Abstract
Gender differentials in commuting have been reported in the literature, often couched within the household responsibility hypothesis. This hypothesis attributes shorter commutes to females due to a disproportionate load of household responsibilities. The objective of the present study is to report research regarding commuting time in São Paulo Metropolitan Region, in Brazil. Based on microdata from the Demographic Census of 2010 the focus of the present study is on the role of marital status and presence of dependents on gender differentials in commuting time. Specifically, the research seeks to determine whether there is empirical support in this region for the household responsibility hypothesis. The results suggest that marital status exerts a stronger influence on the commuting time of working women, with the number of dependents (children and elderly) exerting a smaller influence on commuting time. Gender differentials are observed also for single and formerly married working females, which suggests other cultural or environmental factors not fully captured by the household responsibility hypothesis. Most studies, however, are set in North America. This research contributes towards the development of a broader, international knowledge foundation regarding gender and commuting patterns.
Keywords
Introduction
With a population of approximately 19.8 million, São Paulo Metropolitan Region (SPMR) is the largest urban centre in Brazil. As a megacity, it belongs in a class with other cities such as Tokyo, New York, Seoul and Mexico City. SPMR, not unlike some of these megacities, is known for chronic urban issues associated with very high levels of agglomeration, including traffic congestion and many associated externalities (Wheeler, 2003; Zheng, 1998). In 2009, according to the official annual Brazilian household survey (Instituto Brasileiro de Geografia e estatística (IBGE), 2010), approximately 55.3% of commuters in the SPMR spent more than 30 minutes travelling, and about 23% of them (approximately 1.8 million people) spent more than one hour going from home to work every day.
The proportions of commuters who travel more than 30 minutes and more than one hour exceed those observed at other metropolitan regions in Brazil and are also high for international standards. Considering the next four biggest metropolitan regions in the country, the percentage of commuters spending more than 30 minutes travelling was 46.6% for Recife, 44.5% for Rio de Janeiro and Belo Horizonte and 33.0% for Porto Alegre. Commuting time is also high compared with international cities. For instance, according to McKenzie and Rapino (2011), the percentage of workers in American cities in 2009 who spent 30 minutes or more commuting was 33.5%, and only 7.8% of these workers spent an hour or more commuting. At 34.6 min, the average time spent commuting in New York Metropolitan Area (comprised of New York City, Northern New Jersey and Long Island) is well below the value observed for SPMR, which stands at approximately 40.2 minutes.
The literature on the determinants of commuting length has identified a number of variables associated with commute times. A consistent finding is that the personal characteristics of individuals influence commuting patterns (Ericksen, 1977; Hanson and Johnston, 1985; Turner and Niemeier, 1997). A recent, and extensive study of US metropolitan regions (Crane, 2007), shows that positive covariates of commuting time include income, age and educational achievement. These factors, however, can operate in different ways based on gender and marital status. A key finding of Crane is that gender differences in commuting identified earlier in the literature (e.g. Madden, 1981; Turner and Niemeier, 1997; White, 1986) are tenaciously persistent, and women still tend to have shorter commutes measured both in terms of distance and time. Further, whereas marital status (being married) tends to increase commuting time for males, it has the opposite effect for females (Crane, 2007: table 4). These findings are also borne by the recent American Community Survey Report (Rapino and Cooke, 2011), whereby the average commuting time for women in the USA in 2009 was found to be 23.4 minutes, which is well below the average of 26.7 minutes for men.
The existence of differences in commuting patterns between males and females has led to a body of research that attempts to explain this phenomenon, which has important implications for labour force participation, housing and transportation (Crane, 2007; Rosenbloom, 1978). One strand of the literature posits that commuting differentials are a consequence of differences in the way males and females participate in household maintenance activities. According to the household responsibility hypothesis (Johnston-Anumonwo, 1992), women typically bear a greater responsibility for maintaining the household and caring for dependants (e.g. children) than men. Competing demands for time result in women being less mobile – thus their reduced commuting length (Crane, 2007; Giuliano, 1998; Lee and McDonald, 2003; Madden, 1981; Turner and Niemeier, 1997).
In this paper, we aim to investigate the factors that influence commuting times in the SPMR, a heretofore unexplored case study. More concretely, we are interested in empirically testing the household responsibility hypothesis for this region. São Paulo Metropolitan Region is one of the largest urban centres in the world and commuting times are relevant from the perspective of transportation, residential patterns and labour participation. However, to the best of our knowledge, the region has not been studied from this perspective before. A large share of female participation in the workforce (approximately 44% of all employment in the region) allows us to study gender differentials in commuting time under a diversity of urban household and labour market environments. Further, there are clear cultural differences, since a prevailing attitude in Latin America concerns the dominant role of the male (DeSouza et al., 2004). Finally, there is a significant contingent of domestic workers, which comprised 9.3% of total of employed people of this metropolitan region in 2010. The large domestic work component in the workforce can increase the commuting time of women, as domestic workers travel to reach high-income areas where employment opportunities are available.
Analysis is based on data drawn from the 2010 Demographic Census of Brazil. Commuting times in this data set are recorded in a categorical fashion. For this reason, multivariate analysis of the factors that influence commuting time in the SPMR is conducted using an ordinal probit model. In order to test the household responsibility hypothesis, we consider (in addition to other potential confounders) the influence of marital status, spouse employment and presence of dependent family members (children and not active seniors). These latter variables serve as proxies for household responsibilities.
Gender difference in commuting time and the household responsibility hypothesis
In studies on the determinants of commuting time in urban centres, there is fairly consistent evidence regarding gender, with women having shorter commutes than men. In the economic literature, White (1986) and Madden (1981) published pioneering work that theoretically and empirically addressed these differences. In economics commuting time is considered to be derived from decisions regarding the place of residence, place of work, commuting mode, housing costs and wages. Based on these factors, the disparities in commuting time between individuals are thought to reflect differences in preferences for residential space and commuting time. The literature in transportation, geography and planning, on the other hand, emphasises the role of demographic and geographic variables. For instance, MacDonald (1999) has drawn attention to the fact that the shorter commuting time of women may be associated with a lower income level (possibly due to discrimination in the labour market), the spatial distribution of jobs traditionally held by women and the internal allocation of maintenance activities within households. According to the latter explanation, better known as the household responsibility hypothesis, women who traditionally devote more time on housework than men (perhaps because of perceptions of value or roles), must reconcile these activities with paid work. This, in turn, leads women to choose to live closer to work or, in cases where the residential choice is made jointly with other household members, to work closer to home. In either case, the observable outcome is the same, namely a shorter commute than men.
There is no disputing that women’s commute is shorter than men’s. The mechanisms of operation with regards to the household responsibility hypothesis, on the other hand, appear to be diverse. The research of Lee and McDonald (2003) in Seoul indicates that the effect of household responsibility was linked to child care. This finding is similar in other studies that report a negative effect of children on the commuting time of women (Crane, 2007; Ericksen, 1977; Giuliano, 1998; Lee and McDonald, 2003; Madden, 1981; Turner and Niemeier, 1997). However, there are other studies that show a non-significant or even a positive influence of the presence of children on the commuting time of women (Gordon et al., 1989; Hanson and Johnston, 1985; Johnston-Anumonwo, 1992; White, 1986). The study of Lee and McDonald (2003) contains an interesting twist: the presence of supportive adults (i.e. parent or in-laws) contributed to extend the commute of women, which further emphasises the relevance of child care.
The importance of marital status and employment situation of the spouse as determinants of the commuting time of women also shows disparities. For example, Johnston-Anumonwo (1992) suggests for the case of American commuters that household responsibilities are important for understanding the shorter commutes of women only in the case of married women where both partners work, since presumably the additional responsibility of maintenance activities would fall mostly on women. However, Lee and McDonald (2003) in South Korea, and Crane (2007) in US metropolitan areas, emphasise that being married negatively affects the commuting time of women, regardless of whether the spouse works or not.
To the best of our knowledge, this study is the first to address the household responsibility hypothesis for any Brazilian metropolitan region. This is somewhat surprising, since even cursory descriptive analysis, as seen below, shows that both marital status and the presence of children seem to have quite different effects on the working time of women relative to men in this region.
SPMR and the data set
Composed of 39 municipalities, SPMR has a very high population density of approximately 2.6 × 106 inhabitants per square kilometre. The city of São Paulo at the core has an economy that specialises in the service sector, and concentrates about 57% of its population, in only about 19% of the area of the region (IBGE, 2012).
Similar to most large Brazilian Metropolitan Regions, rapid urban development of SPMR occurred without adequate expansion of basic household and transport infrastructure (Meyer et al., 2004; Nadalin and Iglori, 2010). Nadalin and Igliori (2010) contend that this explains, at least partially, the pattern of location of families according to income, with more affluent ones less distant from employment centres.
The SPMR also present a distinctive pattern of location of employment. Urban growth has been accompanied by increasing price of the land which pushes activities that are more intensive in use of space out of the city of São Paulo. Thus, manufacturing activities have moved to more peripheral municipalities inside of SPMR while occupations in the service economy have concentrated in the city of São Paulo (Meyer et al., 2004). To control for the influence of specific characteristics of SPMR in the following analysis we use a long set of variables.
Data for this research were drawn from a sample of employed individuals from the 2010 Demographic Census of Brazil. This data base is released by the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e estatística (IBGE)), a federal government agency responsible for generating most of the country’s social information. A representative sample of the Brazilian population in 2010 contains information about 10.7% of the country’s households, which represents over 20 million people. For the SPMR, the sampling fractions of the municipalities that are a part of the region ranged from 5% to 20% of households. In 2010, the household sample provided a total of 576,174 occupied persons, who represent approximately 6.1% of the total employed persons in this metropolitan region who are over 10 years old (IBGE, 2012). 1
After considering only those observations corresponding to commuters, the final sample used in this study comprises 549,867 persons employed in the SPMR. The 2010 Demographic Census includes individual attributes, family and household characteristics, as well as occupational status.
Table 1 shows the variables considered for the analysis. These include family and household characteristics (e.g. household income and number of rooms) and characteristics associated with the employment sector and type of occupation. The data base includes in addition age, gender, race and education. These variables are regularly used in empirical research concerning the determinants of commuting time (e.g. Crane, 2007; Lee and McDonald, 2003). More generally, they are also used in mobility research, since personal characteristics relate to the ability to travel (e.g. age; Schwanen and Paez, 2010), and the distribution of occupations according to education in urban areas and/or potential spatial segmentation according to race (q.v. Kain, 1968). Variables associated with occupation type (informal employment, self-employed, entrepreneur and formal employment) and those associated with the business type (industry, services, construction and commerce) help to measure the influence of these occupation distributions and types of professional activity on commuting time.
Definition of the independent variables used in the analysis.
Given our interest on the family responsibility hypothesis, the focus of the analysis is on gender, marital and work status, and presence of additional household members.
Additionally, a set of variables relate to the characteristics of the place of residence (number of rooms in the household, number of residents and per capita household income) and the use of residential urban space. The inclusion of these variables is motivated by results in urban economics (Alonso, 1964). Traditional urban economics models (Fujita, 1989) indicate that land use intensity decreases with distance from employment centres and anticipate the possibility of urban segmentation of urban space according to income. The first occurrence is explained by higher price of the unity of space in centres, the second by the possibility of different reactions to the trade-off between accessibility and space according to income.
Descriptive statistics
Based on the sample of commuters for SPMR, Table 2 presents key descriptive statistics regarding division of labour, marital status and household heads for both genders, and presence of children. These statistics provide some preliminary information regarding the different dimensions of family structure and employment status in the sample. Thus, while female labour share is relatively high, it is still below that for males. Further, single women are more frequently occupied than married women. As discussed below, this helps to explain why a simple comparison between the commuting time of men and women may not reveal significant differences. We also note that most employed men are household heads. In contrast, less than 19% of all couples are headed by women. Finally, it is apparent that the percentage of employed persons with children was similar for men and women.
Composition of the occupied and family structure stratified by gender – São Paulo Metropolitan Region, 2010.
Source: Authors’ calculations based on 2010 Census microdata. Sample of employed persons. Occupied means that the person was working during the week of Census inquiry; participation rate corresponds to the weight of people in active age that are working; household head is the person recognised as responsible for residence by the others living in it; married is the person living with conjugal status with another and the presence of children correspond to the presence at home of at least a person under 16 years old.
Table 3 presents additional information regarding gender. There are some points worthwhile highlighting. With respect to work participation by gender and marital status we note that rates are higher for females in dual-earner households, while males are more likely to be in single-earner households. Relatedly, it is interesting to note that for the couples where only one partner works, around 68% of the working partner is the male. Conversely, of those formerly married approximately 61% of individuals were women.
Distribution of employed persons: gender and marital status categories (%) – São Paulo Metropolitan Region, 2010.
Source: Authors’ calculations based on 2010 Census microdata.
For each marital status classes in Table 3, Table 4 presents the distributions of females and males across commuting time categories. From Table 4, we note first that females are more frequently present in the shortest commute category (‘up to 30 minutes’), in agreement with evidence in other international studies. Furthermore, the statistics strongly suggest that, in the case of SPMR, marital status affects commuting time differently according to the gender.
Marital status and commuting time in São Paulo Metropolitan Region – Distribution among time commuting categories (%), 2010.
Source: Authors’ calculations based on 2010 Census microdata.
With regards to the latter point, it can be appreciated that gender differences tend to grow for married individuals and to diminish for single or formerly married ones. Furthermore, even for married individuals, it is also possible to note that the difference between genders is more salient for couples when only one member works. The statistics indicate that when both partners work, the percentage of employed persons that spend less than 30 minutes commuting increases from 51.8% to 53.0% for men and from 57.6% to 58.9% for women, relative to couples where only one partner work. This suggests that the distribution of commuting time was associated with the amount of work outside the home by the couple, as noted by Johnston-Anumonwo (1992). Finally, among the four marital status classes, both for female and male, the highest percentage of individuals in the shortest commute time category (‘Up to 30 minutes’) is found for individuals that are formerly married.
Table 5 presents the distribution of occupied workers in SPMR by the presence of other family members. It can be seen that around 26% of workers live in households with the presence of only one child (14 years old or younger). The proportion of households with more than one child is lower still. The numbers also show that around 10% of occupied individuals in SPMR live in households where there is an inactive senior. An inactive senior is an individual aged 65 years or older who does not work at home or outside the home. In principle, we believe that the presence of an inactive senior can reduce the burden of responsibilities and thus allow for longer commutes for other household members. In contrast, more than 50% of individuals live in households without children or inactive seniors.
Distribution of occupied workers by presence of family members (%) – São Paulo Metropolitan Region, 2010.
Source: Authors’ calculations based on 2010 Census microdata.
Table 6 presents the frequency of commuting time categories by presence of other family members. It is clear that gender differentials in commuting times tend to be smaller when seen from this dimension, as opposed to marital status. Nevertheless, even at the level of descriptive analysis, it is still possible to perceive that the presence of a child appears to affect differently the commuting time of females and males.
Family responsibility hypothesis variables and commuting time in São Paulo Metropolitan Region – Distribution among time commuting categories (%), 2010.
Source: Authors’ calculations based on 2010 Census microdata.
More specifically, as seen from the frequency in the category of the shortest commuting time, the presence of children is associated with increased commuting time in the case of males. The association is less clear in the case of females. The opposite situation appears to prevail for the case of the presence of inactive senior at home; here, the evidence suggests that this situation permits a longer commuting time for females and is associated with a shorter commuting time for males. Finally, we note that in the absence of both children and inactive seniors commuting times tend to be more egalitarian between genders.
Determinants of commuting time in SPMR
Modelling approach
In this section we present the results of the multivariate analysis of commuting time in SPMR. As noted above, the dependent variable is categorical and ordinal (i.e. commuting time categories). Therefore, an appropriate modelling technique is the ordered probit model. The dependent variable was organised in four groups of commuting time categories, as follows: (i) less than 30 minutes, (ii) more than 30 minutes and less than 1 hour, (iii) more than 1 hour and less than 2 hours and (iv) 2 hours or more.
The probit model is obtained by assuming that the random terms in an ordinal utility function follow the standard normal distribution. The model is well known and technical details can be consulted in several standard texts, including Train (2003).
Based on estimates for the parameters of the model it is possible to calculate marginal effects, which allow the measurement of the effect of explanatory variables on the probabilities that the individual’s commuting time corresponds to each of the different categories, i.e., that
where the β’s are parameters associated with covariates, the α’s are threshold parameters, and
Gender differences, household responsibility hypothesis and commuting time in SPMR
To implement the analysis we account for a large set of potential confounders. For example, individuals who are single are also generally younger, and individual age can potentially influence commuting time. In the same way, if there are gender differences by occupation sectors and if these sectors are distributed differently across the region, we may observe gender difference of commuting time. The results of estimating the model are displayed in Table 7. To assess the robustness of coefficients for gender, marital status and presence of dependents, we present three different models. For model (I), we consider only the effects of marital status variables on commuting time, in model (II) we also includes variables associated to the presence of children and inactive senior and for model (III) we include an extended set of confounders. In the models, variables are interacted with an indicator variable for female (i.e. a gender interaction effect). Therefore the coefficient for females, where such is the case, is the sum of the two columns.
Determinants of commuting time – SPRM, 2010 − coefficient estimates of ordered probit models (standard errors in brackets).
Notes: ** indicates p value < 0.05, * indicates p value < 0.1. Ro2 corresponds to the ordinal explained variation statistic proposed by Lacy (2006).
From Table 7 and model (I), we observe negative coefficients for both married and formerly married men and women, which implies that, relative to single man (reference), they have a higher probability of being in the categories of shortest commuting time and lower probability of being in the categories of longest commuting time. Since the coefficient for a female is the sum of the coefficients in the two columns, it can be seen that this effect is stronger for women. On the other hand, we initially obtain a positive coefficient for a single female.
Model (II) includes, in addition to marital status variables, variables related to the household responsibility hypothesis. From Table 7, we can see few qualitative differences for the coefficients associated to marital status variables (only the interaction coefficient for formerly married individuals is not significantly different from zero at conventional levels of confidence). In other words, the results regarding marital status are robust to the presence of children or inactive seniors in the household. The coefficients obtained for these variables also present important difference between genders. Specifically, while the presence of a child tends to increase commuting time (both for males and females), the negative coefficients obtained for the gender interactions indicate that these effects are stronger for males than for females. We also note a different effect on commuting time of females and males associated with the presence of an inactive senior: this tends to reduce commuting time for males and but it tends to increase the commuting time of females.
Model (III) includes an extended set of confounding variables. Including these confounders has implications for the statistical significance of marital status variables coefficients: with the exception of married individuals in single-earner households, being married or formerly married only matters for females (the direct effect associated with males is not significant). Furthermore, the effect is a tendency towards shorter commutes by females, relative to single males. Likewise, compared with single males, single females are prone to shorter commuting times.
To quantify the effect of the variables on the probit probabilities for each commuting time class, the marginal effects corresponding to model (III) are shown in Table 8. As noted above, marital status variables are significant only for females. Specifically, relative to single males, being a married female implies an increase of around 7% on the probability of being in the shortest commute category (up to 30 minutes) and a decrease of 1.5% on the probability of commuting more than 2 hours. There is little difference between married females in single- or dual-earner households. Confirming the importance of the married condition for women commuting time, we also note that single females are 3% more likely to be in the shorter commute category relative to single men.
Determinants of commuting time – Marginal effects of the variables of the ordered probit model – SPRM, 2010.
Notes: ** indicates p value < 0.05, * indicates p value < 0.1. The marginal effects are obtained for coefficients of model (III) of Table 7 and they represent mean of marginal effects across observations.
For the variables regarding the presence of dependents we also observe very different marginal effects for females and males. Thus, the presence of children has a small and negative or not significant effect on the probability of a male having short commutes. For females, in contrast, the presence of one, two and three or more children, increase the probability of shorter commutes by 0.7, 1.7 and 2.2%, respectively. Intriguingly, as noted above, the presence of an inactive senior at home has the opposite effect, since it pulls the male towards the home (i.e. an increased probability of shorter commutes) but releases the female for longer commutes (i.e. a slightly reduced probability of shorter commutes).
Over all, the results appear to confirm for the case of SPMR the importance of family structure for explaining shorter commutes observed for females. Married females tend to have shorter commutes, and the probability of a female’s commute to be in the shortest category is further increased by the presence of children. Thus, even in an urban environment characterised by the strong presence of domestic employment, these results are consistent with those obtained by Lee and McDonald (2003) in South Korea and Crane (2007) in US metropolitan areas.
Nevertheless, our results also indicate that single and formerly married women are prone to shorter commutes relative to single or formerly married men. These results are not explained by our long set of control variables. Thus, it appears that in the case of SPMR the shorter commuting time of women cannot be entirely attributed to marital status and to the presence of dependents at home. A possible complementary explanation could be the perception or reality of crime and safety in SPMR, which may affect the disposition of females to commute longer. Supporting evidence for this conjecture is offered by Moura (2012), who finds that longer commuting times are positively associated with the probability of being the victim of a violent robbery in a set of major Brazilian cities.
Interestingly, from Table 8, we also note that there are gender differences in the way other attributes affect commuting time. Specifically, for ethnicity, education and age variables we have obtained additional gender effects that increase the tendency of females towards shorter commutes. This suggests that the situation for females in SPMR is influenced by marital status and the presence of children at home, but is also affected by broader cultural differences regarding participation in society by members of both sexes, and as noted above in relation to the risk of crime, to how urban environments are experienced by females.
Two additional results are worthwhile highlighting, because they stand in contrast to previous findings.
First, we note that a white male has a slightly higher probability of being in the shortest commute class, relative to a non-white male (by approximately 3%). This effect is larger for females (+4.3%). This is in contrast to what has been found in American cities (Crane, 2007) where whites tend to display longer commutes. The result, however, is not unexpected, and reflects a key aspect of Brazilian cities, where income tends to be centralised (Nadalin and Iglori, 2010).
Second, Madden (1981) observed that females employed part-time tended to have shorter commutes. This was explained as the outcome of a calculation whereby a long commute would not be deemed worthwhile for just a few hours of work. In contrast, we find that females who are employed part-time in the SPMR have a higher probability of commuting longer times. This can be explained by the large proportion of domestic workers, who tend to be low-income females who live far away from employment opportunities (i.e. affluent residential areas).
Concluding remarks
The objective of this paper has been to investigate the determinants of commuting time in SPMR. A particular focus of the research was on commuting gender differentials that have been observed in case studies in the international literature. This issue is important because of the changing economic fortunes of Brazil in recent years, and the implications for female participation in the labour force. The international literature contains a large number of studies in North America. However, as suggested by Lee and McDonald (2003) research in other national settings can help to identify familial and social structures that may operate differently across countries and/or cultures.
The results obtained for the case of São Paulo Metropolitan Region indicate that marital status is a significant predictor of commuting time for females. While relative to males, females are more likely to have shorter commutes regardless of household structure, the effect is more pronounced for married females especially in dual-earner households. Similar to the findings of Lee and McDonald (2003) in Seoul, the presence of children also seems to decrease the ability of females to commute longer distances. As noted above, the same is not observed for males. In general terms, our results are consistent with the household responsibility hypothesis. It must be noted, however, that single and formerly married women are also prone to shorter commutes relative to similar males. Therefore, while the household responsibility hypothesis seems to be part of the reason for shorter commutes of females, there appear to be broader cultural and environmental factors. The results may in this way be reflective of attitudes regarding gender in Brazil, whereby men tend to be socially dominant and women are expected to be self-sacrificing and submissive (DeSouza et al., 2004).
It is worthwhile to note some of the potential implications associated with the results of the present study. First, it is important to note that although the marginal effects are relatively small (consistently less than 10%), the aggregate effect over millions of residents in São Paulo Metropolitan Region is not negligible. On the one hand, as the economy of the region continues to grow, there are implications for the ability of females to find suitable jobs, since shorter commute times imply a preference for proximate jobs, or the inability to exercise geographically more extensive job searches. For businesses, this may introduce strictures in their access to a suitable labour pool. At the same time, there is a downward trend in the percentage of married women among employed females in recent years (46.2% of employed women were married in 2001 and 41.6% in 2009). Since single women are less likely to be in the shortest commute category, it is possible that overall the commuting time of women in the SPMR will increase in the coming years, which can directly affect congestion in the region, and the quality of life of females in particular. The net effect is not apparent, but we suggest that this is a matter for further research.
Footnotes
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors