Scholarly article on topic 'Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo'

Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo Academic research paper on "Social and economic geography"

CC BY
0
0
Share paper
Academic journal
Regional Science and Urban Economics
OECD Field of science
Keywords
{"Female labor force participation" / "Commute times" / "Spatial patterns" / "Spatial statistics" / Tokyo}

Abstract of research paper on Social and economic geography, author of scientific article — Mizuki Kawabata, Yukiko Abe

Abstract We explore intra-metropolitan spatial patterns of female labor force participation, and examine how they relate to commute times in Tokyo. The spatial patterns differ markedly by marital status and the presence of children. For married mothers, the spatial clusters of low participation and regular employment rates are largely located in the inner suburbs, many of which overlap with the spatial clusters of long male commute times. The spatial regression results indicate that for married mothers, a longer commute time is significantly associated with lower participation and regular employment rates, while for unmarried and childless married women, these associations are mostly nonsignificant. Among married mothers, the magnitude of the negative associations is greater for college graduates than for those with a high school education or less, suggesting that highly educated mothers are especially sensitive to commute times. We argue that the spatial transportation constraint intensifies the household division of labor, resulting in unique patterns for married mothers.

Academic research paper on topic "Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo"

Author's Accepted Manuscript

Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo

Mizuki Kawabata, Yukiko Abe

www.elsevier.com

PII: S0166-0462(17)30187-4

DOI: https://doi.org/10.1016/j .regsciurbeco.2017.11.003

Reference: REGEC3312

To appear in: Regional Science and Urban Economics

Received date: 26 May 2017 Revised date: 29 S eptember 2017 Accepted date: 1 November 2017

Cite this article as: Mizuki Kawabata and Yukiko Abe, Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo, Regional Science and Urban Economics,

https://doi.org/10.1016/j.regsciurbeco.2017.11.003

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Intra-metropolitan spatial patterns of female labor force participation and commute times in Tokyo^

Mizuki Kawabataa*, Yukiko Abeb

aFaculty of Economics, Keio University, Tokyo 108-8345, Japan

bGraduate School of Economics and Business Administration, Hokkaido University, Hokkaido 060-0809 Japan

Corresponding author. mizuki@econ.keio.ac.jp

Abstract

We explore intra-metropolitan spatial patterns of female labor force participation, and examine how they relate to commute times in Tokyo. The spatial patterns differ markedly by marital status and the presence of children. For married mothers, the spatial clusters of low participation and regular employment rates are largely located in the inner suburbs, many of which overlap with the spatial clusters of long male commute times. The spatial regression results indicate that for married mothers, a longer commute time is significantly associated with lower participation and regular employment rates, while for unmarried and childless married women, these associations are mostly nonsignificant. Among married mothers, the magnitude of the negative associations is greater for college graduates than for those with a high school education

* This research was supported by JSPS KAKENHI Grand Number JP16K13363, JP26590045, and JP15H03358.

or less, suggesting that highly educated mothers are especially sensitive to commute times. We argue that the spatial transportation constraint intensifies the household division of labor, resulting in unique patterns for married mothers.

JEL classification J21, R12, C31

Keywords

Female labor force participation; commute times; spatial patterns; spatial statistics; Tokyo

1 Introduction

Geographic disparities in female participation in the labor market have drawn increased attention in urban and labor economics. Recent studies show that female participation varies considerably across regions, such as metropolitan areas, counties, and prefectures (Abe, 2011; Fogli and Veldkamp, 2011; Black et al., 2014); the regional variations in female participation are much greater than those in male participation (Abe, 2013). Some have argued that commute time is an important factor in the large regional variations in female participation (Abe, 2011; Black et al., 2014). Using inter-metropolitan data on commute times in the US, Black et al. (2014) find negative associations between commute times and labor force participation of married women. However, empirical evidence of the role of commute times in female participation is limited. In particular, the intra-metropolitan geographic disparities in female

participation and their associations with commute times have not been explored using data for small geographical units.

In this study, we aim to shed new light on geographical disparities in female participation by examining intra-metropolitan spatial patterns of female participation and studying how they relate to commute times. Specifically, we ask the following three questions: (1) Are there specific intra-metropolitan spatial patterns of female participation? (2) Are the spatial patterns of female participation related to commute times? (3) Do the spatial patterns of female participation and their associations with commute times differ by marital status and the presence of children? We examine these questions for women aged 25-54 in Tokyo, the world's most populous metropolitan area with long commute times.

We pay particular attention to the importance of commuting for women, since female commutes are shorter than male commutes (this point is reviewed in the next section). The shorter female commutes suggest that making long commutes is not feasible for many women, because they either face greater spatial constraints or have stronger preferences for short commutes. In a large metropolitan area like Tokyo, commute times differ considerably by residential location. In the suburbs, many people spend a long time commuting to work in the central business district (CBD) via heavily congested public transportation. It is conceivable that such long commutes impede women, especially women with children, from participating in the labor market.

Our contribution to the literature is three-fold. First, we examine intra-metropolitan spatial patterns of female participation by identifying spatial clusters of high and low levels of female participation. The levels of female participation are unlikely to be evenly distributed within a metropolitan area; they may have distinct spatial patterns vis-à-vis urban spatial structure. For example, there may be disproportionately more married women with children who need more housing space living in the suburbs, where housing costs are modest. Such suburban mothers are more likely to be housewives or to be working locally rather than working full time and commuting to the CBD. By contrast, households in which both spouses work in the CBD have greater incentives to live nearby. In order to examine the spatial patterns, we employ geographic information systems (GIS) and spatial statistics—Global Moran's I and Getis-Ord Gi*.

Second, we use municipal-level data, which are more spatially disaggregated than the inter-metropolitan data used in previous research (Black et al., 2014). The use of a smaller geographical unit (in this study, the municipality) enables us to examine intra-metropolitan geographic disparities in female participation and commute times as well as their spatial patterns. Moreover, analyzing municipal data within a metropolitan area allows us to study the role of commute times in an environment where relocation (i.e., sorting by choosing residential location) is more realistic. For example, it is unlikely that people would migrate from New York City to Minneapolis, where commuting costs are lower; however, it is likely that people would choose from locations within the Tokyo metropolitan area (e.g., Kamakura1 versus central Tokyo) when considering commuting distance and the housing costs in each place.

Finally, we examine three groups of women aged 25-54 (unmarried, married without children, and married with children), and three rates for female participation (labor force participation, regular employment, and part-time employment). A comparison of multiple demographic groups of women and these three types of female participation is rare in the literature.

We find considerable intra-metropolitan disparities in female participation. The spatial patterns of labor force participation and regular employment rates differ markedly by marital status and the presence of children. Married mothers have unique spatial patterns of participation and regular employment rates. These rates are negatively and significantly associated with male commute times, while the associations are mostly nonsignificant for unmarried and childless married women. Among married mothers, the magnitude of the negative associations is greater for college graduates than for those with a high school education or less, suggesting that highly educated mothers are especially sensitive to commute times.

The next section reviews the related literature. Section 3 explains the methods, study area, and data. Section 4 presents the findings, and Section 5 concludes.

1 Kamakura is a popular suburban city in Kanagawa Prefecture.

2 In this paper, female participation includes both regular and part-time employment.

2 Related literature 2.1 Female commuting

Female commuting is shorter than male commuting. Studies consistently find this gender difference across regions and countries (e.g., Madden, 1981; Gordon et al., 1989; Hjorthol, 2000; Lee and McDonald, 2003; Crane, 2007; Roberts et al., 2011; Neto et al., 2015). The gender commute gap is large for parents: fathers have long commutes, whereas mothers have short ones (Madden, 1981; McLafferty and Preston, 1997; Hjorthol, 2000).

Researchers examine various reasons why women work closer to home. The primary explanation is the household responsibility hypothesis (HRH) based on the household division of labor in which women shoulder more housework and childcare than men (e.g., Madden, 1981; Johnston-Anumonwo, 1992; McLafferty and Preston, 1997; Turner and Neimeier, 1997). Empirical evidence on the HRH is mixed; some results support it (Turner and Niemeier, 1997; Neto et al., 2015), while others show little or blended levels of support (Gordon et al., 1989; Shingell and Lillydahl, 1986). Turner and Neimeier (1997) critically review the HRH.

A similar but more specific explanation for women's shorter commutes is based on family support trips, such as shopping and transporting children to and from childcare centers or schools. Women make more family support trips than men, and are more likely to combine non-work trips with work trips (Hanson and Hanson, 1981; Hanson and Johnston, 1985; McGuckin and Murakami, 1999; Hjorthol, 2000). The gender difference in trip-chaining behavior is particularly noticeable when children are present. McGuckin and Murakami (1999) find that women with young children are far more likely than men (and women without children) to make multiple stops linked to their commutes. Boarnet and Hsu (2015) discover that within households with children, women make considerably more chauffeuring trips than men, while the non-work trips of men and women do not differ much when they do not have children. These findings suggest that mothers are more sensitive to commute times than men or women without children.

Evidence also suggests that long commutes are especially burdensome for women with young children (Rouwendal, 1999). When traveling with infants and toddlers, mobility becomes limited, and spatial constraints increase in severity. Commuting may involve unexpected trips to and from the hospital when a child falls ill. Roberts et al. (2011) demonstrate that commute time adversely influences the psychological health of women, especially those with preschool-aged children. They find no such negative effects for single men and women without children, those working with flexible schedules, or those who rely on their partners for childcare. Their findings imply that women who do more child-rearing and housework face greater psychological barriers for long commutes than those who have less housework. These psychological barriers may prevent women from working jobs that require long

commutes.

2.2 Commuting time, female labor force participation, and urban spatial structure

Basic urban models indicate that housing and commuting costs are closely related to urban spatial structure. In the monocentric city model formulated by Alonso (1964), Muth (1969), and Mills (1972), there is a tradeoff between housing and commuting costs. The farther the distance from the CBD, the lower the housing prices, and the higher the commuting costs. This trend is found in metropolitan areas around the world, although actual patterns are more complex.

The monocentric city model has three implications for spatial patterns of labor supply by married women. First, to secure enough residential space, families may

3 The hypotheses are based on the household division of labor (i.e., labor supply factors). Other explanations for women's shorter commutes come from labor demand. For example, women tend to have lower wages, higher job turnover rates, and shorter work hours, which reduce economic incentives for making long commutes (Madden, 1981; White, 1986; MacDonald, 1999). In another instance, differences in the spatial distributions of jobs suitable for women and men could explain gender disparities in commuting: female jobs are more dispersed, whereas male jobs are more concentrated (Hanson and Johnston, 1985; MacDonald, 1999).

decide to live farther away from the CBD. Consistent with this hypothesis, Madden (1981) finds that compared to unmarried women, married women live in larger homes and farther from the city center. Suburban living makes it costly for both spouses commute to the CBD, especially in large metropolitan areas. Of people who commute to the Tokyo ward area (referred to as the CBD in this study), half spend 60 minutes or longer, and 15% spend 90 minutes or longer for a one-way commute (TMRTPC, 2010). The likely allocation between spouses is that the one with the comparative advantage for market work (i.e., the husband) would commute to the CBD, while the other (i.e., the wife) would either stay home or work locally. When this division within household occurs, the husband spends a long time commuting, while the labor market participation of the wife is low. McLafferty and Preston (1997) find that married men, more than unmarried men, spend longer times commuting, and argue that the unequal division of labor in which women do more housework frees up time for men to spend on long commutes.

Second, for households with children, the need for space and parents' time value at home are both high. Households with children may also value residential amenities that benefit their children, such as schools and neighborhoods (Brasington, forthcoming; Gamsu, 2016). In the suburbs, a division of the market and household work between parents—the father commutes to the CBD and the mother stays home or works locally—is more likely to take place.

Finally, dual-earner households in which both spouses work in the CBD are willing to pay both spouses' sum of time cost to pay for locations close to the CBD; therefore, such households are likely to live near the CBD and have shorter commutes (Abe 2011). Hjorthol (2000) reports that in Oslo, the commuting distance of married women and men is the shortest in the central parts of the metropolitan area.

These implications from the monocentric city model suggest that costly commuting and lower housing prices in the suburbs induce a sharp division of labor between husbands and wives in the suburbs, especially for couples with children.

3 Methods

3.1 Spatial patterns of female participation

We examine the spatial patterns of female participation by calculating the Global Moran's I and Getis-Ord Gi statistics. Global Moran's I (Moran, 1950) is a global measure of spatial autocorrelation.4 In this study, we use Global Moran's I to evaluate whether the spatial patterns are random, clustered, or dispersed. The Moran's I value (I) is calculated as:

where n is the number of spatial units indexed by i and j, x denotes the variable of interest, X is the mean of x, and w^ indicates the spatial weight between i and j. The null hypothesis is that the values being analyzed are randomly distributed across space (or no spatial autocorrelation). If the Moran's I statistic is significant, a positive Moran's I value denotes spatial clustering, whereas a negative Moran's I value indicates spatial dispersion. Note that the Moran's I value is a global measure; it is a single measure for an entire study area (in this study, the Tokyo metropolitan area) and does not illustrate the locations of spatial clustering within the study area.

4 As a global measure of spatial autocorrelation, we employed Global Moran's I rather than Getis-Ord General G (Getis and Ord, 1992) since our data exhibit clustering of both high and low values. When both high and low values cluster, the General G is less appropriate, since the high and low values are likely to cancel each other out.

On the other hand, the Getis-Ord Gi statistic (Getis and Ord, 1992; Ord and Getis, 1995) is a local measure of spatial autocorrelation, which is calculated for each spatial unit (the municipality in this study) within the study area.5 Therefore, the Gi* statistics identify the locations of spatial clusters of high values (hot spots) and low values (cold spots) if such spatial clusters exist. The Gi* statistic is calculated as:

Y"=i WjjXj-XjV^wu

nzj=1wlr(zn=iwijy

^ _ Y"=l*j

JY" x?

n is the number of spatial units indexed by i and j, xj is the value for j, and wij is the spatial weight between i and j. The Gi* statistic is essentially a z-score. A positive and significant Gi* statistic indicates a spatial clustering of high values, wherein a larger statistic indicates a more intense clustering of high values. A negative and significant Gi* statistic denotes a spatial clustering of low values, wherein a smaller statistic denotes a more intense clustering of low values. Using GIS, we plot the Gi* statistics on maps to examine the spatial patterns of hot and cold spots.

5 As a local measure of spatial autocorrelation, we employed Getis-Ord Gi* rather than local Moran's I (Anselin, 1995) since we are interested in the spatial patterns of the clustering of high and low values. In local Moran's I, a positive value represents a clustering of high or low values. A negative value indicates a spatial outlier. Since detecting spatial outliers is not the focus of this analysis, we use the Gi*. Also, the Gi* statistic, which represents a z-score, is straightforward to interpret.

For the spatial weight for the Moran's I and Gi* statistics, we use the first-order binary contiguity matrix based on the rook criterion, in which two spatial units are defined as neighbors when they share a common border. The contiguity matrix is commonly used as the spatial weight matrix for data represented by areal units (polygons) that vary in size. We also experiment with other spatial weights to check whether the results are sensitive to the choice of weights. In calculating Moran's I, the row elements of the spatial weight matrix are standardized so that their sum equals one. For the Gi* statistics, the row standardization is irrelevant; the statistics are invariant to the standardization.

In the spatial weight matrix, the number of neighbors for each municipality is small, ranging from 1 to 11 (mean: 5.28; standard deviation: 1.88) for the 243 municipalities in our study area (explained in Section 3.3). Therefore, for the local measure (Getis-Ord Gi*), we calculate the pseudo p-values based on permutation (number of permutations: 99,999) and determine the significance based on the pseudo p-values. With 99,999 permutations, the numbers of hot and cold spots at the 1% significance level are robust to the particular random permutation.

3.2 Spatial regression

Next, we use regression analysis to examine the relationships between commute times and the three participation measures (female labor force participation, regular employment, and part-time employment rates). For the regression analysis, we employ the spatial Durbin model (SDM), which includes spatial lags of the dependent variable as well as the explanatory variables (Anselin, 1988; LeSage and Pace, 2009). Female participation at a specific location may depend on female participation at neighboring locations, and vice versa. Fogli and Veldkamp (2011) show that women learn participation in the labor market from nearby employed women. Female participation at a particular location may also depend on commute time and other characteristics at neighboring locations, and vice versa. Vega and Elhorst (2017) demonstrate that characteristics in neighboring regions play a significant role in female participation. LeSage (2014) suggests the use of the SDM when endogenous interactions and feedback effects, or global spatial spillovers, are present. We also

estimate the standard model with ordinary least squares (OLS), a non-spatial specification, for comparison.

The SDM model and its data generating process are specified as follows (LeSage and Pace, 2009; Elhorst, 2014):

y = pWy + atn + WX6 + 8 (5)

y=(In- pW)-1(ain + + WX6 + e) (6)

E~N(0,a2In)

In equation (5), y is an n x 1 vector of observations on the dependent variable, W represents the n x n spatial weight matrix, p is the spatial autoregressive coefficient, X denotes an n x k vector of observations on explanatory variables, fi and d are k x 1 vectors of parameters to be estimated, £ is an n x 1 vector of disturbance terms, 0 represents an n x 1 vector of zeros, and ain indicates ann x 1 vector of ones associated with the constant term parameter a.

When p is nonsignificant, we proceed to estimate the spatial lag of X (SLX) model, which includes the spatial lags of the explanatory variables (WX) but does not have the spatial lag of the dependent variable (Wy). The SLX model is specified as follows (LeSage and Pace, 2009; Elhorst, 2014):

y = ain+X$ + WX6 + 8 (7)

The SDM and SLX models are estimated with the maximum-likelihood (ML) method. As in the previous section, we use the row-standardized first-order binary contiguity matrix as the spatial weight matrix, W.

In the standard model estimated with OLS, the marginal effect of an explanatory variable equals the coefficient estimate of that variable. For the SDM model, on the other hand, the marginal effect does not equal the coefficient estimate because of

the feedback effects (spillovers) induced by the data generating process. Therefore, we calculate the total effect, the sum of the direct effect and indirect effect, as the marginal effect. The direct and indirect effects for the rth explanatory variable in the matrix X are expressed as follows (LeSage and Pace, 2009, LeSage, 2014):

dy/dX'r = (In - pW)-1(Inpr + W6r ) (8)

(In - PW)-1 = In + pW + p2W2 ... (9)

A change in each explanatory variable is associated with an n x n matrix of partial derivatives. The direct effect is a scalar summary measure of the own-partial derivatives (dyi/OXf) provided by an average of the main diagonal elements of the matrix. The indirect effect is a scalar summary measure of the cross-partial derivatives (dyj/dX{) (spillovers) given by an average of the off-diagonal elements, where i^j. For the SLX model, the direct effect equals the coefficient estimate (ft), while the indirect effect corresponds to the coefficient estimate of the spatial lag of the explanatory variable (6), since W is row-standardized and its main diagonal elements are zero (Elhorst, 2014; LeSage, 2014).

3.3 The study area and data

The study area is the Tokyo metropolitan area (often referred to as Tokyo in this study), comprised of Tokyo Metropolis (the metropolitan prefecture) and its three neighboring prefectures: Chiba, Kanagawa, and Saitama, excluding remote islands. Tokyo is an ideal case for this study, since it is a large metropolitan area with long commute times. In fact, population wise, Tokyo is the largest metropolitan area in the world, with 35.6 million people recorded in the 2010 Population Census. Furthermore, Tokyo's urban spatial structure resembles a monocentric city, although several sub-centers exist. The core of the metropolitan area is the Tokyo ward area (the 23-special ward area), with a densely concentrated population and business districts. In this study, the Tokyo ward area is regarded as the CBD.

The spatial unit of analysis is the municipality (shi = city; ku = ward; machi = town; and mura = village) in the 2010 census, which is the smallest level of spatial detail available for the data used in this study. In 2010, the Tokyo metropolitan area contained 243 municipalities. We restrict the sample to municipalities with a population of 50 people or more to minimize sampling errors in participation statistics. Consequently, the number of observations (municipalities) is 243 or less, depending on the sample.

In this study, commute time is the average one-way travel time to work by municipality of residence (in minutes). We derive data on commute times for men and women aged 25-54 from the special tabulations of the 2008 Tokyo Metropolitan Region Person Trip Survey, obtained from the Tokyo Metropolitan Region Transportation Planning Commission (TMRTPC). The survey is conducted every ten years, and 2008 is closest to 2010. In examining relationships between commute times and the three participation measures, we use male commute times. Since most men work, employing male commute times is likely to alleviate selection bias arising from the use of female commute times.6

Data on labor force participation and regular and part-time employment for women aged 25-54 by marital status, the presence of children, and education are from publicly available and order-made tabulations of the 2010 Population Census. The National Statistics Center of Japan provided the order-made tabulations. The regular employment rate is the proportion of those who work as regular employees among the population, and the part-time employment rate is the similar rate of part-time employees. The rates are in percentages.

In the regression analysis, the dependent variable is the labor force participation rate, the regular employment rate, or the part-time employment rate. The independent variables are commute time (the variable of interest) and control variables. Commute time is the average one-way commute time for men (in minutes), as explained above. The men and women in our data are 25-54 years old, unless otherwise noted.

6 To mitigate selection bias, Black et al. (2014) use commute times for white married men. Commuting time data by marital status are not available in Japan.

The control variables include local housing prices, local income, and local unemployment rates for men, which Black et al. (2014) suggest as possible explanations for the large spatial variations in the labor force participation of married women. In addition, we use two more variables: the proportion of households with two or more children, and the availability of childcare centers. This last variable is important for this study since childcare access is found to be significantly associated with female participation in Tokyo (Kawabata, 2014). In Japan, eliminating long waiting lists for childcare centers has been an urgent policy issue. Since most childcare centers are publicly subsidized, availability is more likely to be an issue than cost, as in the cases of Italy and Germany (Del Boca and Vuri, 2007; Kreyenfeld and Hank, 2000).

Local housing prices come from the average residential land price in the 2010 Nikkei Electronic Economic Databank System (NEEDS). In the regression models, the log of the average residential land price is used.7 The local unemployment rates for men are from the 2010 Population Census. The proportion of households with two or more children represents the fraction of households with two or more children among all households (in percentage), from the 2010 Population Census. The availability of childcare centers is represented by the ratio (in percentage) of the capacity of licensed childcare centers to the population of preschool-aged children (under 6 years old) in 2010. The data on childcare centers come from prefectural governments and municipalities, and the data on the preschool-age population are from the 2010 Population Census.

7 As a measure of local income, we used average annual income per person, which is calculated as total taxable annual income divided by number of taxpayers. The data are from the 2010 municipality taxation status and others [Shichoson Kazei Joukyoutou no Shirabe], available from the Ministry of Internal Affairs and Communications of Japan. However, we do not include this income variable in regressions since it is highly correlated with the average residential land price (with a correlation coefficient of 0.89).

4 Empirical results

4.1 Spatial disparities in commute times and female participation

Table 1 presents the summary statistics of commute times and the three participation measures. Women have shorter commute times than men. The median one-way commute time is 38.8 minutes for women and 49.9 minutes for men. The three participation measures for women differ by marital status and the presence of children. Among the three groups of women, the labor force participation and regular employment rates are highest for unmarried women (followed by married women without children), and lowest for married women with children. The median labor force participation rate is 84.0% for unmarried women, 64.8% for married women without children, and 55.1% for married women with children. The median regular employment rate is 45.8% for unmarried women, 26.0% for childless married women, and 14.9% for married mothers. In contrast, the median part-time employment rate is the lowest at 17.3% for unmarried women, 23.9% for childless married women, and highest at 31.6% for married mothers.

Table 2 reports the Global Moran's I statistics for the three participation measures. The Moran's I values are all highly significant, suggesting that the female labor force participation, regular employment, and part-time employment rates are not evenly distributed across municipalities within the metropolitan area. The positive Moran's I values imply that high and low values of these rates are spatially clustered. The intensity of the spatial clustering is greater for married women with children than for unmarried women and married women without children. In addition to the first-order contiguity, we experimented with the following spatial weights: (a) inverse distance; (b) inverse distance squared; and (c) fixed distance band (determined to include at least one municipality as a neighbor). The above results did not change when we used different spatial weights.

Table 1. Summary statistics

Median Mean ^ Min. Max.

Commute time (min.) Men Women

Labor force participartion rate (%)

Unmarried women

Married women No

children

With children

Regul ar employment rate (%)

Unmarried women

Married women No

49.9 47.7 10.4 21.4 67.6

38.8 37.0 9.2 11.7 59.7

84.0 82.9 5.2 61.7 94.0

64.8 65.5 5.2 56.1 85.7

55.1 57.4 8.7 40.5 79.6

45.8 45.5 5.0 32.5 66.7

26.0 26.2 4.4 14.3 40.0

children With

children 14.9 16.2 4.5 10.4 37.9

Part-time employment rate (%) Unmarried

women 17.3 17.3 4.2 6.0 32.1 Married women No

children 23.9 24.3 6.3 8.5 46.2 With

children 31.6 31.0 6.6 9.2 47.5

Note: The men and women in the samples are 25-54 years old. Commute time is the average one-way travel time to work. The number of observations (municipalities) is 243, except for married women without children, in which case that figure is 242.

Table 2. Global Moran's I.

Moran's

z-score p-value

Labor force participartion rates

Unmarried women

Married women No

children

With children

Regular employment rates

Unmarried women

Married women No

children

With children

0.48 11.77 0.00

0.33 8.13 0.00

0.76 18.77 0.00

0.31 7.72 0.00

0.15 3.82 0.00

0.70 17.46 0.00

Part-t ime employment rates

Unmarried

Married women

children

With children

Note: The women in the samples are 25-54 years old. The number of observations (municipalities) is 243, except for married women without children, in which case that figure is 242.

Figure 1 depicts the spatial patterns of (a) male commute times and (b) their hot spots (spatial clusters of high values) and cold spots (spatial clusters of low values) based on the Getis-Ord Gi* statistics. Male commute times exhibit considerable intra-metropolitan disparities and distinct spatial patterns. Most hot spots are located around the inner suburbs (within approximately 30 kilometers of the Tokyo ward area), suggesting that many men residing in the inner suburbs travel to the CBD, enduring long commute times. Cold spots of male commute times are located around the core parts of the CBD and in the outer suburbs around the peripheries of the metropolitan area, indicating that men residing in the outer suburbs work in the suburbs rather than commute to the CBD. The spatial pattern of female commute times (not shown) is similar to that for men but is less conspicuous; for women, there exist a lesser number of hot and cold spots.

Tokyo ward aiea CBD)

O Prefecture D Municipality

One-way

ConniMil^ tmw

Male commute times

Getis-Ord G, BB Cold spot (p*> 001) ■I Cold spot (p = 005) GD Cold spot (p - 010} ! ■ Not significant EI Hoi »pot ip - 0 10)

■ Hot spot fp = 005)

■ Hot spot (p = 0 01)

l b i Hot and cold spots of male commute times

Fig. 1. Male commute times

Sore: The men arc 25-54 years old iColoi should be uved for this figuie in prill I >

<d> Ueosrnsi -»wn» (e) Married wsa® -a-ttti » HiWM (f) M^ffjsd -=.-CrrTj3D wift Aittsa

Regular employment rates Fig. 2. Hot and cold spots of female labor force participation and regular employrrient rates

•"Ce'jet ¿evij be <aed fee -TTit t

Figure 2 portrays the hot and cold spots of female labor force participation and regular employment rates. The spatial patterns of these rates differ notably by marital status and the presence of children. In particular, striking differences exist between married women with and without children. For married women with children, the rates of labor force participation and regular employment show more distinctive spatial

patterns. Many cold spots of these rates are located around the inner suburbs, while most hot spots are located around the outer suburbs. On the other hand, for married women without children, these rates are more evenly distributed across municipalities, in accordance with their smaller and less significant Moran's I values (see Table 2). Compared with the spatial pattern of male commute times (Figure 1), for married women with children, many of the cold spots of the labor force participation and regular employment rates overlap with the hot spots of male commute times and vice versa (i.e., many of the hot spots of the labor force participation and regular employment rates overlap with the cold spots of male commute times). The spatial patterns of the part-time employment rates are not presented; they do not differ much by marital status and the presence of children, and have no obvious spatial relationships with male commute times.

The correlation coefficients in Table 3 support the visual relationships from the maps (Figures 1 and 2). For married women with children, the Gi values of the labor force participation and regular-employment rates are highly and negatively correlated with the Gi values of male commute times (with correlation coefficients of -0.71 and -0.78, respectively). For married women without children, the correlations are also negative but weaker. For unmarried women, the correlations are smaller, and the sign is positive. For the part-time employment rates, the correlations do not differ much by marital status and the presence of children.

Table 3. Getis-Ord Gi*: correlations with male commute times

Labor force participartion rates

Unmarried women 0.02

Married women

No children -0.59

With children -0.71

Re gular employment rates

Unmarried women 0.22 Married women

No children -0.20

With children -0.78

Part-t ime employment rates

Unmarried women -0.26 Married women

No children -0.27

With children -0.25

Note: The men and women in the samples are 25-54 years old. The number of observations (municipalities) is 243, except for married women without children, in which case that figure is 242.

4.2 Regression results

Table 4 reports the estimation results of the standard models and the SDM or SLX models for labor force participation rates. For comparison, we present the results for married men.

One of the statistical tests to select a model that better describes the data is the likelihood ratio (LR) test (Elhorst, 2014). The value of the LR test (-2*(logLrestricted-logLunrestrected)) follows a chi-squared distribution with degrees of freedom (df) equal to the number of restrictions. For each sample, the LR test of the SDM versus standard model specifications takes a value greater than 12.6 (the 5% critical value with 6 df), suggesting the SDM as a preferred specification. Since the estimates for the spatial autoregressive coefficients (p) for college-educated married women without children and married women with children are nonsignificant, for those groups of women we also estimated the SLX models. The values of the LR test of the SLX models versus the standard models are all greater than 11.1 (the 5% critical value with 5 df). The LR test statistics further suggest that the SLX is a preferred specification

to the SDM. For these groups of women, therefore, we report the estimation results of the SLX models instead of those of the SDM models.

Table 4. Regression of labor force participation

Ma Mar

rri ried

ed wo

me me

8 The p-values of the LR test of the SLX models versus the standard models are smaller than the p-values of the LR test of the SDM models versus the standard models.

children

With children under 6

With children, none under 6

HS or less

Col leg e

HS or less

HS or less

Col leg e

Sta Sta Sta

nd. S nd. S nd.

mo M mo M mo

del del del

S ta n d.

m o d el

Each sample excludes municipalities with populations of less than 50 people or no neighboring municipalities.

Significant at 1%; ^Significant at 5%.

Table 5. Marginal effects of commute time on labor force participation rates

Total effect ^

Direct

Indirect

Married men

Unmarried women

Married women

No children

HS or less

College

With children under 6

HS or less

College

effect effect

0.02 -0.01 0.03 243

(0.37) (0.62) (0.60)

0.11 -0.06 0.17 243

(1.41) (-1.44) (1.95)

-0.23 ** -0.01 -0.23 * 239

(-2.65) (-0.12) (-2.12)

-0.13 0.01 -0.15 206

(-1.18) (0.14) (-0.97)

-0.41 ** -0.05 -0.37 ** 238

(-5.47) (-0.62) (-3.52)

-0.54 ** -0.12 -0.42 ** 209

(-5.18) (-1.14) (-2.79)

With children, none under 6

HS or less -0.12 * -0.04 -0.08 243

(-2.29) (-0.84) (-1.07)

College -0.31 ** -0.16 -0.15 218

(-3.39) (-1.92) (-1.17)

Note : **S ignificant at 1%; *Significant at 5%. t-values are in parentheses. W is the binary contiguity matrix.

Each sample excludes municipalities with populations of less than 50 people or no neighboring municipalities.

Table 5 presents the total, direct, and indirect effects of commute time (the variable of interest) on the labor force participation rates in the SDM or SLX models, in accordance with the preferred models in Table 4.9 The total effects differ markedly by marital status and the presence of children. For married men and unmarried women, the association between commute time and the participation rate is positive and nonsignificant. In contrast, for married women with children, commute time is negatively and significantly associated with the participation rate. For married women

9 In fact, the total, direct, and indirect effects in the SLX models are similar to those in the SDM models (not presented).

without children, the association is also negative and significant for those with a high school education or less but is nonsignificant for college graduates.

Among married women with children (for both under 6 years old as well as 6 and above), the magnitude of the negative association is greater for college graduates than for those with a high school education or less. This result indicates that highly educated mothers are more sensitive to commute times when participating in the labor market. For college-educated married women with children under 6 years old, for instance, a one-minute increase in commute time is associated with a 0.54 percentage point decrease in the labor force participation rate. Since the range of commute times across municipalities is 46.2 minutes (see Table 1), the commute time difference results in a 24.9 percentage point difference in the labor force participation rate. This is not a small difference given that the range of the participation rate for these women is 64.6% (the maximum of 84.6% minus the minimum of 20.0%).

It is interesting to note that the direct effects are nonsignificant for all groups, suggesting that labor force participation is not significantly associated with commute time in one's own municipality. On the other hand, for childless married women with a high school education or less, and married women with children under 6 years old, the indirect effects (spillover effects) are significant. This result implies that for these groups of women, labor force participation is more responsive to commute time in neighboring municipalities than to commute time in one's own municipality. This could indicate that those women tend to work in the vicinity of their residences, and as a result, an increase in commute time in neighboring municipalities might be associated with a decrease in the labor force participation rate. For married women with children aged 6 and older, while the direct and indirect effects are both nonsignificant, the total effects are significant. This result indicates that the associations between commute time and labor force participation are significant when the impact of commute time combines the impact of commute time in one's own municipality with that of neighboring municipalities. In fact, for these women, the magnitude of the total effects is notably greater than that of the direct and indirect effects.10

10 The total effect is the sum of the direct and indirect effects. For married women with children aged 6 and older, the total effect is -0.12 (the sum of -0.04 and -0.08) for those with a high school education or

When the total effects in the SDM or SLX models are compared with the coefficients (marginal effects) of commute time in the standard models (Table 4), the signs and significance are consistent. Compared with the coefficients in the standard models, the magnitude of the total effects in the SDM or SLX models is greater or almost the same. This result suggests that the associations between commute time and the labor force participation rate tend to become greater when the models incorporate the spatial dependence (spillovers).

Table 6 shows the estimation results of the standard models and the SDM or SLX models for regular employment rates. For married women with children under 6 years of age and college-educated married women with children aged 6 and older, the spatial autoregressive coefficients (p) are nonsignificant. For these groups of women, we also estimate the SLX models. Since the LR test statistics indicate SLX as a preferable specification to SDM, we report the results of the SLX models.

Except for childless married women with a high school education or less and college-educated married women with children under 6 years old, the values of the LR test of the SDM or SLX models versus the standard models exceed the 5% critical values (12.6 for the SDM models and 11.1 for the SLX model), indicating the SDM or SLX models as preferred alternatives. For childless married women with a high school education or less and for college-educated married women with children under 6 years old, the values of the LR test (7.8 and 9.4, respectively) are smaller than the 5% critical value, implying that the standard models need not be rejected in favor of the SDM or SLX model. For these two groups of women, the signs and significance of the total effects in the SDM or SLX model are consistent with those of the coefficients in the standard models, and the differences in the magnitude of the total effects and coefficients in the standard models are small.

Table 7 shows the total, direct, and indirect effects of commute time on the regular employment rates in the SDM or SLX models, corresponding to the preferred

less, and is -0.31 (the sum of -0.16 and -0.15) for those with a college education. The significant total effect indicates that the total effect is significantly different from zero, while the direct and indirect effects are not.

models in Table 6.11 The total effects of commute time indicate notable differences by marital status and the presence of children. For unmarried women and married women without children, commute time is not significantly associated with the regular employment rate. For married women with children, on the other hand, commute time is negatively and significantly associated with the regular employment rate, except for married women with children aged 6 or older and high school education or less; for this group of women, the association is nonsignificant. When the sample is for married women with children in general (not shown in Table 7), the total effect is negative and significant (-0.10 with a t-value of -2.02).

Among married women with children, the negative associations between commute time and the regular employment rate are greater in absolute values for college graduates than for those with a high school education or less, as in the case of labor force participation. For college-educated married women with children aged 6 or older, for example, a one-minute increase in commute time is associated with a 0.32 percentage point decrease in the regular employment rate. Incorporating the range of commute time (46.2 minutes) results in a 14.8 percentage point difference in the regular employment rate.

Table 6. Regression of regular employment

11 As in the case of labor force participation, the total, direct, and indirect effects in the SLX models are similar to those in the SDM (not presented).

12 The total effect is estimated with the SDM since the autoregressive coefficient is significant.

No With children, none

children

children under 6

under 6

Stan d.

mod el

HS or less

HS or less

College

HS or less

C ol le

S Stan D d.mo M del

St an d.

m od el

St a n d.

m o d el

Sta nd. mo del

St an d.

m od el

Each sample excludes municipalities with populations of less than 50 people or no neighboring municipalities.

**Significant at 1%; *Significant at 5%.

Table 7. Marginal effects of commute time on regular employment rates

Total effect Direct effect Indirect effect N

Unmarried women

0.11 * -0.12

Married women

No children

(-0.15) (2.47) (-1.37)

HS or less

-0. 09

-0. 02 -0.07

(-1.16) (-0.36) (-0.75)

College

(-1.82) (0.78) (-1.73)

With children under 6

HS or less

-0. 18

0.00 -0.18

(-3.85) (0.00) (-2.78)

College

-0.07 -0.15

(-2.17) (-0.69) (-1.02)

With children, none under 6

HS or less -0.08 -0. 06 -0.02 243

(-1.74) (-1.80) (-0.40)

College -0.32 (-2.99) ** -0.16 -0.15 218

(-1.64) (-1.07)

Note : **S ignificant at 1%; *Significant at 5%. t-values are in parentheses. W is the binary contiguity matrix.

Each sample excludes municipalities with populations of less than 50 people or no neighboring municipalities.

Interestingly, the direct effect is significant and positive for unmarried women, while the direct effects for the other groups of women are all nonsignificant and mostly negative. This finding could indicate that unmarried women have lower spatial constraints than married women; unmarried women might be able to travel longer to work as regular employees. For married women with children under 6 years of age and a high school education or less, the indirect effect is significant, whereas the direct effect is nonsignificant. This implies that these women are more responsive to commute times in neighboring municipalities than to commute time in their own municipality. This might be partly because those women are likely to work in nearby municipalities as regular employees. For college-educated married women with children (under 6 years of age as well as 6 and above), the total effects are significant, while both the direct and indirect effects are nonsignificant. This result suggests that for these college-educated women, commute time is significantly associated with regular employment rates when

the impact of commute time combines that in one's own municipality with that in

neighboring municipalities (spatial spillovers)13.

When we compare the total effects with the coefficients in the standard models (Table 6), the sign and significance are mostly consistent. The differences are for unmarried women and for married women with children aged 6 or older as well as a high school education or less; for these groups of women, the coefficients in the standard models are significant but the total effects in the SDM are nonsignificant.

We do not present the results for part-time employment rates since their associations with commute time are mostly nonsignificant.

The household division of labor hypothesis is most relevant when one spouse (husband) makes a long commute and the other spouse (wife) makes a short commute or stays home. While many of the suburban municipalities in our sample seem to fit this pattern, there are some for which the commuting distance from the CBD is so long that the household division described above is less likely to hold. To address this concern, we estimate our models using the sample of municipalities in which 10% or more of male commuters travel to the CBD for work. The results are presented in Appendices A and B. Figures A1 and A2 show the corresponding maps for male commute times and the female labor force participation and regular employment rates, respectively. The spatial patterns of the hot and cold spots are largely consistent with those for the entire metropolitan area, except for the regular employment rate for married women with children (Figure A2(f)). For these married mothers, hot spots of the regular employment rate appear in the CBD, while the hot spots are non-existent in the CBD for the sample of the entire metropolitan area including outer suburbs with high regular employment rates (Figure 2(f)).

13 The total effect is -0.32 (not -0.31 due to rounding), which is the sum of the direct effect (-0.16) and indirect effect (-0.15). The total effect is markedly larger than the direct and indirect effects, in absolute values. The significant total effect denotes that the total effect is significantly different from zero, whereas the direct and indirect effects are not.

The estimated total, direct, and indirect effects of commute time on female labor force participation and regular employment rates in the SDM or SLX models are reported in Tables B1 and B2, respectively. The results of the SLX models are presented when the autoregressive coefficients (p) in the SDM models are nonsignificant and when the LR test values indicate the SLX models as preferable to the SDM models. The results are largely consistent with those for the entire metropolitan area. For the labor force participation rates, the signs and significance of the total effects remain the same. For the regular employment rates, the signs are consistent but the significance shows some changes. The changes, however, do not alter the finding that the associations between commute time and the regular employment rates are negative and significant for married women with children. Of note is that the magnitude and significance of the total effects for college-educated married women are greater than those for the samples of the entire metropolitan area (Table 7). This result suggests that within the commuting distance from the CBD, highly educated women who work as regular employees are particularly sensitive to commute time.

5 Conclusions

Our municipal-level analysis revealed considerable intra-metropolitan disparities in female labor force participation, regular employment, and part-time employment rates. As implied by the Global Moran's I statistics, these rates are not evenly distributed within the metropolitan area. There are spatial clusters of high and low rates. The hot and cold spot maps show that the spatial patterns of the labor force participation and regular employment rates differ markedly by marital status and the presence of children. Compared with unmarried women and married women without children, married women with children exhibit more significant spatial clustering of high and low rates of the participation and regular employment. For married mothers, the spatial clusters of low participation and regular employment rates are largely located in the inner suburbs, many of which overlap with the spatial clusters of long male commute times.

The spatial regression results indicated that for married mothers, a longer commute is significantly associated with lower labor force participation and regular employment rates, while for unmarried and childless married women, the associations

are mostly nonsignificant. The findings support the view that mothers' labor market participation is sensitive to commute time. Since residential decisions are endogenous, the effect of commute time on participation is not causal. Rather, the circumstances in the Tokyo metropolitan area induce households to simultaneously decide on location and labor market participation for both spouses. The typical choices are: (1) living close to the CBD and both spouses work there; (2) living in the inner suburbs, whereby the husband commutes to the CBD and the wife stays home or works locally; and (3) living in the outer suburbs and both spouses work in the suburbs.

The spatial regression results also indicated that among married mothers, the negative associations between commute time and the labor force participation and regular employment rates are greater for college graduates than for those with a high school education or less. These findings differ from those of Black et al. (2014), who find greater associations for high school graduates than for college graduates among married women.14 These contradictory results may have arisen partly because our study is based on intra-metropolitan data, while the study by Black et al. (2014) is based on inter-metropolitan data. In the inter-metropolitan analysis, highly educated women are perhaps more likely to live and work in larger metropolitan areas where they have longer commute times. In contrast, in our intra-metropolitan analysis of Tokyo, highly educated women are more likely to live and work closer to the CBD, with shorter commute times.

Our findings imply that for married mothers, intra-metropolitan disparities in commute times play an important role in their participation in the labor market. The inner suburbs, which are farther from the CBD but from where men nonetheless regularly commute, have high concentrations of lower labor force participation and regular employment rates for married mothers. Since commute time is not significantly associated with the labor force participation rate for married men (see Table 4), suburban living that entails a long commute for the father intensifies the household

14 In a model for women with children under 5 in the study by Black et al. (2014, Panel B2 in Table 6, p. 68), the association is greater for college than for high school graduates, but in all other models, including married women in general, the associations are greater for high school than college graduates.

division of labor in which the father travels to the CBD and the mother either stays home or works locally. We argue that the spatial transportation constraint induces this gender division, resulting in unique patterns for married mothers.

In many countries, among women with children, preferences for participation in the labor market are much higher than actual participation rates (Jaumotte, 2003). A national survey in Japan shows that among couples with children under 15, most women (86%) who are not currently working would want to (National Institute of Population and Social Security Research, 2016). Our findings suggest that implementing policies, which alleviate commuting constraints, could help women with children participate more actively in the labor market. Examples include improving employment accessibility, reducing congestion, promoting flexible working hours, increasing the housing supply around employment centers, and encouraging male commitment to housework and childcare.

In recent years, the number of dual-earner couples in Japan has risen dramatically. Spatio-temporal analysis using data after 2010 is a possible direction for further research. Another is to conduct a comparative analysis of Tokyo and other metropolitan areas. Our research revealed that within a metropolitan area, the levels of female participation differ by location, and the intra-metropolitan disparities have unique spatial patterns. Spatially disaggregated analysis potentially unveils important dimensions of the urban labor market that deserve more attention.

Appendix A

O Prefecture □ Municipality

Oik-way commuting tune

□ <- 30 (

■ 30-40

■ 40-50

■ 50-60

■ >60

Tokyo \v;utl iirca

(a) Male commute tinier

Getis-Ord Gt ■I Cold »pot (p- 0.0 It

■ C old spot (p - 0 05) F1 Cold spot ip = 0 10) 1 I Not ugmikitnt

I I Hot spot (p-0 10)

■ Hot spot <p- 0.05)

■ Hot spot (p - 0.01) No data I no ueiyhbon»

(b) Hot and cold spots of male commute times

Fig. VI. Male commute times for municipalities m which 10 percent or more of male commuters travel to the Tokyo ward area to work

Note The men aie 25*54 years old < Color should be used for tht> tisiue in print >

Figure A1. Male commute times for municipalities in which 10 percent of more of male commuters travel to the Tokyo ward area to work.

U -Anc;«; ûeea-Otas

H Cold jç<xtS -ûûti ■ Ce« -ftiîîi Cola jçot'y -ttlûi >'e: i^uScs îiet »^t^p-4101

< Î > Uamanied "•orr-'ac

(b) M*fT»i •W04T«C ÛD

Hot »pot» and col<î spocsoalabor force participation rat»»

(d) lTafli*fr>8d (s) -R-OÎT« -tvtSh ao cbfldree (0 nwi» cbâdrec

Hot »pot» and cold »pet»ofrégulai employment rate»

Fig. A2. L abor force participation and regular employment rates farwamen azed 2 5-54 years for municipalities in which 10 percent or more maie commutais travel to the Tokyo vv-ard area to work. (Color ibcoiS te aiad fof 2» Sgan » ceux. )

Appendix B

Table B1. Marginal effects of commute time on labor force participation rates for

municipalities in which 10% or more of male commuters travel to the Tokyo ward area for work

Total effect Direct effect Indirect effect N Model

Labor force participation rate

Unmarried women -0.05 0.02 -0.07 162 SDM

(-0.54) (0.33) (-0.63)

Married women

No children

HS or less

College

(-2.06)

(-1.63)

0.10 -0.30

(0.78) (-1.82)

-0.01 -0.16

(-0.08) (-0.90)

162 SLX

160 SLX

With children under 6

HS or less

0.11 -0.44

162 SDM

(-3.96)

College

(-5.23)

With children, none under 6

(0.82) (-2.72)

-0.50 ** 0.00 -0.50 ** 161 SDM

(0.03) (-2.82)

HS or less

-0.25 ** 0.02 162 SLX

(-3.36) (-2.76) (0.13)

College

-0.18 -0.08

162 SDM

(-2.03) (-1.31) (-0.45)

Table B2. Marginal effects of commute time on regular employment rates for municipalities in which 10% or more of male commuters travel to the Tokyo ward area for work.

Direct

Total effect Indirect effect N Model

effect

Regular employment rate

Unmarried women -0.13

(-1.56

Married women No children

HS or less -0.06

(-0.66 )

College -0.29 *

(-2.23 )

0.12 -0.25 * 162 SDM

(1.45) (-2.14)

-0.02 -0.04 162 SDM

(-0.22) (-0.30)

-0.06 -0.22 160 SLX

(-0.37) (-1.03)

With children under 6

HS or less -0.11

0.06 -0.18 162 SLX

(-1.94

(0.83) (-1.78)

College -0.39

0.16 -0.55 * 161 SLX

(-4.35

) (1.18) (-3.30)

With children, none under 6

HS or less

■0.15 *

-0.02 -0.13 162 SLX

(-3.79 )

(-0.44) (-1.90)

College -0.36

0.03 -0.39 * 162 SLX

(-4.28

(0.25) (-2.72)

Note : ** Significant at 1%; *Significant at 5%. t-values are in parentheses. W is the binary contiguity matrix.

Each sample excludes municipalities with populations of less than 50 people or no neighboring municipalities.

References

Abe Y (2011) Family labor supply, commuting time, and residential decisions: The case of the Tokyo metropolitan area. Journal of Housing Economics 20: 49-63.

Abe Y (2013) Regional variations in labor force behavior of women in Japan. Japan and the World Economy 28: 112-124.

Alonso W (1964) Location and land use: towards a general theory of land rent. Harvard University Press, Cambridge.

Anselin L (1988) Spatial econometrics: methods and models. Kluwer Academic Publishers, Dordrecht.

Anselin L (1995) Local Indicators of Spatial Association-LISA. Geographical Analysis 27: 93-115.

Black DA, Kolesnikova N, Taylor LJ (2014) Why do so few women work in New York (and so many in Minneapolis)? Labor supply of married women across US cities. Journal of Urban Economics 79: 59-71.

Boarnet MG, Hsu H-P (2015) The gender gap in non-work travel: The relative roles of income earning potential and land use. Journal of Urban Economics 86: 111-127.

Brasington DM (forthcoming) What types of people sort to which public services? Papers in Regional Science.

Crane R (2007) Is there a quiet revolution in women' s travel? Revisiting the gender gap in commuting. Journal of the American Planning Association 73: 298-316.

Del Boca D, Vuri D (2007) The mismatch between employment and child care in Italy: the impact of rationing. Journal of Population Economics 20: 805-832.

Elhorst, J. P. (2014) Spatial Econometrics: From Cross-Sectional Data to Spatial Panels. Springer, Heidelberg.

Fogli A, Veldkamp L (2011) Nature of nature? Learning and the geography of female labor force participation. Econometrica 79(4): 1103-1138.

Gamsu S (2016) Moving up and moving out: The re-location of elite and middle-class schools from central London to the suburbs. Urban Studies 53: 2921-2938.

Getis A, Ord JK (1992) The analysis of spatial association by use of distance statistics. Geographical Analysis 24(3): 189-206.

Gordon P, Kumar A, Richardson H (1989) Gender differences in metropolitan travel behaviour. Regional Studies 23 (6): 499-510.

Hanson S, Hanson P (1981) The impact of married women's employment on household travel patterns: a Swedish example. Transportation 10:165-183.

Hanson S, Johnston I (1985) Gender differences in work-trip length: explanations and implications. Urban Geography 6(3): 193-219.

Hjorthol RJ (2000) Same city - different options: an analysis of the work trips of married couples in the metropolitan area of Oslo. Journal of Transportation Geography 8: 213-220.

Jaumotte F (2003) Labour force participation of women: empirical evidence on the role of policy and other determinants in OECD countries. OECD Economic Studies 37: 51-108.

Johnston-Anumonwo I (1992) The influence of household type on gender differences in work trip distance. Professional Geographer 44(2): 161-169.

Kawabata M (2014) Childcare access and employment: the case of women with preschool-aged children in Tokyo. Review of Urban & Regional Development Studies 26: 40-56.

Kreyenfeld M, Hank K (2000) Does the availability of child care influence the

employment of mothers? Findings from western Germany. Population Research and Policy Review 19: 317-337.

Lee BS, McDonald JF (2003) Determinants of commuting time and distance for Seoul residents: The impact of family status on the commuting of women. Urban Studies 40(7): 1283-1302.

LeSage, J, Pace, R. K. (2009) Introduction to Spatial Econometrics. CRC Press, FL.

LeSage, J. P. (2014) What regional scientists need to know about spatial econometrics. The Review of Regional Studies 44: 13-32.

MacDonald HI (1999) Women's employment and commuting: explaining the links. Journal of Planning Literature 13(3): 267-283.

Madden JF (1981) Why women work closer to home. Urban Studies 18: 181-194.

McGuckin N, Murakami E (1999) Examining trip-chaining behavior: comparison of travel by men and women. Transportation Research Record 1693:79-85.

McLafferty S, Preston V (1997) Gender, race, and the determinants of commuting: New York in 1990. Urban Geography 18(3):192-212.

Mills ES (1972) Studies in the structure of the urban economy. Johns Hopkins Press, Baltimore

Moran PAP (1950) Notes on continuous stochastic phenomena. Biometrika 37(1): 1723.

Muth RF 1969. Cities and housing: the spatial pattern of urban residential land use. The University of Chicago Press, Chicago.

National Institute of Population and Social Security Research (2016) Summary Results from the 15th National Fertility Survey [Dai 15 Kai Syusyoudoukoukihonchosa Kekka no Gaiyou]. (in Japanese)

Neto RS, Duarte G and Paez A (2015) Gender and commuting time in Sao Paulo metropolitan Region. Urban Studies 52: 298-313.

Ord JK and Getis A (1995) Local spatial autocorrelation statistics: distributional issues and an application. Geographical Analysis 27(4): 286-306.

Roberts J, Hodgson R and Dolan P (2011) "It's driving her mad' ": gender differences in the effects of commuting on psychological health. Journal of Health Economics 30: 1064-1076.

Rouwendal, J (1999) Spatial job search and commuting distances. Regional Science and Urban Economics 29: 491-517.

Singell LD and Lillydahl JH (1986) An empirical analysis of the commute to work patterns of males and females in two-earner households. Urban Studies 2: 119129.

Tokyo Metropolitan Region Transportation Planning Commission (TMRTPC) (2010) 5th Tokyo Metropolitan Region Person Trip Survey: Tokyo Metropolitan Region as Seen in the Movements of People [Hito No Ugokikaramieru Tokyo Toshiken]. Tokyo Metropolitan Region Transportation News Vol. 22. (in Japanese)

Turner T, Niemeier D (1997) Travel to work and household responsibility: new evidence. Transportation 24: 397-419.

Vega SH and Elhorst JP (2017) Regional labour force participation across the European Union: a time-space recursive modelling approach with endogenous regressors. Spatial Economic Analysis 12: 138-160.

White MJ (1986) Sex differences in urban commuting patterns. American Economic Review 76(2): 368-372.

Highlights

• We explore intra-metropolitan spatial patterns of female labor force participation.

• The spatial patterns differ by marital status and the presence of children.

• Married mothers have unique spatial patterns of female participation.

• Highly-educated mothers are especially sensitive to commute time.

• Findings shed new light on geographic disparities in female participation.