types of data including socio-economic status. There are also many pieces of research into the occurrence factors of vacant housing, they usually focus on a specific city and clarify the inclination of vacant housing at a city scale. Since each neighborhood of the city has different characteristics, it is difficult to extrapolate the results to other cities. It is true that our study cannot explain the characteristics of each neighborhood within a city, but we illustrate the inclination of the vacant house ratio without any extrapolations. One example of regional quantitative analysis is that Ai (2016) [23] scored the living conditions using an index of population transition by neighborhood unit. Yet, since the study is restricted to the Tokyo metropolitan region, there is a lack of research comparing the factors affecting spatial urban shrinkage between regions. Since the target area of this study is the whole country, it is possible to describe regional differences in spatial urban shrinkage using multiple factors.

3. Method 3.1. Categorization of cities by Urban Employment Area Understanding the characteristics of urban shrinkage heavily depends on how cities are categorized. The main driver which stipulates the scale of urban areas would be the economic activity of local residents such as commuting. Kanemoto and Kurima (2005) [24] defined Urban Employment Area (UEA), whose boundary is based on the commuting of workers and densely populated district (DID). In a UEA, there are two types of areas: core and suburb. The core of a UEA is made up of densely settled municipalities with a population of more than 10,000 in the DID. The suburb of a UEA is the area where at least 10% of workers commute to a specific core. A UEA, whose core has a population of more than 50,000 in the DID, is called a Metropolitan Employment Area (MEA). Also, a UEA, whose core has a population of 10,000 to 50,000 in the DID, is called a Micropolitan Employment Area (McEA). In this study, we categorized cities from the perspective of UEAs. As mentioned in the introduction, 771 cities are targeted as the categorization excludes cities with populations of 500,000 or more granted special rights by government ordinance. First, the three largest UEAs (Tokyo, , and metropolitan areas) are categorized as “Three major UEAs,” because the population of these UEAs is at least 5,000,000, which is much larger than the others. Since we excluded the cities with a population of 500,000 or more granted special rights by government ordinance, this area does not represent the comprehensive characteristics of the Figure 2. Conceptual diagram of UEAs. However, three major UEAs are different categorization by UEAs from the other UEA categories because they are Urban and Regional Planning Review Vol. 4, 2017 | 256 influenced by strong cores and most of the workers commute to the core cities despite having many hours of commute time. Also, due to the high population density, public transportation is developed. Second, cities including MEAs are called “Large core UEAs.” A substantial number of workers commute to the core cities although the influence of the core cities is weaker than the one in three major UEAs. It is thought that local residents are more likely to use a private vehicle and are not restricted to the disposition of public transportation. Third, cities including McEAs are “Small core UEAs.” Cities in small core UEAs are located in rural areas rather than urbanized areas. Supposedly, the daily mode of transportation is a private vehicle, and public transportation is not frequently used. Lastly, the other cities are categorized as “Independent cities,” which retain no cores and do not strongly depend on other cities (Figure 2). Independent cities do not have any core cities nearby and local residents are spread out. Because of the low population density, most of the local residents use a private vehicle for their daily lives.

3.2. Data selection This paper deals with multiple factors shared by cities with increasing amounts of vacant housing. The factors ranges from socio-economic to built-environment ones. Upon the selection of the factors, we referred to the factors used in previous research. According to built-environment factors, building structure, floor area, and the width of adjacent road are related to the occurrence factors of vacant housing (Sakamoto and Yokohari 2016; Nakai et al. 2012) [17], [22]. A better dwelling environment also lowers the ratio of vacant housing (Katayama et al. 2006) [21]. Focusing on socio-economic factors, the existence of aging populations in neighborhoods results in an increase in the vacant house ratio (Takeshima et al. 2004) [20], and land value is negatively related to the quantity of underused land (Abe et al. 2012) [18]. Reflecting on the studies mentioned above, we selected the factors for the analysis. There follows a brief explanation of and hypotheses regarding the factors. Ratio of vacant housing: This ratio is a response variable. The definition of vacant house could be classified as in temporary use, secondary use, leased, and others. We employ the “others” classification of vacant housing by 2013 Housing and Land Survey, because most of the vacant housing that should be demolished are included in this definition. Ratio of young residents (AGE): This factor indicates the population ratio of the 20 to 39 age group. As illustrated in the literature review, age structure is affected by the quantity of vacant housing. The ratio of elderly people also explains the vacant house ratio well. However, we found multicollinearity between the ratios of elderly people and young residents, so we chose the ratio of young residents, which contributes to the activation of the second-hand housing market. Since young residents are willing to rent second-hand housing, it is supposed that this factor has a negative impact on the vacant house ratio regardless of the difference in UEA categories. Assessed wooden housing value per area (VALUE): This demonstrates wooden housing value per floor area. According to Japanese law, the value decreases if buildings are old and not maintained well. In Japan, property tax is decided by the value of the property. Therefore, if the value is high, housings in the city are well maintained, and we hypothesize that the vacant house ratio would be low in every UEA category. Ratio of housings adjacent to road of less than 4m width (ROAD): This factor is the ratio of housings adjacent to a road whose width is less than 4 meters. As illustrated in the literature review, vacant housing tend to be adjacent to narrower roads. It is presumed that narrow road has a positive effect on vacant house ratio regardless of UEA categories, and that the relationship differs Urban and Regional Planning Review Vol. 4, 2017 | 257 depending on the types of UEAs. Ratio of wooden structure (WOOD): This is the ratio of non-fireproof wooden housings. As mentioned in the literature review, it would have a positive impact on vacant house ratio. Also, this factor could be related to easiness of regenerating cities because wooden housings are relatively easier to demolish than those made out of concrete or steel. Mean floor area per housing (FL): This indicator shows mean floor area per housing in a city. We hypothesize that if the mean floor area is substantially low, the vacant house ratio would be high, because small housings would not meet the demand of purchasers who need to accommodate their families. Ratio of housings within 500m from day care facilities (CARE): This is the number of housings located within 500 meters of day care facilities, and one of the factors indicating a pleasant dwelling environment. We tried other factors on dwelling environment such as clinic, kindergarten, and community facility, but day care facility is the best factor in fitting the model. It is hypothesized that low ratio of housings within 500m from day care facilities are related to high vacant house ratio. Number of grocery stores per habitable area (GR): This factor also indicates pleasant dwelling environment. Since it expresses the density of grocery stores, it could be thought that the high value of this factor is associated with a high standard of accessibility. We hypothesized that in “small core UEAs” and “independent cities,” local residents usually use their car, and this factor is not statistically significant. Table 1 shows the descriptive statistics of data acquired. There are 7 explanatory variables, whose data are retrieved from government offices in Japan.

Table 1. Descriptive statistics

Variable (N=771)abbreviationunitmean standard deviation MedianMinimum Maximum vacant house ratio a VH -0.0720.0390.0640.0020.227 Age (20-39) ratio b AGE -0.1980.0310.1990.1100.317 Assessed wooden housing VALUE 1,000 yen /㎡ 18.6815.30017.8527.74934.038 value per floorarea c

Adjacent to 4m road ratio a ROAD-0.3880.1370.3720.0120.751 Wooden structure ratio a WOOD-0.3780.1920.3630.0010.871 Mean floor area a FL ㎡ 113.46025.090110.22062.020221.670 Day care within 500m ratio a CARE-0.2320.1500.1980.0000.768 Grocery per Habitable area d GR store / 100㎡ 4.1024.5262.4570.21042.077 Source: (a) Ministry of Land, Infrastructure, Transport, and Tourism(MILT) (2013) Housing and Land Survey [25]; (b) Ministry of Internal Affairs and Communications(MIAC) (2015) National Population Census [26]; (c) MIAC (2015) Survey on property value [27]; (d) MIAC (2014) Economic Census [28]

3.3. Generalized linear mixed models for vacant house ratio in each category The analysis estimating the fixed effects of vacant house ratio is well fitted for a generalized linear mixed model whose link function is a logit model, because the number of vacant housing is count data with the limitation of the total number of housings. Since the ratio of vacant housing takes the value of [0, 1], a logistic function should be a suitable model for estimating fixed effects for the vacant house ratio (Kubo 2012) [29]. The analysis is a multivariate regression which is well fitted when the dependent variable is binomial. Hence, a generalized linear mixed model (GLMM) is applied to the presence of vacant housing. For instance, Morckel (2013) [30] Urban and Regional Planning Review Vol. 4, 2017 | 258 estimated the factors of housing abandonment in Columbus and Youngstown, the United States, by logistic regression model, which is one kind of GLMM.

Let “푖” denote the identification of a city, and “푦푖” denotes the amount of vacant housing in 푖, and “푁푖” is the total amount of housings in 푖. The expected ratio of vacant housing is expressed by binomial distribution. 푁푖 푦푖 푁푖 − 푦푖, Pr(푦푖|푁푖, 푝푖) = ( ) 푝푖 (1 − 푝푖) 푦푖 Pr (푦푖 |푁푖, 푝푖): occurrence probability of vacant housing out of total amount of housings in 푖, 푝푖 : occurrence probability of vacancy in each housing in 푖. In the case of vacant house ratio, the distribution tends to be overdispersed. Therefore, we included a random variable which follows Gaussian distribution. In the model, logit link function is as follows. 푝푖 logit(푝푖) = log ( ) = 훽0 + ∑ 훽푘 푥푘,푖 + 푟푖 1 − 푝푖 푘 훽0: constant term, 훽푘: coefficients to be estimated, 푥푘푖 socio-economic and built-environment factors (k = 1, 2, … ,7), 푟푖: random effect term in 푖 The expected ratio of vacant housing is the following. 1 푝푖(푦푖|{훽푘}, 푟푖) = , 푧푖 = 훽0 + ∑ 훽푘 푥푘,푖 + 푟푖 1 + 푒푥푝(−푧푖) 푘

1 2 − 푟푖 푔 (푟 |푠) = (2휋푠2) 2 exp(− ) 푖 푖 2푠2 Then, estimated coefficients 훽̂푘 are calculated such that a log-likelihood function is maximized. Likelihood and log-likelihood functions are as follows. ∞ 퐿푖 = ∫ 푝푖(푦푖|{훽푘}, 푟푖)푔푖(푟푖|푠)d푟푖 −∞

퐿({훽푘}, 푠) = ∏ 퐿푖 푖

훽̂푘 = arg max log 퐿 ({훽푘, 푠}) 훽 In the study, the number of explanatory variables is decided by Akaike’s Information Criterion (AIC). For all the categories, AIC was not the lowest in the models in case all the explanatory variables are contained. However, because the gap of AIC between the best model and the model containing all the explanatory variables was less than 4.0, we applied the model with all the variables for all categories, because it is understandable to compare each variable.

4. Result 4.1. Fundamental statistics of each category After classifying cities into each category, we obtained 197 samples of the three major UEAs category, 318 large core UEAs, 157 small core UEAs, and 99 independent cities respectively. The ratio of vacant house is different depending on categories (Figure 3), whose means are statistically significant by ANOVA. The three major UEAs have the lowest mean vacant house ratio among the 4 categories, while independent cities have the highest. Spatial distribution is attached in Appendix 1 along with the distribution of vacant house ratio (Appendix 2). Urban and Regional Planning Review Vol. 4, 2017 | 259

Figure 3. Box plot of vacant house ratio and each category defined by UEAs

There are some trends on the mean values of explanatory variables. All of the mean values are statistically significant by ANOVA (Table 2). It is shown that a strong hierarchy exists from the “three major UEAs” to “independent cities.” Age (20-39) ratio, assessed wooden housing value, house ratio to day care within 500m, and grocery store density decrease from “three major UEAs” to “independent cities.” On the other hand, house ratio adjacent to less than 4m road, wooden structure ratio, and mean floor area increase when the scale of UEA is larger. Looking at variables individually, age (20-39) ratio and mean floor area do not change dramatically compared to the other variables. It could be considered that such variables are not affected by geographical difference. However, the values of wooden structure ratio and house ratio to day care within 500m in three major UEAs are twice as big as in independent cities, and the value of grocery store density in three major cities is more than quadruple that in independent cities. Such variables would be strongly affected by geographical difference. Focusing on standard deviation (S.D.) on “three major UEAs,” S.D. is smaller than the other categories except for assessed wooden housing

Table 2. Result of ANOVA

Three major Large core Small core Independent (upper) F-ratio Variable UEAs UEAs UEAs cities (lower) p-value n=197 n=318 n=157 n=99 mean 0.221 0.202 0.184 0.166 124.500 AGE S.D. 0.025 0.025 0.027 0.028 0.000 mean 24.511 18.270 15.510 13.430 267.583 VALUE S.D. 3.716 4.098 3.313 2.901 0.000 mean 0.323 0.375 0.414 0.516 56.327 ROAD S.D. 0.099 0.130 0.136 0.135 0.000 mean 0.243 0.380 0.406 0.597 109.640 WOOD S.D. 0.132 0.164 0.186 0.156 0.000 mean 93.588 116.784 121.762 129.132 80.803 FL S.D. 18.464 23.236 21.111 24.937 0.000 mean 0.336 0.231 0.169 0.127 72.502 CARE S.D. 0.186 0.125 0.086 0.080 0.000 mean 8.358 3.161 2.124 1.795 118.334 GR S.D. 6.295 2.815 1.545 1.149 0.000

Urban and Regional Planning Review Vol. 4, 2017 | 260 value, house ratio to day care within 500m, and grocery store density. Since the center areas of “three major UEAs” are highly urbanized, high property value and high quality of accessibility would be retained in such areas. When looking at S.D. on “independent cities,” house ratio adjacent to a road of less than 4m and mean floor area are dispersed compared to the other categories. Supposedly, “independent cities” would include densely populated districts, so that housings sometimes tend to be adjacent to narrow roads and smaller than ones in new towns. 4.2. Characteristics of each category The characteristics of each category are clarified by estimated coefficients of logistic regression models (Table 3). The following are the descriptions of categories. Category I: Three major UEAs This category represents a large-scaled urban area, so it includes both city centers and city suburbs. AGE, VALUE, and CARE have a negative impact on vacant house ratio. Especially, when AGE increases one additional percent, vacant house ratio decreases by , which means approximately 3.7% decrease of vacant house ratio by 1% increase of 20-39 age population. Wooden housing value per housing is also a significant factor related to the ratio of vacant housing. Although this variable is also statistically significant in the other categories, “three major UEAs” has the lowest odds ratio among categories. It could be thought that UEAs such as Tokyo, Osaka, and Nagoya metropolitan areas have a huge area and there is a massive socio-economic gap between the center and suburb. Thus, a gradation of socio- economic gap would be generated between areas. ROAD, WOOD, FL, and GR seem to be related to high vacant house ratio, although these variables are not statistically significant. This means built-environmental factors are not related to vacant house ratio in this category. Category II: Large core UEAs This category includes many medium-scaled cities located in rural areas. In this category, only ROAD is positively related to the ratio of vacant house. Other variables such as AGE, VALUE, FL, and GR have a negative impact on vacant house ratio. Especially, the population ratio of 30 – 39 age is the lowest among the 4 categories, whose odds ratio is 0.925. Only in this category, the number of grocery stores per area is statistically significant among categories. This means that an area with a high ratio of vacant housing sometimes does not have a sufficient number of grocery stores, which reduces accessibility in the area. Category III: Small core UEAs This category is a set of small-scaled urban employment areas, which means that these are basically located in rural areas and the core of the UEA does not have a strong attractive force for workers. In the “small core UEAs,” 6 variables out of 7 are statistically significant. AGE, VALUE, FL, and CARE have a negative impact on vacant house ratio. On the other hand, high values of ROAD and WOOD are related to a high ratio of vacant housing. Especially the coefficient of CARE is the highest among the 4 categories. High housing ratio within 500m of day care facility means a high potential of accessibility and it leads to a low ratio of vacant housing. The ratio of housings made of wood is statistically significant only in this category. It is considered that the area, where plenty of wooden structured housings are built, would be an old densely populated district. Such an area would have narrow roads and housings with small floor area, which coincides with the coefficients of variables, such as ROAD and FL. Category IV: Independent cities Cities in this category do not compose an urban employment area. Such independent cities are prone to be dispersed in rural areas. AGE, VALUE, and FL have a negative effect on the ratio Urban and Regional Planning Review Vol. 4, 2017 | 261

Table 3. Result of GLMM

Three major UEAs Variable Coef.Std. errorz value Pr ( > | z | ) odds ratio per unit intercept -1.697 0.667 -2.5440.011 * - AGE -3.723 1.326 -2.8080.005 ** 1% → 0.963 VALUE -0.056 0.013 -4.4300.000 *** 1000yen/㎡→ 0.946 ROAD 0.438 0.360 1.2160.224 - WOOD 0.708 0.395 1.7920.073 - FL 0.005 0.003 1.4470.148 - CARE -0.396 0.190 -2.0830.037 * 1% → 0.996 GR 0.008 0.007 1.1760.239 - N 197 AIC 1304 Scale parameter 0.358 in mixing distribution

Large core UEAs Variable Coef.Std. errorz value Pr ( > | z | ) odds ratio per unit intercept -0.913 0.220 -4.1440.000 *** - AGE -7.825 0.772 -10.1370.000 *** 1% → 0.925 VALUE -0.023 0.005 -5.0120.000 *** 1000yen/㎡→ 0.977 ROAD 1.334 0.151 8.8340.000 *** 1% → 1.013 WOOD 0.177 0.133 1.3310.183 - FL -0.002 0.001 -2.6310.009 ** ㎡→ 0.998 CARE -0.105 0.135 -0.7760.437 - GR -0.018 0.006 -2.9370.003 ** store/100㎡→ 0.982 N 318 AIC 1881 Scale parameter 0.246 in mixing distribution

Small core UEAs Variable Coef.Std. errorz value Pr ( > | z | ) odds ratio per unit intercept -1.024 0.245 -4.1860.000 *** - AGE -4.348 1.022 -4.2530.000 *** 1% → 0.957 VALUE -0.036 0.009 -4.1280.000 *** 1000yen/㎡→ 0.965 ROAD 0.782 0.218 3.5890.000 *** 1% → 1.008 WOOD 0.464 0.147 3.1510.002 ** 1% → 1.005 FL -0.003 0.001 -3.1210.002 ** ㎡→ 0.997 CARE -0.911 0.245 -3.7240.000 *** 1% → 0.991 GR 0.009 0.015 0.5970.550 - N 157 AIC 901 Scale parameter 0.239 in mixing distribution

Independent cities Variable Coef.Std. errorz value Pr ( > | z | ) odds ratio per unit intercept -0.163 0.284 -0.5730.567 - AGE -7.541 1.418 -5.3200.000 *** 1% → 0.927 VALUE -0.050 0.014 -3.6770.000 *** 1000yen/㎡→ 0.951 ROAD 1.416 0.262 5.4060.000 *** 1% → 1.014 WOOD -0.209 0.212 -0.9840.325 - FL -0.004 0.001 -3.6940.000 *** ㎡→ 0.996 CARE -0.357 0.412 -0.8680.385 - GR -0.009 0.028 -0.3250.745 - N 99 AIC 558 Scale parameter 0.242 in mixing distribution statistically significant: ***0.1%, **1%, *5%

Urban and Regional Planning Review Vol. 4, 2017 | 262 of vacant housing. Only ROAD influences the high ratio of vacant housing. ROAD marks the highest value among the 4 categories. This means that there are many dense residential areas located, where housings tend to be vacant due to inconvenience. Mean floor area is the lowest among the 4 categories, although we could not observe a large numerical gap between the 4 categories. Still, the increase of 10 m2 of mean floor area leads to 4.0% of the decrease of vacant house ratio, which indicates that mean floor area has a huge impact on the vacant house ratio. Perhaps, local residents are capable of affording a housing with a large floor area in this category, and the demand for housings with small floor area does not exist to such an extent.

5. Conclusion We analyzed the regional differences in socio-economic and built-environment factors of the vacant house ratio. Since the vacant house ratio represents the spatial situation of shrinking cities, the factors of the ratio indicate the extent of spatial urban shrinkage. Employing a generalized linear mixed model, coefficients of the factors of vacant house ratio were estimated. In this study, cities were categorized according to urban employment areas, which stipulates the urban boundaries by economic activities. Consequently, the following findings were clarified. First, strong trends of vacant house ratio and its factors among categories were found. “Three major UEAs” marked the lowest vacant house ratio and values of ROAD, WOOD, and FL. On the other hand, “independent cities” had the highest values of those variables. The values of AGE, VALUE, CARE, and GR showed opposite results to ROAD, WOOD, and FL. Yet, the larger the scale of UEAs was, the higher or lower the variables changed, and there was no transition of such order. This means that the larger scale of UEAs is inclined to have lower vacant house ratios, and that they retain a similar trend of socio-economic and built-environment factors. Second, socio-economic factors affecting vacant house ratio in each category were statistically significant in all the UEA categories. Age composition ratio of 20-39 years old was negatively related to the vacant house ratio for all the categories. Especially, “large core UEAs” and “independent cities” had high values of coefficients, -3.723 and -7.825, respectively. It is thought that segregation of age groups occurs and elderly people tend to live in the areas where urban shrinkage is happening. Wooden housing value per floor area also has a negative effect on vacant house ratio for every category. “Three major UEAs” had the lowest value of the variable (-0.056) among the 4 categories. The higher the value is, the more local residents tend to live in well maintained housing. Thus, the low value of wooden housing indicates that neighborhoods in the city have deteriorated, and it is attributed to the high ratio of vacant housing. Third, the statistical significance of built-environment factors also had a clear relationship with vacant house ratio, but the influences depended on the categories. The ratio of housings adjacent to a road of less than 4m was positive except in the “three major UEAs.” We suppose that this variable would be compatible with wooden house ratio and mean floor area. It is considered that the districts, where urban shrinkage is occurring, have narrow roads and wooden housings with small floor area. “Large core UEAs,”, “small core UEAs,” and “independent cities” would have such shrinking neighborhoods, not the case of the “three major UEAs.” Especially, “independent cities” had a strong tendency of containing such shrinking neighborhoods, because values of both ROAD and FL were the worst among the 4 categories. House ratio within 500m to day care facility is the proxy variable indicating pleasant dwelling environment. This variable was negatively significant for the “three major UEAs” and “small core UEAs.” We think that in those areas, the ratio of vacant housing is high if day care facilities are less accessible. Grocery store Urban and Regional Planning Review Vol. 4, 2017 | 263 density was negatively related to vacant house ratio only in “large core UEAs.” In such UEAs, high ratio of vacant housing occurs in the area where the number of grocery stores is relatively low. Considering the result of CARE and GR, an area, where accessibility to facilities with daily necessities is low, tends to suffer from a high level of vacancy. This finding follows previous research (Katayama et al. 2006) [21]. However, only “independent cities” did not show any relationship with vacant house ratio. Supposedly, local residents usually use their car for daily journeys, and they would not consider accessibility so much. In order to prepare for the impacts of spatial urban shrinkage, it is important to know where future areas of spatial urban shrinkage are. For each UEA category, factors driving cities to shrink would be different. In the “three major UEAs,” spatial urban shrinkage is not advancing as much as in the other categories. However, as a proactive action, we could focus on the area where a majority of elderly people live, and such an area has a high ratio of vacant housing. Also, property value would be one of the predictors of urban shrinkage. Areas where the property value is low, tend to have a high ratio of vacant housing. In “large core UEAs,” age composition and property value are also important factors in urban shrinkage. Density of grocery stores is one of the important factors because only this category is negatively correlated to grocery store density. This means that commercial inconvenience is associated with the high ratio of vacant housing. In “small core UEAs,” in addition to the factors of AGE and VALUE, built-environment factors are important. We should focus on spatial differences such as narrow roads, wooden structured housings, and the size of housings. Also, dwelling environment such as the proximity to day care facilities is sensitive to an increase in vacant housing in this category. In “independent cities”, focusing factors are similar to those of “small core UEAs.” However, current situations of built- environment are more sensitive in “independent cities,” so cities in which there are such neighborhoods of narrow roads and small housings, should be careful of spatial urban shrinkage. There are several issues to be discussed. First, considering variables associated with accessibility to public transportation would be beneficial to this research. Some papers argue that the distance to a train station is negatively related to an increase in vacant housing (Sakamoto and Yokohari 2016) [17]. However, due to the limitation of available data, we could not define variables that indicate accessibility to public transportation. Second, analyses on property value should be conducted further. According to Japanese law, if a property owner leaves a house vacant, the land property tax with buildings is cheaper than the tax without buildings. Therefore, if we consider land value instead of housing value, vacant house ratio would vary depending on region. However, we had multicollinearity between land value and grocery density. Also, we supposed that the analysis unit would be too large to observe such an effect. Consequently, we only considered property value for housing, excluding property value for land. Third, geographical differences of the factors within a city should be considered. In this study, the unit of analysis was a city, so that it was difficult to grasp the differences of the factors among neighborhoods. Although data on vacant housing is available city by city, we are able to divide it proportionately into neighborhoods, which enables us to analyze spatial urban shrinkage using each neighborhood unit.

Acknowledgments We are grateful to the two anonymous referees and participants at the 2017 spring seminar of the urban operations research for their helpful comments.

Urban and Regional Planning Review Vol. 4, 2017 | 264

Appendix 1. UEA category in Japan

Urban and Regional Planning Review Vol. 4, 2017 | 265

Appendix 2. Vacant house ratio in Japan

Urban and Regional Planning Review Vol. 4, 2017 | 266

References [1] Hollander, J. B., Pallagst, K., Schwarz, T., & Popper, F. Planning shrinking cities. Progress in Planning, 72(4), 223-232, 2009 [2] National Institute of Population and Social Security Research (NIPSSR), Estimated population in Japan, as of March, 2012 (in Japanese), 2012 [3] Asami, Y. Urban issues of vacant houses (in Japanese), Evaluation, 52, 1-6, 2014 [4] Haase, D. Urban ecology of shrinking cities: An unrecognized opportunity? Nature and Culture, 3(1), 1-8, 2008 [5] Rieniets, T. Shrinking cities: Causes and effects of urban population losses in the twentieth century. Nature and Culture, 4(3), 231-254, 2009 [6] Spelman, W. Abandoned buildings: Magnets for crime? Journal of Criminal Justice, 21(5), 481-495, 1993 [7] Wiechmann, T., & Pallagst, K. M. Urban shrinkage in Germany and the USA: A comparison of transformation patterns and local strategies. International Journal of Urban and Regional Research, 36(2), 261-280, 2012 [8] Schilling, J., & Logan, J. Greening the rust belt: A green infrastructure model for right sizing America's shrinking cities. Journal of the American Planning Association, 74(4), 451-466, 2008 [9] Martinez-Fernandez, C., Weyman, T., Fol, S., Audirac, I., Cunningham-Sabot, E., Wiechmann, T., & Yahagi, H. Shrinking cities in Australia, Japan, Europe and the USA: From a global process to local policy responses. Progress in Planning, 105, 1-48, 2016 [10] Rink, D., Haase, A., Grossmann, K., Couch, C., & Cocks, M. From long-term shrinkage to re-growth? The urban development trajectories of Liverpool and Leipzig. Built Environment, 38(2), 162-178, 2012 [11] Haase, A., Herfert, G., Kabisch, S., & Steinführer, A. Reurbanizing Leipzig (Germany): Context conditions and residential actors (2000–2007). European Planning Studies, 20(7), 1173-1196, 2012 [12] Bontje, M. Facing the challenge of shrinking cities in East Germany: The case of Leipzig. GeoJournal, 61(1), 13-21, 2005 [13] Grossmann, K., Bontje, M., Haase, A., & Mykhnenko, V. Shrinking cities: Notes for the further research agenda. Cities, 35, 221-225, 2013 [14] Haase, D., Haase, A., & Rink, D. Conceptualizing the nexus between urban shrinkage and ecosystem services. Landscape and Urban Planning, 132, 159-169, 2014 [15] Ujihara, T., Abe, H., Murata, N., & Washio, N. Reality of “spongy urban area” in local city: Based on situation of development and loss, vacant house. Journal of Japan Society of Civil Engineers, Ser. D3 (Infrastructure Planning and Management), 72(1), 62-72, 2016 (in Japanese) [16] Ujihara, T., Abe, H., & Nonak, S. Reality of spongy urban area based on state analysis of residential districts. Journal of the City Planning Institute of Japan, 51(3), 466-473, 2016 (in Japanese) [17] Sakamoto, K., & Yokohari, M. Dynamic characteristics of emerging vacant houses and abandoned lands in residential neighborhoods: A case study in the residential neighborhoods of , Tochigi. Journal of the City Planning Institute of Japan, 51(3), 854-859, 2016 (in Japanese) [18] Abe, S., Nakagawa, D., Matsunaka, R., & Oba, T. Analysis of factors influencing the change of underused land area and transformation from underused land to residential land in central areas of local cities. Journal of Japan Society of Civil Engineers, Ser. D3 (Infrastructure Planning and Management), 68(5), I_467-I_477, 2012 (in Japanese) [19] Kanamori, Y., Ariga, T., & Matsuhashi, K. Factor analysis and estimation of rate of vacant dwellings. Journal of the City Planning Institute of Japan, 50(3), 1017-1024, 2015 (in Japanese) [20] Takeshima, H., Omi, T., & Ishizaka, K. Study on the prediction of empty house and vacant lot occurrence in suburban residential area, in case of city. AIJ Journal of Technology and Design, 10(20), 325-330, 2004 (in Japanese) [21] Katayama, N., Kaido, K., Murakami, S., & Maeda, Y. The study on the sustainability of suburban housing areas focusing on the conditions of blank lots and empty houses: Case study at Kani City and Tajimi City which are located in Nagoya region. Urban Housing Sciences, 2006(55), 70-75, 2006 (in Japanese) [22] Nakai, S., Kana, K., & Sakuma, Y. Research on actual condition of existing vacant houses and study on possibility of utilizing them as "spaces for disaster prevention" in the densely built-up area: The case of Tsuruhashi Area in Osaka City. Journal of the City Planning Institute of Japan, 47(3), 1063-1068, 2012 (in Japanese) [23] Ai, H. Indexing of living environment attracts young and productive age generations: An analysis based on local districts within the Greater Tokyo Metropolitan Area. Journal of the City Planning Institute of Japan, 51(3), 860-866, 2016 (in Japanese) Urban and Regional Planning Review Vol. 4, 2017 | 267

[24] Kanemoto, Y., & Kurima, R. Urban employment areas: Defining Japanese metropolitan areas and constructing the statistical database for them. GIS-based Studies in the Humanities and Social Sciences, 85-97, 2005 [25] Ministry of Land, Infrastructure, Transport, and Tourism. 2013 Housing and Land Survey, http:// www.e-stat.go.jp/SG1/estat/GL08020101.do?_toGL08020101_&tstatCode=000001063455& requestSender=search, (accessed 2017-11-17). [26] Ministry of Internal Affairs and Communications. 2015 National Population Census, http://www.e-st at.go.jp/SG1/estat/GL08020101.do?_toGL08020101_&tstatCode=000001080615&requestSender= search, (accessed 2017-11-17). [27] Ministry of Internal Affairs and Communications. 2015 Survey on Property Value, http://www.sou mu.go.jp/main_sosiki/jichi_zeisei/czaisei/czaisei_seido/ichiran08_h27_00.html, (accessed 2017-11-17). [28] Ministry of Internal Affairs and Communications. 2014 Economic Census for Business Frame, http:// www.e-stat.go.jp/SG1/estat/GL08020101.do?_toGL08020101_&tstatCode=000001072573&request Sender=search, (accessed 2017-11-17). [29] Kubo, T. Introduction to Modeling of statistics for data analysis – generalized linear model, hierarchical Bayesian model, MCMC, Iwanami Shoten, ISBN: 400006973X, 144-167, 2012 (in Japanese) [30] Morckel, V. C. Empty neighborhoods: Using constructs to predict the probability of housing abandonment. Housing Policy Debate, 23(3), 469-496, 2013