Residential mobility is often viewed as a disruptive process in the lives of children and adolescents, as moving often means severing ties with friends, schools, and communities (Raviv et al. 1990
). An overwhelming body of literature suggests that adolescents who move tend to experience higher levels of psychological duress, perform poorly in school, and display higher levels of problematic behaviors relative to their residentially stable peers (Haynie and South 2005
; Metzger et al. 2015
). As such, there is a long tradition of social science research aimed at identifying the factors that place some adolescents at a higher risk of moving than others (e.g., South et al. 1998
; Gasper et al. 2010
; Porter and Vogel 2014
). Scholars typically point to the role of family and community factors as ‘pushes’ that motivate relocation. For instance, changes in family structure or parental employment have been consistently associated with mobility intentions and follow-through (Lee et al. 1994
; South and Crowder 1997
; Geist and McManus 2008
; Vandersmissen et al. 2009
; Feijten and van Ham 2013
), as have broader indicators of neighborhood quality, such as racial composition, socioeconomic disadvantage, and crime rates (South and Crowder 1997
; Xie and McDowall 2008
; Hipp and Steenbeek 2015
Emerging research suggests that characteristics of sending and receiving neighborhoods may be more consequential than the simple act of moving (Sharkey and Sampson 2010
; Vogel et al. 2017
). This research has primarily focused on the effects of moving into and out of deprived neighborhoods, as prolonged exposure to socioeconomic disadvantage has been consistently linked to detrimental outcomes in later life (Ellen and Turner 1997
; Leventhal and Brooks-Gunn 2000
; Clark and Morrison 2012
). Generally speaking, this literature demonstrates that families living in impoverished neighborhoods move with greater frequency than those from more affluent areas, and, perhaps not surprisingly, adolescents who are able to leave distressed neighborhoods fare better than those who remain (Crowder and South 2011
; Roy et al. 2014
). It seems that residential mobility, especially when coupled with downward shifts in neighborhood quality, has profound implications for adolescent development and well-being.
As we elaborate below, two observations complicate prior research in this area. For one, selection bias looms large in both the neighborhood effects and mobility literatures, as there are a number of factors that influence the neighborhoods in which people live as well as their subsequent likelihood of moving out (Sampson 2002
; Van Ham and Manley 2012
). It remains unclear whether and how neighborhood characteristics influence mobility above and beyond the factors that disproportionately funnel families into particular communities. Second, much of the extant research has treated neighborhoods as islands, examining the effect of residential
neighborhood characteristics on subsequent mobility. However, neighborhood processes are rarely spatially independent (Graif et al. 2014
) and given the high levels of segregation in many American cities, socioeconomic disadvantage in areas that are nearby residential neighborhoods—oftentimes referred to as extralocal neighborhoods (e.g., Crowder and South 2011
)—likely influence the ability of families to relocate. However, limited research has examined the role of spatial dynamics in the relationship between neighborhood deprivation and residential mobility.
The present study builds upon prior research by considering the relationship between neighborhood socioeconomic disadvantage and residential mobility among a representative sample of American adolescents. We move beyond existing scholarship by employing a longitudinal spatial modeling strategy that allows us to assess how (1) changes in socioeconomic disadvantage between sending and receiving neighborhoods and (2) levels of disadvantage in extralocal neighborhoods influence mobility. To our knowledge, this is the first attempt to simultaneously account for selection bias and spatial dynamics in the relationship between neighborhood deprivation and mobility.
3. Current Study
While a large body of literature demonstrates a seemingly clear link between neighborhood deprivation and residential mobility, prior work is limited by the threats posed by selection bias and the inability to consider the interactions between broader community features and residential neighborhood characteristics. Thus, it remains unclear whether the oft-observed relationship between neighborhood disadvantage and subsequent mobility reflects a true neighborhood ‘effect’ or if it is attributable to the differential sorting of residentially instable individuals and families into socioeconomically deprived neighborhoods. Moreover, even if the threats posed by endogeneity are negligible, it is unclear how levels of disadvantage in extralocal areas enable or constrain mobility. In this sense, the spatial patterning of neighborhood disadvantage may have profoundly different effects on residential mobility if neighborhoods in the nearby community differ from the neighborhoods in which people currently reside.
Guided by these limitations, the present study builds upon and moves beyond existing scholarship in two key regards. First, we utilize a spatial modeling strategy that allows us to specify unique effects of socioeconomic disadvantage in residential and extralocal neighborhoods. This allows us to (1) compare the effects of local and extralocal neighborhood disadvantage on adolescent mobility and (2) examine the degree to which socioeconomic disadvantage in extralocal areas conditions the effect of local neighborhood disadvantage on out-mobility. Second, we employ a hybrid random effects model that allows us to decompose within and between individual differences in the effect of local and extralocal neighborhood deprivation on residential mobility. Unlike more traditional regression models, this approach allows us to hold constant time-invariant individual differences that might differentially sort certain families into specific neighborhoods.
Our spatial modeling strategy allows us to detect community influences that would otherwise be missed if we considered only residential neighborhood characteristics. The choice of the hybrid regression model allows us to examine not only how neighborhood deprivation influences mobility on the whole, but how changes in neighborhood socioeconomic status between sending and receiving neighborhoods influence subsequent mobility. Departing from prior studies, this empirical framework allows us to address not only selection into neighborhoods, but also selection processes that differentially sort individuals into neighborhoods characterized by varying proximity to other disadvantaged neighborhoods.
Data for this analysis are drawn from the first six waves of the National Longitudinal Survey of Youth 1997 (NLSY97). The NLSY97 is a nationally representative sample of American youth who were between the ages of 12 and 16 as of 31 December 1996. The first round of surveys was administered during 1997. Youth respondents have been interviewed on an annual basis since 1997 (with data collection ongoing). As described below, we append tract-level data from the 2000 decennial censuses to the NLSY97 individual records in order to capture the level of socioeconomic disadvantage in the immediate and extralocal neighborhoods in which respondents resided at each of the first six waves of the survey. We limit our analyses to respondents 18 years and younger at each of the survey waves, as the mechanisms driving mobility experiences likely change once individuals graduate high school and transition to higher education and the labor market. Respondents missing on any of the covariates were removed from the analysis via listwise deletion. As discussed below, we also limit our models to respondents who have moved at least once during the observation period (yielding a final sample of 1785 respondents).
is dichotomous variable differentiating respondents who moved census tracts between each wave of the survey from those who did not. We focus on census tracts for two reasons. First, the census tract is the geographic aggregation that most closely approximates a neighborhood and is used most frequently in studies of neighborhood effects (Sampson 2002
). Second, recent studies examining neighborhood effects on youth outcomes in the NLSY97 have consistently relied on tract-level measures of neighborhood characteristics (e.g., Vogel and South 2016
; Vogel et al. 2017
Our index of neighborhood disadvantage
combines the percent of families below the poverty line, the percent of households receiving public assistance, the percent of households headed by women, the percent of the population that is unemployed, and the percent of the population over the age of 25 lacking a high school diploma. These variables are highly intercorrelated and all load on a single factor (alpha = 0.924). We combine the variables into a single index such that higher scores indicate higher levels of neighborhood disadvantage
. We use census data from the year 2000 to generate our measures of disadvantage. Our rationale here is twofold. First, the 2000 census was conducted at the midpoint of our observation period, serving as the least biased, consistent measure of tract disadvantage across waves. Second, using a single point-in-time measure of neighborhood disadvantage, rather than an interpolated value from multiple censuses, allows us to isolate the effect of sending and receiving neighborhood characteristics on subsequent mobility (the main purpose of the analyses) rather than inadvertently measuring the effect of small-scale neighborhood change on mobility. As neighborhood status is rather stable over time (Zwiers et al. 2016
), significant changes in neighborhood disadvantage during our observation period are unlikely.
Extralocal neighborhood disadvantage is measured as a spatially lagged, distance-weighted index capturing the level of socioeconomic disadvantage in all census tracts within 100 miles of each respondent’s tract of residence. We employ an inverse distance decay function that assigns more weight to levels of disadvantage in geographically proximate neighborhoods and less weight to those that are further away. We then apply the spatial weights generated from this matrix to the index of neighborhood socioeconomic disadvantage described above to generate a measure of extralocal neighborhood disadvantage.
The regression models control for a number of variables that might be associated with both neighborhood disadvantage and residential mobility. Age reflects the age of respondents at the Wave 1 interview. Race differentiates respondents who self-identified as non-Hispanic black, non-Hispanic white, Hispanic, and non-Hispanic other race. Gender is coded 1 for males and 0 for females. The empirical models also control for whether the respondent resided in an urban area (relative to a suburban or rural area).
The models also include several variables intended to control for family background. Measures of family socioeconomic status (SES) include: home ownership, receipt of public assistance, and parent’s educational status. Home ownership differentiates homeowners from respondents whose parents rent or lease. Receipt of public assistance is measured as a dichotomous variable indicating whether any member of the respondent’s family received food stamps, supplemental security income, or aid to families with dependent children in the 12 months prior to the Wave 1 interview. Family structure differentiates respondents who lived with both biological parents from other living arrangements (single-parent households, step-families, and other family arrangements). The models also control for a lagged measure of mobility, differentiating respondents who moved at the previous wave from those who had not.
4.3. Analytic Strategy
We utilize hybrid random effects logistic regression models to analyze the associations between local and extralocal disadvantage on residential mobility. The models take on the form:
where α is the regression constant, Zi
is the time invariant covariates, LDit
refers to the level of local neighborhood disadvantage for respondent i at time, t, EDit
refers to the level of extralocal disadvantage, and Mit−1
refers to the one-year lag of prior residential mobility. The error term, uit
, can be decomposed into two components:
The first component, τi
, captures individual, time-stable characteristics. The random error term, εit
, is assumed to be uncorrelated with any of the predictor variables in the equation. This assumption is often untenable as the threat of endogeneity of unmeasured, time-stable covariates looms large. The hybrid model decomposes the time-varying effects into two parts: one reflecting between-person differences and one reflecting within-person differences. The between-person component is calculated as the average of each covariate for individuals over time (e.g., the average level of disadvantage in a respondent’s residential census tract for the first six years of the NLSY97). The within-person component is computed as the difference between respondents’ scores each wave from their group mean. This is denoted formally as:
These components are included in the hybrid regression such that:
The hybrid model has several advantages over traditional modeling strategies for panel data. For one, it provides estimates of both between and within-person effects. Second, the ‘fixed’ portion of the analysis holds constant any time-invariant variables, thereby reducing the risk of selection bias. Third, the decomposition reduces the potential for the within-individual estimates to be correlated with the time-stable error term, τi. Thus, the random coefficients in the model reflect the between-person differences in local and extralocal disadvantage on the likelihood of outward mobility, while the within-person estimates mirror those from a fixed-effects regression, allowing us to assess whether changes in local and extralocal neighborhood disadvantage associated with intertract mobility influence the subsequent probability of out-mobility within individuals over time.
The NLSY97 includes a number of respondents from the same family. Approximately 25 percent of respondents in the analytic sample are from families in which at least one other sibling also participated in the survey. As such, all models are presented with robust standard errors to account for the disproportionate clustering of respondents within families.
Residential mobility is considered to be a tumultuous event in the lives of children and adolescents. Emerging research suggests that (1) neighborhood disadvantage increases the likelihood of experiencing a residential move above and beyond individual and family risk factors and (2) decreases in neighborhood quality between sending and receiving neighborhoods may be especially harmful for later life outcomes. Conspicuously absent from this literature is a discussion of the processes that funnel families into particular neighborhoods and an understanding of how levels of socioeconomic disadvantage in the broader community contribute to out-mobility. Accordingly, this study attended to this gap in the literature by applying a longitudinal spatial modeling strategy to address both selection bias and spatial dynamics in the relationship between neighborhood disadvantage and residential mobility among a representative sample of American adolescents.
The results from the random effects models indicate that local neighborhood deprivation is positively associated with residential mobility, further highlighting the elevated levels of mobility among adolescents from economically disadvantaged neighborhoods often observed in prior research. Moving beyond the prior literature, the inclusion of the spatially lagged measure of extralocal disadvantage indicates that levels of socioeconomic disadvantage in nearby neighborhoods magnify the effect of local disadvantage on outmigration. In this sense, the strongest effects of neighborhood deprivation are found among adolescents who reside in areas of concentrated poverty. This suggests that families who live in economically deprived communities are more likely to move when the surrounding community resembles their own neighborhood, potentially pointing to affordable houses nearby, and are less likely to move when the surrounding area is more affluent than their own, perhaps underscoring the limited availability of alternative housing options. Of note, the inclusion of the product term for local X extralocal disadvantage strengthened the parameter estimates for family structure and black and Hispanic race-ethnicity. This may be attributable to the increasing representation of minorities and single and step-families in areas of greater concentrated disadvantage. Collectively, this spatial modeling strategy allows us to capture community influences that are often overlooked, either implicitly or explicitly, in prior research.
The ‘fixed’ component of the models allows us to assess how changes in neighborhood disadvantage between sending and receiving neighborhoods influence subsequent mobility. This approach addresses selection bias by holding constant time-invariant factors that may differ between respondents living in neighborhoods with varying degrees of disadvantage, thus providing a less-biased estimate of the effect of neighborhood disadvantage on mobility. Departing significantly from prior research as well as the estimates presented in the random effects models, our findings reveal that increases in socioeconomic disadvantage between sending and receiving neighborhoods are negatively associated with future mobility. We speculate that two complementary processes may be at play here. One the one hand, declines in neighborhood quality associated with residential moves may act as ‘traps,’ decreasing the likelihood that an individual will experience a subsequent move. However, the coefficient also seems to indicate that families who are able to ‘move up’ are unlikely to hold onto their improved neighborhood statuses, and thus face a greater likelihood of moving in the future. The parameter estimates for the spatially lagged indicator of extralocal disadvantage indicate that the negative effect of neighborhood deprivation is most pronounced among respondents who move to neighborhoods surrounded by other, poorer neighborhoods, suggesting that the mobility-hampering effect of neighborhood deprivation is strongest among respondents who move to areas characterized by the highest levels of concentrated poverty, both locally and extralocally.
This finding is especially striking when considered alongside the wealth of literature on the deleterious influences of neighborhood disadvantage on adolescent health and well-being (e.g., Leventhal and Brooks-Gunn 2000
). The results presented here suggest that, ceteris paribus, families who move to areas more disadvantaged than where they began are less likely to escape. Similarly, families who experience upward mobility are unlikely to maintain their improved residential status for long, potentially slipping back into economically deprived neighborhoods within a relatively short period of time. Adolescents who become ensnared in such spatial poverty traps may have prolonged exposure to myriad community risk factors, increasing their likelihood of engaging in high-risk behaviors (Sampson et al. 1997
) and diminishing their likelihood of graduating high school (Crowder and South 2011
), attending college (Garner and Raudenbush 1991
), and securing gainful employment in later life (Bauer et al. 2011
). From this vantage point, downward mobility and the spatial dynamics of neighborhood disadvantage may be especially salient factors in the perpetuation of socioeconomic inequality among American youth.
We would be remiss not to acknowledge several caveats that potentially undermine the findings reported here. First, as with most studies of neighborhood effects, we rely on administrative designations to capture neighborhood boundaries. The census tract is the most widely used approximation of residential neighborhoods in American scholarship (see Sampson 2002
) and is consistent with prior studies linking residential area characteristics to the individual survey responses of NLSY participants (Vogel and South 2016
). Of course, administrative boundaries do not perfectly map on to resident perceptions of their “neighborhood” (e.g., Basta et al. 2010
) and the level at which neighborhood characteristics are measured can have non-trivial consequences for the observed association between neighborhood features and the outcomes of interest (e.g., Hipp 2007
; Petrovic et al. 2018
; Vogel 2016
). Second, data collection for the NLSY 97 began approximately 20 years ago. A lot has changed in the United States in the intervening years. To our knowledge, these data still reflect one of the most comprehensive and representative longitudinal samples of American adolescents and one of the few national data surveys for which detailed address data are available. Third, while our modeling strategy removes much of the threat posed by omitted time-invariant covariates, the parameter estimates are sensitive to the omission of factors that might vary within respondents over time. In this sense, the models might miss some of the more proximate mechanisms that explain why some families move to affluent areas and others to disadvantaged areas. However, the combination of the fixed-effects regression and the lagged indicator of mobility provide a rather strong correction for endogeneity, further underscoring the robustness of the results presented here. Finally, as our analysis examines adolescent mobility patterns over a relatively short period of time (a maximum of five years per respondent), we can say little about how the spatial dimensions of neighborhood deprivation contribute to mobility over the life course.
We caution readers to interpret our findings with these caveats in mind. Rather than critical limitations, we view these as fruitful avenues for future research. Beyond the purview of the current study, it would be instructive to see whether the findings reported here are sensitive to variation in the aggregation of neighborhood boundaries; for instance, whether the spatial patterning of neighborhood effects holds when measures of socioeconomic disadvantage are aggregated to smaller (or larger) spatial scales. Similarly, given the dated nature of the NLSY97 and the relatively short follow-up, it is important to determine whether these findings can be replicated among more recent samples of American adolescents and whether the observed patterns hold over longer follow-up periods.
Limitations aside, this study expands on scholarship on neighborhood deprivation and residential mobility in several key regards. To our knowledge, this is the first attempt to jointly address the role of selection bias and spatial dynamics in the neighborhood deprivation–mobility relationship. The results from the spatial component of our analyses suggest that researchers interested in neighborhood effects on the behavior of residents need to consider not only the characteristics of immediate neighborhood environments, but also characteristics of geographically proximate neighborhoods. As demonstrated here, failure to consider such spatial processes may generate misleading conclusions about the nature of the relationship between neighborhood socioeconomic disadvantage and the mobility patterns of persons living in such areas. As most studies fail to control for residential selection, neighborhood disadvantage may be incorrectly associated with heightened mobility. Once holding constant unobserved time-invariant factors that sort certain families into specific neighborhoods, we find that neighborhood disadvantage actually decreases the odds of subsequent mobility, providing a stronger indication of a ‘true’ contextual effect. To this end, we strongly encourage future researchers to consider the role of selection bias and spatial dynamics when examining the relationships between neighborhood characteristics and the behavior of residents.