Tell me who's your neighbour and I'll tell you how much time you've got: The spatiotemporal consequences of residential segregation

Abstract Relying on data from the United States Census and the American Time Use Survey (2010–2017), we examine how residential segregation influences per capita discretionary time availability in Los Angeles, New York City and Miami. We find a sizable disadvantage of being Latinx for discretionary time availability. Non‐Latinx Whites have 182 extra hours of per capita discretionary time per year than do Latinx individuals. Both within‐neighbourhood and adjacent‐neighbourhood influences matter. In most neighbourhoods, segregation is correlated with having more discretionary time. Individuals in highly segregated areas have approximately 80 more hours of discretionary time per year than those living in diverse areas. This suggests that in addition to socioeconomic, cultural and well‐being benefits, ethnic enclaves may also impart temporal advantages. However, we find that there may be diminishing marginal returns with increasing segregation in surrounding areas. Sociodemographic characteristics explain over one‐quarter of the variance between segregation and discretionary time availability.

This study seeks to reorient current theorizing on Latinx residential segregation by centering sociotemporal inequalities.
We introduce the idea that it is imperative to consider time when theorizing about how ethnic residential segregation affects needed resources. Focusing on per capita discretionary time left over after meeting daily survival needs, we examine the puzzle of ethnic residential segregation and discretionary time availability. Though we know that neighbourhood-level sociodemographic characteristics are central for the social processes that drive stratification (Castañeda et al., 2015;Harding, 2007), the spatial distribution of time availability is still understudied (Castañeda, 2018). Methodological complexities and data limitations have thus far prevented us from concretely examining how ethnic segregation shapes time use disparities (Castañeda, 2018). This is unfortunate, since discretionary time is necessary for combating entrenched inequalities, garnering socioeconomic resources, and for all aspects of individual and community well-being (Giurge et al., 2020;Goodin et al., 2008;Kalenkoski & Hamrick, 2013;Williams et al., 2016).
Furthermore, neighbourhoods are not isolated islands. They are influenced by each other: neighbourhood mobility and sociodemographic patterns are altered by the ethnic compositions of adjacent neighbourhoods (Crowder & South, 2008;Wilson & Taub, 2007). Yet, we still do not know how ethnic segregation in adjacent neighbourhoods may shape discretionary time availability in a nearby neighbourhood. Ethnographic research suggests that segregation matters for betweenneighbourhood time use, influencing the well-being of residents (e.g., commuting patterns and waiting times) (Castañeda, 2018;Edwards, 2017). However, until now, we have been unable to examine this systematically via large-scale quantitative data.
Thus, our understanding is incomplete when it comes to how the interplay between ethnic segregation and neighbourhoodlevel sociodemographic conditions shapes discretionary time availability.
We begin by putting prevalent theoretical perspectives from the neighbourhood effects literature in conversation with theories on segregation and from the sociology of time, explicitly touching on the hypotheses undergirding our study. Next, we provide an overview of the data, followed by our downscaling method producing local estimates of discretionary time, then review and compute our segregation measures. We ask three related questions: 1. Does per capita discretionary time availability vary spatially, and if so, does residential ethnic segregation matter for discretionary time availability? 2. How do sociodemographic characteristics influence the above?
3. Do the characteristics of adjacent neighbourhoods affect the relationship between segregation and time availability? 2 | DISCRETIONARY TIME AND NEIGHBOURHOOD EFFECTS While no one has more than 24 h in a day, not everyone has the same amount of discretionary time. We offer a more precise operationalization of discretionary time in our methods section, but the construct can be efficiently summarized as the time left over for discretionary activities after the minimal amount of time needed to satisfy bodily, financial and household needs has been spent.
Overall, those with more economic and cultural resources have more control over their time and more autonomy over their discretionary time. Those with less resources spend extra time to make enough money. Those with higher status can purchase time by outsourcing menial tasks (Hamermesh, 2019). Thus, discretionary time is intimately intertwined with social inequalities. Per capita discretionary time availability is a highly responsive measure to sociodemographic constraints and opportunities, while also being sensitive to historical and structural conditions: when structural inequality impedes on discretionary time availability, this has both individual and societal ripple effects (Goodin et al., 2008;Hamermesh, 2019;Williams et al., 2016).

| Ethnic residential segregation and discretionary time availability
Residential segregation refers to the sorting of large, homogeneous groups of people into specific spatial concentrations (DuBois, 1899; Massey & Denton, 1988). Explanations as to how residential segregation affects the lives of the Latinx population in the United States can be roughly grouped into two categories: studies looking at the detrimental effects or those noting potential benefits. Existing literature has not explicitly examined how ethnic segregation may be detrimental for discretionary time availability, but it does point to multiple key propositions.
Residential segregation may undermine the socioeconomic outcomes of the Latinx population, by constraining them to live in neighbourhoods with less public investment, higher poverty, underemployment and unemployment, underfunded institutions and limited access to stable jobs (Borjas, 1994;Castañeda, 2018;Edwards, 2017). Socioeconomic status (often measured by income and education attainment levels) may force individuals to live in more impoverished neighbourhoods. These neighbourhoods have less stable, owner-occupied housing, higher population density and high levels of ethnic segregation (Alba & Foner, 2015;Castañeda, 2018). The above could lead to a higher likelihood of segregation being detrimental for per capita discretionary time availability, as individuals may need to spend extra time to mitigate the effects of neighbourhood instability by commuting longer distances and working multiple jobs (Castañeda, 2018;Cutler et al., 2008;Edin et al., 2003;Giurge et al., 2020;Holzer, 1991).
Lower-income, segregated neighbourhoods also tend to experience disproportionately higher rates of disruption (Sharkey et al., 2014). These can range from lower high school completion rates due to familial stressors to higher rates of single women-led households. In such situations, extended family members may need to take on the responsibility of caring for dependents at the expense of their per capita discretionary time availability (Feldmeyer, 2010;Shihadeh & Barranco, 2010). Thus, ethnic residential segregation may further exacerbate already-present gender differences in free time (Bianchi et al., 2000, Castañeda, 2018. Women living in segregated neighbourhoods may need to take on extra jobs to support their families, cutting into their discretionary time availability at home. Women may also feel more obligated to perform gendered labour for extended family nearby (Pinto & Ortiz, 2018).
However, research on ethnic enclaves has suggested that in some circumstances, ethnic segregation may be beneficial.
The presence of extended kinship networks in ethnically segregated communities could potentially distribute economic risks (Menjivar, 1997). People choosing to live in communities where their ethnicity is over-represented may tap into social networks more efficiently. These could include job connections, carpool networks and emotional support, reducing the likelihood of family instability (Fomby et al., 2010). This could increase discretionary time availability as community members could receive time and money-saving assistance. Residential segregation could also contribute to community well-being by offering eldercare, childcare and market labour (Waldfogel, 1999).
Through the above mechanisms, residential ethnic segregation can create self-contained markets for ethnic goods and access to coethnic sources of capital, leading to higher median income levels, lower unemployment rates and higher rates of owneroccupied housing (Cutler et al., 2008;Edin et al., 2003;Menjivar, 1997). This could have positive consequences for discretionary time availability.
Ethnically segregated neighbourhoods may also increase discretionary time availability in more tangible ways. The ability to speak a common language makes a concrete difference when it comes to efficiently navigating one's environment. When norms and customs are familiar to all and communities can establish ethnic institutions (churches, stores, etc.), common economic and well-being goals are accomplished more efficiently (Fine, 2002;Lazear, 1999). This is particularly beneficial for lower SES individuals (Edin et al., 2003). Thus, residential segregation could reduce the chance of lacking needed discretionary time.

| ADJACENT NEIGHBOURHOOD EFFECTS AND DISCRETIONARY TIME AVAILABILITY
While multiple researchers have explored how segregation may matter for neighbourhood-level infrastructure and resources (Crowder & South, 2008;Denton & Massey, 1991), it is unclear how segregation in adjacent neighbourhoods may influence discretionary time availability in an area. This is imperative to consider, as existing research shows that neighbourhoods influence each other, with residential segregation having important adjacentneighbourhood consequences (Castañeda, 2018). This literature can be sorted into two broad categories: studies looking at spillover effects and those focusing on containment. When we refer to spillover effects, we mean that the characteristics of one neighbourhood may influence those of an adjacent neighbourhood. Studies on the spillover effects of segregation have found that both proximity and population size matter. The proximity of a neighbourhood to one with a highly segregated large minority population strongly predicts the likelihood of future change in local neighbourhood characteristics (Denton & Massey, 1991). This can ensue from 'white flight', resulting in minority population members moving into an adjacent neighbourhood previously inaccessible to them (Crowder & South, 2008).
Ethnically segregated areas can also be influenced by conditions prevalent in the neighbourhoods adjacent to them. For example, if group-level inequality is high between two neighbourhoods, these conditions may contribute to housing instability, displacement, and the necessity to dedicate needed discretionary time to labour, just to be able to afford continuing to live in one's own neighbourhood (Ocejo, 2011;Policy Link, 2016).
To understand how ethnic segregation in adjacent neighbourhoods could spill into the time use profile of another, we need to consider that time and money are distinct while being intimately intertwined (Sampson et al., 1999). For example, an economically well-off, less segregated locality may 'buy time' via care or household labour from a nearby economically less well-off, potentially segregated neighbourhood. This would lead to the necessity for those living in the segregated, less economically privileged area to commute to other neighbourhoods for work, influencing their discretionary time.
Adjacent neighbourhood characteristics could also have the opposite effects on local characteristics. They may curtail the tendency for one population to interact with adjoining populations, containing population and neighbourhood-characteristics to a particular area (Lee, 1966;South & Crowder, 1997). When it comes to the containment of discretionary time availability to neighbourhoods, we can think of this as such: The concentration of an ethnically segregated, possibly impoverished group to a neighbourhood may lead members of nearby more economically well-off, perhaps less ethnically segregated neighbourhood to avoid the area. This will lead to the entrenchment of both economic and discretionary time scarcity in the area avoided, as the likelihood of economic investment BÓ AND DUKHOVNOV | 3 of 18 and job creation decreases (Lee, 1966;South & Crowder, 1997).
Here, individuals may need to work harder to survive: juggling multiple low-paying jobs, while navigating crumbling infrastructure and institutions. Their efforts will influence their per capita discretionary time availability.
The ethnic composition of adjacent neighbourhoods may also affect the conditions of a local neighbourhood (Crowder & South, 2008). In the case of discretionary time availability being contained, this could happen through ethnic segregation undermining between-neighbourhood time exchanges. Or conversely, an ethnically segregated neighbourhood may be more socially cohesive, which could contain discretionary time availability to the area. In this scenario, members of segregated yet cohesive neighbourhoods would be less likely to form new social ties (which require time to establish) in adjacent neighbourhoods (DiPrete et al., 2011). As the literature above illustrates, maintaining focus on discretionary time availability is imperative for a better understanding of the repercussions of ethnic segregation.

| RESIDENTIAL SEGREGATION AND INEQUALITY
Ethnic residential segregation is consequential for multiple reasons: it influences how individuals navigate and experience neighbourhoods, undergirds the stability of social networks, shapes health outcomes, access to resources and interactions with institutions (Bayer et al., 2008;Durlauf, 2004;Massey, 2001). The above consequences all have the potential to influence inequalities in discretionary time availability. In lieu of outlining the extensive body of existing literature on residential segregation, we review aspects relevant to our study.
The existing literature is in disaccord as to whether segregation is harmful or beneficial for the Latinx population when it comes to their noneconomic outcomes. Some argue that segregation for them may be beneficial (network cohesion, access to fresh vegetables, etc.), while others assert that as the segregated Latinx tend to be disproportionately exposed to neighbourhood disadvantage, this carries detrimental health consequences (Do et al., 2017). This presents a puzzle: Latinx ethnic residential segregation may have different consequences, depending on how it shapes the noneconomic outcome of discretionary time. If ethnic residential segregation is beneficial for discretionary time availability, it may be beneficial for population-level mental and physical well-being.
However, if ethnic residential segregation is detrimental for time availability, the segregated group may be doubly disadvantaged (Giurge et al., 2020;Rumbaut, 2011).
There are several reasons to suspect that ethnic segregation might have an independent effect on per capita discretionary time.
Starting with Durkheim, multiple researchers have argued that distinct groups experience and share time differently (Durkheim & Swain, 2008;Gell, 1992). Segregation creates self-contained social groups, curtailing a group's ability to meaningfully interact with others outside of their segregated social networks, limiting betweengroup transmission of time and temporal perspectives (Bourdieu, 1964;Echenique & Fryer, 2007;Massey & Denton, 1988). Notwithstanding the extensive literature on the topic, pressing gaps remain.
Despite knowing that economic factors are salient, and that discretionary time is as important for individual and societal wellbeing as money is, we do not know how structural conditions-such as ethnic segregation-influence the necessary resource of discretionary time. While we know that individual-level sociodemographic variable shape time availability, how neighbourhood-level sociodemographic characteristics-such as population density and housing type, group-level income inequality, median income, unemployment rates, area-level education attainment and household compositionmay influence the relationship between segregation and per capita discretionary time is still uncharted. Similarly, though the existing literature has noted that neighbourhoods influence each other (Crowder & South, 2008;Wilson & Taub, 2007), it is unclear how segregation in adjacent neighbourhoods, along with their socioeconomic environment, may predict discretionary time availability at a given location.
To study the relationship between ethnic segregation and per capita discretionary time, we use data from three highly segregated localities in the United States when it comes to Latinx segregation: New York City, Miami and Los Angeles (Alba & Foner, 2015). Many areas in these cities rank high on the Information Theory Index, an entropy-based measure of diversity capturing how segregation varies between neighbourhoods (Massey & Denton, 1989;Reardon & Firebaugh, 2002). Specifically, we use the above cities' labour market commuting zones, as they delineate local experiences better (as counties reflect political borders) 1 . Though highly segregated, these localities also contain integrated areas. Further, our labour market commuting zones house dissimilar Latinx groups, allowing us to honour the cultural heterogeneity of the population. The Los Angeles region contains individuals with mainly Mexican and Salvadoran origins. Caribbean-origin Latinx are prevalent in the New York City area, whereas many with Cuban origins live in the Miami region (Alba & Foner, 2015). These localities serve as great case studies for the examination of the linkages between Latinx residential segregation and per capita discretionary time availability.

| DATA
Our analyses rest on two data sources. The first consists of 8 consecutive years of the ATUS (2010-2017, N ≈ 96,000). This is a nationally representative, diary-based survey of American adults, recording the amount of time spent on various activities, where, and with whom. The annual survey sample is drawn from the participants of the Current Population Survey (CPS), conducted by the U.S. Census Bureau. A crucial feature of the survey is geographically 1 https://www.ers.usda.gov/data-products/commuting-zones-and-labor-market-areas/ linking respondents to their county of residence. This link exists for 10% of counties, encompassing the most populous suburban and urban neighbourhoods. This feature undergirds our methods of choice and the scope of our analyses.
Our second data source is the 2010 U.S. Census county and census block TIGER shape files-with corresponding population attributes by age, sex, race and ethnicity-assembled by the National Historical Geographic Information System (NHGIS) (Manson et al., 2017). This enables us to scale down the time use rates to the specific demographic groups (considering age, sex and ethnicity) living in the subcounty local areas, and to calculate our segregation indexes. Both datasets allow individuals to self-select into our Latinx group if 'they are Hispanic or Latino, of any race'. 2 6 | METHODS

| Discretionary time availability by demographic group
Our analysis centers three highly ethnically segregated localities in the United States: Los Angeles, CA (LA), Miami, FL and the New York City, NY (NYC). Since the CPS only has geographic links for respondents in populated urban and suburban counties, this precludes us from including the rest the country. Following precedent, we define discretionary time as time dedicated to leisure, socializing, religious activities and exercise (Goodin et al., 2008;Williams et al., 2016). For each city, we consider 12 subgroups: two ethnicity categories (Latinx and non-Latinx White), two sexes (female and male), and three age groups (30-44, 45-54 and 55-69). The age groups capture seminal stages of working adulthood: early middle age, middle age and early late adulthood (Medley, 1980). Our choice was also influenced by sample size constraints, as the introduction of more detailed breakdowns would have led to dwindling sample sizes for select subgroups. This allows us to generate stable estimates for our 12 groups. Next, we introduce our downscaling method, using ATUS county-level measures to approximate discretionary time for standardized, small areas.
First, we create a standard unit to which county-level estimates of discretionary time will be downscaled. We downscale ATUSderived rates to 1-km-wide hexagonal grids. These hexagonal grids serve as the unit of analysis in our models. Our choice was guided by precedent, heeding the call to shift focus from census tracts as proxies for neighbourhoods, to more theoretically meaningful, standardized, comparable units of analysis (Chaix et al., 2005;Riley, 2018). Such continuous spatial surfaces have been used to characterize neighbourhood contexts in work examining the geographic distribution of resources (Chaix et al., 2005). Sensitivity analyses also show that 1-km-sized grids exhibit the lowest root mean squared error, producing the most accurate population characteristics (Lloyd et al., 2017). This is relevant, as census tract sizes can vary considerably within a county. Our uniform-sized grids correct for this, allowing for accurate comparisons between areas.
This scale does not require additional statistical techniques to regularize the data for comparability (Le Bras, 2008). As an added advantage, hexagonal grids correspond to coastal regions better than square grid cells, which is relevant given that our cities have extensive coastal areas. In sum, for our purposes, grids are superior alternatives to irregular geography (such as census tracts or block groups) (Lloyd et al., 2017).
Next, we fill the standardized hexagonal grids with our populations of interest. We spatially overlay the census blocks along One assumption that underlies this process is that the population is distributed homogeneously throughout a census block. This a reasonable assumption, given that our 1-km-wide hexagonal grids are comparable to the population size of census tracts in places with high-to-moderate population density. Most urban blocks are quite small, generally containing housing of similar density and type, minimizing error.
Third, we generate county-level estimates of discretionary time, employing recommended ATUS methodology (Bureau of Labour Statistics, 2020, p. 37).

| Ethnic segregation
To measure ethnic segregation, we rely on two widely applied measures of unevenness and inequality. First, we adopt a measure of inequality developed by Roberto (2015), the Divergence Index (Roberto's D). To ensure that our results are comparable with much of the literature on residential segregation, we also use the Information Theory Index (H), describing the relative diversity or homogeneity in an area (Reardon & Firebaugh, 2002). Although both measures are area-decomposable, allowing for multi-group score calculations, we are interested in Latinx-White segregation. Thus, we calculate two-group D and H indices, for Latinx and non-Latinx Whites. The H index is particularly fitting for our analysis, as it compares the degree of evenness in one group relative to the population of both groups (Massey & Denton, 1989). The U.S. Census Bureau also uses the H index to measure how large, homogeneous groups of people are sorted into specific spatial concentrations (Massey & Denton, 1988). 6 Alternatively, the Divergence Index indicates how 'unusual' is the entropy of an observed value, relative to the theoretical/expected value (Roberto, 2015), reflecting segregation as the over-or under-representation of population groups in a local area. Both indices are robust to the size of the underlying population, allowing for the comparison of various areas.
The H index is top-bounded by the maximum value of 1, denoting maximum relative homogeneity. Although the lower bound is typically 0, the minimum possible value can be negative, representing relative 'hyper-integration' or a complete mixing of groups within an area. The Divergence Index is strictly bounded between 0 and 1, denoting maximum integration and maximum segregation. To achieve the greatest possible accuracy in our D and H indexes, we use census blocks as underlying geographic units to proportionally allocate the population to the cells that contain them.
Following Reardon and Firebaugh (2002), we first compute individual-cell entropy scores where E i is the entropy score of a cell i within metropolitan region, and π g is the proportion of group g (Latinx or non-Latinx White) in the cell. To obtain the Information Theory Index for each cell, we compute the difference between the cell and regional level entropy scores, standardized by regional entropy.
F I G U R E 1 Downscaling and spatial adjustment procedures 6 https://www2.census.gov/programs-surveys/demo/about/housing-patterns/multigroup_ Following Roberto (2015), the Divergence index for a cell is calculated as: where π ig is the proportion of group g population in cell i and π g is the proportion of group g in the overall population.

| Spatial modelling
Since we are interested in predicting the gap on average discretionary time availability between two populations, while accounting for local clustering or spillover effects, we rely on spatial econometric models, namely cross-regressive (local) Spatial Lag-X (SLX), simultaneous autoregressive (SAR) and Spatial Error (SEM) models (Anselin, 2002). Ordinary linear mixed-effects models (Table SA4) with individual-level and county-level group effects justify our use of individual attributes, such as age, sex and ethnicity as the basis on which to spatially downscale discretionary time. At the same time, such models do not allow us to carry out individuallevel analysis on sub-county geography. Moreover, segregation is a descriptive attribute of an area, such as a neighbourhood, and its impacts are more conceptually meaningful on that level, rather than a large administrative unit, like a county. Importantly, smaller geographic scale ensures that variability in local associations between discretionary time and segregation are not lost to 'averaging out' at a county level. Unlike conventional Ordinary Least Squares (OLS) models, our spatial models account for autocorrelation in the structure of data: SLX focuses on the local association of observed covariates, SAR emphasizes the global spatial dependence between the dependent variables, whereas SEM focuses on spatial autocorrelation in the error term. The three spatial models are complementary. For all three spatial models, we compute queen (1) contiguity weights matrices, meaning that neighbouring pairs are identified between immediately adjacent cells if they share a common border or point. Other forms of spatial contiguity or orientation (e.g., rook) may be used, there is overall no inherent reason to constrain spatial dependence structure for time availability, nor we have reasons to believe the effects would spread beyond immediate neighbours. The cross-regressive Spatial Lag model is represented by the following equation where X is a n k × matrix of independent variables with a corresponding vector of coefficients β, WX is a cross-regressive term obtained through multiplication of n n × matrix of weights regulating the effect of neighbouring covariates X on the time availability outcome of a cell. The SAR model is defined by the following equation where Wy is the autoregressive term scaled by the global spatial correlation coefficient ρ, indicating the degree of spatial dependence between the lagged dependent variable. In other words, the model predicts the impact of time availability in one cell, because of time availability patterns in surrounding regions. The SEM is shown in the following equation in which u is the error term with nonspherical error variance, and λ is the error parameter. Following precedent, our model selection is based on the global Moran's I test for spatial autocorrelation and Lagrange multiplier tests for Spatial Lag and SEM (Anselin, 2002).
Both tests indicate that our models are good candidates for explaining how ethnic segregation matters for discretionary time availability (p < 0.0001), considering spatial dependence between the variables. We also perform the recommended spatial Hausman test for model selection, which is designed to indicate preference for the SEM model over OLS (Pace & LeSage, 2008

| Control variables
In addition to the above descriptive analyses, we also examine neighbourhood-level drivers of discretionary time availability, by taking both segregation and macro-level sociodemographic characteristics into account. Covariates include Divergence Index (Table 3), H Index (Table SA5)

| Residential segregation and the spatial variation of per capita discretionary time
We find that the neighbourhoods located inside the urban core of both New York City and Los Angeles are highly ethnically segregated, with ample variation between neighbourhoods. Miami's urban core is less ethnically segregated than the areas immediately next to it.  There is significant variation across the regions when it comes to the spatial distribution of discretionary time. Figure 3 illustrates that compared to non-Latinx Whites, there is a sizeable disadvantage of being Latinx when it comes to discretionary time availability. Our descriptive spatial results show that there is likely a relationship between discretionary time availability and residential ethnic segregation. 8

| Sociodemographic characteristics and neighbourhood influences
Zooming in from the bird's-eye-view of the figures, Table 1   neighbourhood and adjacent neighbourhoods. For example, focusing on the Divergence Index for NYC, the difference between a typical low-segregated and high-segregated area is 16 min of per capita discretionary time per day (assuming a 60-percentage point difference between the two segregation states). In other words, a person from a highly segregated area has 16 more minutes of discretionary time per day than their counterpart living in a highly diverse area. Per average week, this translates to over an hour-and-a-half of discretionary time. In a year, this gives us approximately 82 extra hours of discretionary time, pointing to a both substantively and statistically important difference.
Our SAR model is governed by a high degree of spatial autocorrelation (ρ > 0.65) . This is not surprising, given that our downscaling procedure involves adapting county-level rates of discretionary time to the underlying cells, under the assumption of spatial uniformity. The SAR model would be more appropriate with T A B L E 2 Ordinary Least Squares (OLS), Spatial Lag -X (SLX), simultaneous autoregressive (SAR) and spatial error ( (Cutler et al., 2008;Edin et al., 2003;Fine, 2002;. It is likely that in addition to distributing economic risks, the presence of extended kinship and close-knit peer networks in ethnically segregated communities translate to time-saving and time-sharing benefits (Menjivar, 1997 This is outside the scope of this paper, and we were also restricted by data limitations. BÓ AND DUKHOVNOV | 15 of 18 are possible. With greater availability of spatial time use data, it would be possible to include additional factors (e.g., income or household structure), from which to compute estimates for downscaling. Likewise, improvements in spatial accuracy could be achieved with a yet more nuanced post-downscaling adjustment that includes as its smoothing weight a composite index of individual or area-level variables that could impact the availability of discretionary time.
Lastly, future research needs to explore alternative spatial models and specifications with additional areal covariates that could provide superior explanatory power.
Researchers need to revisit the issue of gender when it comes to how segregation affects time availability. As our unit of analysis is the 1 km grid, we focus on how macro-level, contextual factors matter, relying on spatial modelling to understand the geographic distribution of segregation and time availability. We include the effect of percentage of households headed by single women (a socioeconomic disadvantage indicator), but this is not the same as fully tackling gender differences when it comes to how ethnic segregation shapes time availability. 10 Knowing the existing differences in discretionary time availability between the genders (Goodin et al., 2008), there is a possibility that ethnic segregation may matter for women's time availability differently. This is a highly policy-relevant topic that needs more nuanced examination.
Our understanding of how ethnic residential segregation influences life chances remains incomplete without an examination of how it matters for sociotemporal inequalities. Our study contributes to the current literature in several ways. To our knowledge, no prior studies have put neighbourhood and adjacent-neighbourhood effects in conversation with how ethnic segregation may shape the necessary resource of discretionary time. Our findings highlight that is worthwhile to put prevalent theoretical perspectives from the neighbourhood effects literature in conversation with theories on ethnic residential segregation and from the sociology of time. By doing so, we illustrate that it is imperative to consider discretionary time as a resource when theorizing about how ethnic segregation influences other resources. We specifically focus on honouring the experiences of the Latinx population, as considering their population size, their discretionary time availability profiles are vital for shaping the overall time use profile of the U.S. (Bernstein, 2013;Lichter et al., 2015). Our work serves as an important first step in tracing how temporal processes are connected to processes of stratification.