Household crowding, or having more household members than rooms in one’s residence, could potentially affect a child’s educational attainment directly through a number of mechanisms. We use U.S. longitudinal data from the Panel Study of Income Dynamics to derive new measures of childhood crowding and estimate negative associations between crowding during one’s high school years and, respectively, high school graduation by age 19 and maximum education at age 25. These negative relationships persist in multivariate models in which we control for the influence of a variety of factors, including socioeconomic status and housing-cost burden. Given the importance of educational attainment for a range of midlife and later-life outcomes, this study suggests that household crowding during one’s high school years is an engine of cumulative inequality over the life course.
Homeownership is a fundamental aspiration for most Americans, regardless of their socioeconomic status (SES) (Friedman and Rosenbaum 2004; Newport 2013). The Great Recession produced a fundamental change in the economic welfare of many families, especially as it relates to their housing options. Low-income families were particularly vulnerable to the collapse of the U.S. housing market and the concomitant Great Recession (Immergluck and Smith 2006; Lerman and Zhang 2012; Schuetz et al. 2008). From 2003 to 2009, the number of U.S. households that contained multiple families tripled (Eggers and Moumen 2013). Furthermore, recent work on foreclosures and the rise in the cost of housing is starting to shed light on the relationship between housing markets and the well-being of families (Lerman and Zhang 2014). This article adds insights into the importance of housing for the U.S. population by focusing on educational attainment and the enduring disadvantage that household crowding during childhood and adolescence engenders.
Living in crowded housing—usually defined as more than one person per room (Burr et al. 2010; Hall and Greenman 2013; McConnell 2015; Myers et al. 1996; Solari and Mare 2012)—is a common experience in the United States, especially among the poor and near-poor as well as children and adolescents. Although household crowding has declined since the 1970s (Holupka and Newman 2011), analysis of the American Community Survey (ACS) indicates that 15 % to 16 % of the U.S. population still lived in crowded housing between 2008 and 2010 (London and Frazier 2013). Among those living in or near poverty, the rate of crowding was around 30 %, with the highest rate observed among those in deep poverty1 (ranging from 29.17 % in 2008 to 32.47 % in 2010). ACS data also show that the age profile for crowding peaks at more than 30 % among those aged 0 to 5, declines rapidly and linearly from ages 0 to 20, plateaus at approximately 20 % for those aged 20 to 40, and declines nonlinearly among those aged 40 to 60 before leveling off at approximately 5 % among those aged 60 and older (London and Frazier 2013). Not only is crowding concentrated among the young, but evidence spanning the 2008–2010 period has suggested that crowding can change fairly rapidly in response to new social, economic, and policy conditions such as those that emerged during the Great Recession (London and Frazier 2013). During this same period, housing values dropped, foreclosures increased substantially, and households became more crowded (Eggers and Moumen 2013; Ellen and Dastrup 2012). Overall, the change in the rate of crowding for the U.S. population as a whole from 2008 to 2010 was 1.14 percentage points. Among those living below 50 % of the poverty threshold, the change was 3.30 percentage points; the change was 2.44 and 1.39 percentage points, respectively, among those living below the poverty threshold and below 200 % of the poverty threshold.2
Living in crowded housing is often a response to unaffordable housing costs (Bramley 2012), particularly among lower-income households (Lipman 2003) but also among blacks, Hispanics, Asians, and immigrants (Clark et al. 2000; Hall and Greenman 2013; Holupka and Newman 2011; Lipman 2003; Withers 2011). According to data from the ACS, the share of children in the 50 largest cities who lived in cost-burdened households—where more than 30 % of the monthly income was spent on rent, mortgage payments, taxes, insurance, and/or related expenses—increased from 39 % in 2006 to 41 % in 2010, and then declined to 36 % in 2013 (Kids Count Data Center n.d.). Although more than one-third of all children in the 50 largest cities currently live in cost-burdened households, not all cost-burdened households are crowded, and not all crowded households are cost-burdened (Ahrentzen 2003). Tolerance for, and perceptions and experiences of, crowding may vary culturally (Myers et al. 1996; Withers 2011). High housing cost burden for families may be mitigated by moving, including moves to doubling up with family or friends in order to pool resources (Ahrentzen 2003; Bramley 2012; Clark et al. 2000). Although some evidence has indicated that residential moves are not particularly effective at mitigating crowding among poor children (Holupka and Newman 2011), other evidence has indicated that receipt of unit-based housing subsidies reduces residential crowding in low-income single-mother households (Berger et al. 2008; Wells and Harris 2007).
Family-level economic constraints and housing market conditions that contribute to high housing cost burden are primary influences on crowding during childhood (Aratani et al. 2011). Whether and why crowding affects healthy child development and the life course is less clearly articulated in the literature, although over several decades, an important body of research that addresses the effects of crowding on child well-being and early-adulthood outcomes has emerged (Conley 2001; Evans et al. 1998; Gove and Hughes 1983; Leventhal and Newman 2010; Solari and Mare 2012). To help advance this literature, we situate our research within the life course perspective. According to the cumulative exposure model of the life course, earlier-life social conditions directly affect later-life outcomes; however, those influences may be offset or accentuated by the choices individuals make in response to social conditions, participation in social institutions, exposure to formal public policies, and a broad range of other potentially mediating and moderating factors (Berkman et al. 2011; Hendricks 2012). From this perspective, what happens early in life—such as exposure to sustained crowding in childhood or adolescence—may be central to subsequent processes of accumulating inequality; those deprived of resources in one domain become deprived of resources in other domains, while those with access to resources are consistently, and perhaps increasingly, able to obtain additional resources. This differential access to resources—especially when it is rooted early in the life course and affects educational outcomes—may generate and accentuate inequalities across groups over time (Dannefer 1987, 1988, 2003; Ferraro et al. 2009; O’Rand 1996, 2002).
Evidence has suggested that early-life crowding is linked to a range of health, developmental, social, and economic outcomes, with the potential to be an engine of cumulating inequality over the life course (Conley 2001; Evans et al. 1998; Leventhal and Newman 2010; London and Frazier 2013; Office of the Deputy Prime Minister 2004; Regoeczi 2002, 2003, 2008; Solari and Mare 2012; Wells and Harris 2007). Crowding can affect children and their future well-being through a number of mechanisms. Proximity to airborne infectious diseases is one straightforward reason that the health of children and adults may be affected by crowding. Residing in a crowded home also creates an environment with constant stressors, overstimulation, and a lack of privacy. In crowded environments, adults are more likely to be stressed, sleep-deprived, and depressed, which affects their parenting. Moreover, children’s needs frequently differ from the needs of others, potentially increasing conflict. The lack of privacy can create a context where studying and sleep, for example, are frequently interrupted, which could affect children’s behavior and performance in school.
Empirical evidence regarding a relationship between crowding and educational outcomes is mixed. One recent review of evidence from Organisation for Economic Co-operation and Development (OECD) countries on the impact of overcrowding on health and education concluded that “very limited evidence points to an independent relationship between overcrowding and educational attainment” (Office of the Deputy Prime Minister 2004:27). Additionally, Solari and Mare (2012) reported a range of negative associations between crowding and educational and education-related behavioral outcomes in cross-sectional and longitudinal data sets; however, they found that observed negative associations between crowding and math and reading scores were reduced to nonsignificance in fixed-effects models estimated on the longitudinal data.
In contrast to these studies, a small body of research has provided evidence that living in crowded circumstances has negative consequences for educational performance as well as child behaviors, parent–child interactions, and both parental and child health statuses that may affect short- and long-term educational outcomes (Gove et al. 1979; London and Frazier 2013; Office of the Deputy Prime Minister 2004). For example, Goux and Maurin (2005) found that French children living in overcrowded homes (i.e., two or more children per bedroom) were more likely to be held back in school and drop out; at age 15, more than 60 % of children living in overcrowded homes had been held back in primary or middle school. Another study reported that living in crowded circumstances was associated with increased externalizing behavior problems at school (Evans et al. 2001), whereas Solari and Mare (2012) found a positive association between crowding and externalizing behavior problems that persisted in the fixed-effects model they estimated. In addition to reporting negative effects of household crowding on parental physical and mental health, and marital relations, Gove et al. (1979) reported that parents in crowded households were (1) more likely to report that their children were a hassle and that they felt relieved when their children were out of the household, (2) less likely to know the children’s friends or the parents of their children’s friends, and (3) more likely to report that their children had difficulty finding a place in the home to get away from others and to study. Another study found that parents in overcrowded homes speak in less complex and sophisticated ways to children relative to parents in noncrowded homes (Evans et al. 1999). Conley (2001) averaged people per room over a five-year period of childhood and derived a dichotomous indicator of crowding (i.e., average >1 person per room vs. not), which he used to predict years of schooling at age 25, net of demographic characteristics, income, and other housing conditions. He found that those who experienced crowding in childhood had significantly fewer years of schooling than those who did not. His is one of the few studies that has examined how crowding during childhood affects outcomes at older ages.
We know less than we should about the potential life course consequences of growing up in a crowded environment because extant research is often based on cross-sectional, nonrepresentative samples and generally uses a simple measure of household crowding (i.e., more than one person per room) to examine contemporaneous associations with either children’s or adults’ outcomes. Often, the measurement of crowding is dichotomized to indicate the presence or absence of crowding at a point in time. In order to move the field forward and enhance our understanding of the relationship between childhood crowding and well-being across the life course, researchers need to develop a more comprehensive set of indicators of crowding that encompasses the entire childhood period. A more complete assessment of exposure to crowding during childhood has the potential to yield a better understanding of the impact of crowding on children during different critical developmental periods, as well as whether and how crowding during childhood affects a range of other outcomes across the life course. Much of the extant literature has focused on the relationship between crowding during childhood and children’s outcomes when they are young but has not adequately controlled for observed and unobserved potentially confounding variables.
This article adds to the research literature by investigating the relationship between the degree of exposure to crowding at different points in children’s formative years (i.e., ages 0 to 18) and two educational outcomes that mark the transition to adulthood and set individuals on particular life course trajectories: high school graduation by age 19 and educational attainment at age 25. We estimate models without and with fixed effects to ascertain the extent to which crowding exerts an independent influence on these educational outcomes, and we attempt to remove the influence of low SES and housing cost burden from these crowding estimates. We hypothesize that crowding in childhood is associated with decreased high school graduation rates and years of schooling at age 25, and expect that greater exposure to crowding during childhood and/or exposure during particular periods of childhood will be associated with stronger effects on educational attainment.
Data and Methods
The Panel Study of Income Dynamics (PSID) is the data source for this study. The PSID originally surveyed about 4,800 families and includes a large series of demographic questions in addition to information on adult labor force participation. The head of the household for each family was questioned annually from 1968 through 1997, after which the interviews were conducted biennially through 2011.3 The original set of families was composed of a random sample drawn from a national frame and a supplemental group that was low income. Over the years, additional groups were added, but we retain the two original PSID samples for all analyses because of the long time horizon necessary to estimate our models.
As children in the original PSID families grew older, left their homes, and started families of their own, they were retained in the PSID, as were their own children. This unique feature of the PSID allows us to obtain information on children during their childhoods and as they transition to adulthood. We select all children born into a PSID family from 1968 to 1992 in order to construct an annual history of crowding during the child’s first 18 years of life. We also restrict the analytic subsample to those with an individual probability weight to account for the presence of the low-income subsample.
Measuring Crowding During Childhood
Each year that the PSID collected data, the head of the household was asked, “How many rooms do you have for your family (not counting bathrooms)?” Each year the PSID also asked the head of household to report the “number of people (children plus adults) in the family unit.” Using these data, we divide the number of people in the family unit by the number of rooms to construct a continuous measure of persons per room. From this household crowding ratio (HCR), we create three measures to investigate the multifaceted potential influences of crowding on educational outcomes.
First, we use the mean annual HCR calculated from the year of the child’s birth to the year of the child’s 18th birthday. Estimates from models using this measure provide information on the relationship between marginal changes in the HCR and the education outcomes. Second, we generate a variable that measures the proportion of years between the year of the child’s birth and the year the child turned 18 (a total of 19 years) that the individual lived in crowded housing (defined as having a mean HCR > 1). This cumulative crowding measure indicates the proportion of one’s formative years spent in crowded circumstances. This measure lends itself to the cumulative exposure model of the life course by emphasizing the duration of exposure: it allows for a dose-response type of analysis. Next, we examine whether there are developmental periods when crowding has a particularly negative effect on educational attainment. To calculate crowding at different ages, we break each respondent’s childhood and adolescence into four periods. Because this is an inquiry into the educational attainment of individuals, we follow the transitions that many children make during their educational careers through secondary school. More specifically, we divide the child’s formative years into four roughly equal lengths: (1) the year of the child’s birth to the year of the child’s 4th birthday (early youth), (2) the year of the child’s 5th birthday to the year of the child’s 9th birthday (youth), (3) the year of the child’s 10th birthday to the year of the child’s 14th birthday (adolescence), and (4) the year of the child’s 15th birthday to the year of the child’s 18th birthday (high school years). The first period is prior to school, the second roughly covers elementary school, the third corresponds to the child’s middle school years, and the fourth represents the high school years. Over each developmental interval, we calculate the mean HCR.
As explained earlier, previous work in this area used the ever-crowded variable measured at a particular age or over a short period and estimated the mean difference in some outcome for those who lived in a residence with a HCR > 1 compared with those who did not. In this article, we estimate household crowding ratios over a much longer interval and incorporate new measures of household crowding—namely, mean crowding and cumulative crowding (both measured over the child’s first 18 years), and mean crowding at different developmental periods—to provide a much more nuanced investigation of the relationship between household crowding and educational attainment.
Table 1 provides descriptive statistics for the three crowding measures. The mean proportion of time spent in a crowded household was .086. This translates into 1.63 years in a crowded household of the child’s 19 formative years. Among those who ever experienced crowding, the proportion of time spent in a crowded household was .260, which translates into 4.94 of 19 years (not shown in table).
The mean level of the HCR over the formative years that individuals were in the PSID was .740, which is under the typical threshold for crowding. We do observe some changes in the mean HCR over the course of childhood. Before age 10, the HCR is approximately .77 people per room. As children age, the HCR decreases, falling to .67 people per room among those aged 15 to 18. Among those who ever experienced crowding, household crowding levels are much higher. The mean HCR is 1.02 at ages 0–4, 1.00 at ages 5–9, 0.90 at 10–14, and 0.82 at 15–18 (not shown in the table).
The aforementioned crowding measures potentially suffer from measurement error because the only people counted in the numerator are family members. Other residents of the household who are outside the head’s immediate family (e.g., a lodger or a whole other family) are not included. Therefore, our estimates of crowding are conservative: the presence of nonfamily members would increase the household-crowding ratio. From 1984 forward, the PSID did ask whether “nonfamily unit members” shared the housing unit with the household head’s family. Thus, from 1984 forward, we are able to compare results using the crowding measure with family members only in the numerator with the crowding measure calculated with all household residents in the numerator. For all measures, we use the number of rooms available for the family as the denominator because that is what is explicitly asked in the PSID.4 When we compare the HCR for both measures, we find that the correlations between the crowding measures are very high, ranging between .79 and .96. As expected, the HCR is always larger when we include nonfamily members. We ran a number of preliminary models comparing the results using the total number of family members in the numerator with the results obtained when we used all household members: the results were quite similar in all instances. Because the variable measuring the total number of family members is available starting in 1968, whereas the measure of all household residents is available starting only in 1984, we run all our models using the family unit measure. This choice is consistent with how the number of rooms that we use in the denominator is measured in the PSID (i.e., for family members) and increases our sample size considerably.
Educational Attainment Outcomes
We measure educational attainment two ways. The first measure indicates whether the individual graduated from high school by age 19. For individuals born early in the time series, we use the value reported 19 years after the individual’s year of birth. Because the PSID was conducted biennially after 1997, we do not have an education measure at age 19 for those born in 1979, 1981, 1983, 1985, 1987, 1989, and 1991 (our time series stops in 2011). For individuals born in these years, we calculate their educational attainment as the maximum value of education reported between their 18th and 20th years. For example, for individuals born in 1979, we calculated the maximum education between 1997 and 1999. This data structure creates a degree of “lumpiness” because we obtain that maximum differently depending on the individual’s birth cohort (i.e., some of the education measures are for 18-year-olds, some for 19-year-olds, and some for 20-year-olds). These differences occur every other year: we have 19-year-olds in one year, and we have the maximum education reported between the 18th and 20th birthdays in the other. In all our models, we control for year of the individual’s birth to account for this lumpiness in the derivation of the education measure. We classify all individuals who reported 12 or more years of education by their 19th birthday (as qualified earlier) as a high school graduate.
We also use a measure of the individual’s maximum level of education achieved by age 25. Again, because of the biennial surveying done in the PSID, we obtain that maximum differently depending on the individual’s birth cohort. In all instances, maximum education was obtained using the highest recorded education level reported between the year the individual turned 23 and the year the individual turned 27. As shown in Table 1, 72 % of the children born into the PSID had completed high school by age 19, and the mean educational attainment by age 25 was 13.77 years.
For some descriptive analyses, we examine childhood crowding in relation to the family’s income-to-needs ratio (ITNR). To calculate the ITNR, we divided the total family income for the child’s family in a given year by the Census Needs Standard for a family of the child’s size.5 We then took the mean ITNR measured from the year of the child’s birth to the year of the child’s 18th birthday. Among many definitions of poverty, an ITNR of less than 1 is the U.S. Census Bureau’s official definition of a poor family.6 Because we are using means, a family with a mean ITNR of less than 1 would be classified as poor, on average, across all years.
In the multivariate analyses, we include a dichotomous control variable for female and three race/ethnicity variables. The race variables are indicators for white, African American, and “other race” (i.e., neither white nor African American); white is the omitted category in our models. We also include an indicator for Hispanic ethnicity (regardless of race).7 We control for parental educational attainment, using the reported number of years of education for the head of the child’s household in the year the child was born. We also construct an annual per capita total family income measure for each of the developmental stages (i.e., early youth, youth, adolescence, and high school years). Because the HCR can change through adjustments in family structure, we also control for the proportion of each developmental period when the head of the child’s household was married. In addition, we include the mean ratio of the cost of housing—measured as either the mortgage or rent paid annually—to the income of the family.8 We estimate the mean housing cost burden for each developmental period. We then construct an indicator equal to 1 for families that had a housing cost burden greater than .3. The research literature on housing affordability commonly uses this threshold to indicate families with a housing cost burden (Newman and Holupka 2014a, b). We also include an indicator variable equal to 1 for any child who reported that the head was someone other than a parent (such as a grandparent or cousin) during his or her formative years. This measure would likely capture variation from children living in multigenerational families, which are disproportionately immigrant, Hispanic, and Asian (Lofquist 2012). Another characteristic shown to be related to crowding is living in rental housing and public housing (Rosenbaum and Friedman 2004). Each year, the PSID asked the head of the household whether the head owned, rented, or neither owned nor rented the current residence. We include variables for the proportion of each developmental period that the child lived in a residence that the child’s parents owned and another for the proportion of each developmental period that they reported neither owning nor renting. Finally, we control for the proportion of time that the child lived in the Northeast, Midwest, South, or a foreign country for each developmental period; the omitted category is residence in the West. We include these variables to control for regional variation in crowding. Table 1 reports mean values for each of these variables.
For both outcomes—high school graduation by age 19 and educational attainment at age 25—we estimate a series of ordinary least squares (OLS) models. Because the high school graduation outcome is binary, we technically estimate a linear probability model (LPM) for that outcome. We prefer the LPM to a probit or logit model because, as explained later, we include fixed effects in many of the models, and OLS easily accommodates this adjustment. The coefficient estimates for the LPM have the added benefit that they are easy to interpret.9
We begin by estimating the relationship between the mean HCR and each outcome. Next, we estimate the relationship between the proportion of one’s formative years that one lived in a crowded household and each outcome, which may provide evidence for a cumulative effect of crowding. Finally, because we are also interested in determining whether crowding has more influence in certain periods during one’s maturation, we break the crowding measure into four developmental stage–specific age groupings (as described earlier). Our objective with this last set of models is to determine whether crowding is particularly harmful to long-term educational outcomes during certain critical developmental periods.
For all models, we add indicators for female, African American, other race, Hispanic ethnicity, and parental marital status and education. We also include a series of indicators for the year the individual was born. These indicators capture time-specific shocks to the outcome and control for potential idiosyncrasies in the outcomes created by the variable definitions that we generated to accommodate the biennial survey schedule after 1997 (as described earlier). In addition, these models include four continuous measures of the mean per capita family income measured for the four developmental periods: early youth, youth, adolescence, and high school. We also control for the housing cost burden that the family experienced, on average, during the four developmental periods and an indicator of whether the child ever resided with a head of household other than a parent. To capture some contextual variation, we use the regional variables that report the proportion of time that the child lived in the Northeast, Midwest, South, or a foreign country (vs. the West) during his/her formative years. We measure each regional variable during each developmental period. Finally, we include a measure of the proportion of time that the children lived in their own home or neither rented nor owned their home, respectively, during each of his/her four developmental periods.
These initial regression models are descriptive: the crowding measures potentially suffer from bias resulting from omitted variables correlated with crowding and the educational attainment outcomes. For example, food insecurity—which is potentially correlated with crowding and educational attainment—may be the true causal explanation for any observed coefficient estimate. Failure to control for food insecurity may lead to the erroneous conclusion that crowding is harmful to children in terms of their educational attainment when in fact food insecurity is the true explanation.
To address the potential omitted variable bias in our initial models, we estimate two additional fixed-effects models. In the first, we match the children by their SES, operationalized as their mean per capita family income, measured between the child’s birth and age 4. More specifically, we partition the families into quartiles based on their mean per capita family income between birth and age 4. In these fixed-effects models (Model 2), our estimates of crowding are identified by the variation in crowding and the educational attainment outcomes among children in the same quartile of the income distribution when they were young. Preliminary analyses (not shown) indicate that this grouping is more homogeneous than using the variation in the entire sample. In these models, all factors that are common to individuals within these quartiles are removed from the estimation procedure and cannot be a source of bias.10
Although this strategy produces more homogeneous comparisons within the sample, there are potentially omitted variables that might continue to produce bias. For example, there is no reason to believe that food insecurity levels are identical among individuals in the lowest quartile of the per capita family income distribution. Therefore, we refine the categories from the previous model to make them even more homogeneous. In the last set of models, we further cluster children by their family’s crowding level from birth to age 4, which generates eight categories of children. For each quartile of the early-youth per capita income distribution, we create two subgroups composed of those who ever experienced an HCR greater than 1 during early youth and those who always had an HCR less than or equal to 1 during early youth. We estimate a fixed-effects model (Model 3) in which individuals with the same SES and early-youth crowding experience are grouped. Any factors that led to crowding (or to no crowding) in early youth would be constant within these groupings. Because the factors that lead to crowding between birth and age 4 are constant within these groups, the coefficient estimates should suffer from less bias than those estimated with Model 2. The only remaining potentially biasing factors are those omitted variables that affect education and crowding after age 4. For Model 3, because we match on crowding during early youth, we simplify our crowding measure to include only crowding from ages 5 to 18. For all models, we report Huber-White standard errors corrected for intrafamily dependence. All analyses are weighted.
Crowding by Socioeconomic Status, Location, Race, and Family Size
Table 2, panels A–D, describes variation in the three crowding measures by (1) quartiles of the mean ITNR distribution, (2) the region in which the child lived most during his/her formative years, (3) race and Hispanic ethnicity, and (4) the mean family size of the child during his/her formative years. Several findings are noteworthy. First, regardless of the measure we use, crowding is negatively related to the mean ITNR. With few exceptions, we observe declines in the level of crowding as the mean ITNR increases. For this reason, we control per capita family income in all models to distinguish crowding from economic deprivation. Second, crowding appears to be most common in the South (and for those who spent a lot of time in foreign countries, although this is a small subset of the sample). Third, mean HCR levels are much higher among African Americans, the “other” race category, and Hispanics. These socioeconomic and race/ethnicity patterns are broadly consistent with those documented in the literature (e.g., Hall and Greenman 2013). Finally, and not surprisingly, those with the largest families tend to experience more crowding than those with smaller families.
Crowding and Educational Attainment
In Table 3, we provide results from several models reporting the relationship between the mean HCR and the education outcomes. In Model 1, we report a negative and statistically significant association between the mean HCR and the probability of dropping out of high school. A 1 standard deviation increase in the mean HCR (0.218) is associated with a 2.2 percentage point decline in the likelihood of graduating from high school. In Model 2, we add the SES fixed effect; and in Model 3, we add the SES and crowding fixed effect. When we use variation within these fixed effects cells to identify the model, the association between crowding and high school graduation is no longer statistically significant. In addition, the coefficient in Model 3 declines to about one-third its size in Model 1.
The results are much more robust for the maximum education measure. Model 1 indicates that a 1 standard deviation increase in crowding is associated with a 0.29-year decrease in one’s maximum education. Given that the mean education level is 13.77, this is about a 2 % reduction in one’s education. In Models 2 and 3, the coefficient estimates decrease by about one-quarter but remain statistically significant. Using the point estimate from Model 3, we find that a 1 standard deviation increase in the mean HCR is associated with a 0.2-year (1.5 %) reduction in education. Although our measure of crowding is somewhat different, our result is consistent with Conley (2001), who found that individuals who had a mean HCR > 1 (between 1968 and 1972) completed 0.237 years less education by age 25, which translates into a 1.8 % reduction in educational attainment.11
Given the findings for the mean HCR, particularly for maximum education, we next ask whether this effect is cumulative and whether this effect is most pronounced during particular developmental periods.
In Table 4, we present the results that use a cumulative measure of crowding during childhood: the proportion of years that individuals lived in crowded housing (HCR > 1) from the year of birth to the year of their 18th birthday. We find large and precisely measured estimates of the relationship between cumulative crowding and high school graduation in Model 1. The coefficient estimate suggests that a 1 standard deviation increase in the proportion of time spent in crowded housing (0.181) is associated with a 2.8 percentage point decrease in the probability of high school graduation by age 19. In Model 2, we incorporate fixed effects based on the respondent’s SES and observe a small decrease in the point estimate for crowding, although it remains statistically significant. However, the point estimate in Model 3 is less than one-half the size of the Model 1 coefficient and is no longer statistically significant, which suggests that after the children are matched by SES and crowding between birth and age 4, the proportion of time in a crowded home is no longer related to high school graduation.
We observe similar results for maximum education. Based on Model 1, we estimate that a 1 standard deviation increase in the proportion of time spent in crowded housing is associated with a 0.18-year decrease in maximum education by age 25. Given that the average level of education is 13.77 years, this represents a 1.3 percentage point decrease in educational attainment at age 25. As we include additional controls in Models 2 and 3, however, the point estimates decrease in magnitude and become statistically nonsignificant. Thus, we find little evidence that the overall “dose” of crowding matters, particularly after we account for SES and crowding at a young age.
Crowding at Different Developmental Stages
In Tables 5 and 6, we report coefficient estimates for the mean HCR for each developmental stage–specific age interval in order to examine the effect of crowding on high school graduation by age 19 (Table 5) and maximum education at age 25 (Table 6), respectively. Model 1 in Table 5 shows that some developmental periods are more important than others. In fact, crowding between ages 15 and 18 (i.e., during the high school years) appears to be very important in relation to the likelihood of high school completion by age 19. Model 1 suggests that a 1 standard deviation increase in crowding during the high school years is associated with a 3.1 percentage point decrease in the likelihood of high school graduation. Given that 72 % of the individuals in the analytic sample graduated from high school, this represents a 4.4 % decline in the probability of high school graduation. Models 2 and 3 provide similar results quantitatively and in terms of statistical precision. Model 3 suggests that a 1 standard deviation increase in crowding is associated with a 3 percentage point decrease in the likelihood of high school graduation, which translates into a 4.2 % decline.
Table 6 provides findings for maximum educational attainment at age 25. Models 1–3 show that crowding may be more harmful at older than younger ages; crowding during the high school years is negatively and statistically significantly related to educational attainment, whereas crowding at younger ages is not statistically related to educational attainment. Based on the estimate in Model 3, a 1 standard deviation increase in crowding at ages 15 to 18 is associated with a 0.19-year reduction in educational attainment at age 25, which is approximately a 1.4 % decline.12
This study estimates the relationship between household crowding during childhood and two educational outcomes, and differs from the extant literature in several ways. First, we collect information on crowding throughout the child’s formative years (i.e., from the year of the child’s birth to the year of the individual’s 18th birthday). Most previous research has focused on crowding in a single year or a small subset of years during childhood. Second, this study brings three new measures of crowding into the literature. We first estimate the relationship between mean HCR from birth to the year of the individuals’ 18th birthday and our education outcomes. Next, we ask whether cumulative exposure to crowding, measured as the proportion of formative years lived in a crowded household, is related to education. We then estimate the relationship between the mean level of household crowding in four distinct periods of childhood and adolescence. This set of results was designed to examine whether crowding is particularly meaningful with respect to educational outcomes during certain developmental periods.
Consistent with the literature, our data showed that no matter how we measured crowding, it is strongly associated with the SES of the child. As such, one might worry that crowding is simply a proxy for severe deprivation in most estimates. To address this potential confounding and provide evidence of an effect of crowding that is independent of SES during childhood, we control for per capita family income and housing cost burden in our models. Additionally, to address potential bias resulting from unobserved socioeconomic factors, we use fixed effects to match children based on their per capita family income as well as crowding in early youth.
After these adjustments, our findings suggest that mean household crowding has an independent association with educational attainment and that living in a crowded household differentiates children. Descriptively, those who live in a crowded household at any point before their 19th birthday are less likely to graduate from high school and have lower educational attainment at age 25. However, the high school graduation result is not robust after individuals are matched on SES and crowding before age 4. For the maximum education outcome, however, crowding remains highly predictive and negatively associated with one’s highest grade completed.
In our second set of findings, we do not observe a cumulative effect of crowding. Our models show no statistically significant relationship between the proportion of one’s formative years spent in a crowded household and education.
When we estimate the differences in high school completion based on crowding at different developmental stages, we find a robust relationship only for crowding during one’s high school years. A 1 standard deviation increase in mean HCR during one’s high school years is associated with a 3 percentage point decline in the probability of high school graduation, which is equivalent to a 4.2 % decrease. Crowding during other developmental periods does not appear to affect the likelihood of high school graduation. Similarly, crowding during one’s high school years is negatively related to completed education by age 25. We observe that a 1 standard deviation increase in crowding during one’s high school years is associated with a 0.19-year decline in their maximum education by age 25, which translates into a 1.4 % decline. Again, crowding during other developmental periods is not related statistically to one’s educational attainment.
Several potential explanations exist for the link between crowding during high school and educational attainment. Crowding at older ages is more disruptive to studying patterns when homework requirements are more demanding. In addition, older children in crowded homes may be forced to take on more adult responsibilities, such as childcare, which can limit their ability to study. Taking on such childcare roles during the high school years can disrupt educational achievement (Gennetian et al. 2004). As children age, they have more freedom to leave their homes to escape the crowded environment, which might reduce their study time or increase their involvement in activities outside school, which might explain our results. In addition, as teenagers increasingly leave their homes, they will experience less adult supervision, which might increase their likelihood of engaging in risky behaviors and thus could reduce their educational performance and attainment (Gennetian et al. 2008; Lopoo 2005; McLanahan and Sandefur 1994). Even if students who live in crowded homes have grades that are good enough to allow them to graduate, their grades may not be good enough to allow them to be admitted to, or obtain financial aid to attend, college, which may account for some of the difference in educational attainment at age 25 that we document. Future work in this area should explore the mechanisms that explain why crowding during one’s high school years has such a large influence on educational attainment.
This study is not without limitations. One limitation is that we are unable to take specific contextual factors into account, such as degree of urbanicity, which could mitigate or exacerbate the effects of crowding. The PSID did not start collecting this information until very late in the panel, which precludes our ability to control for it in our models. Our models would also benefit from more controls for the neighborhood context, such as local housing costs. Again, this information is not available in the PSID. Although we include fixed effects, which may reduce the importance of these factors, future work might consider whether these factors explain the effects of crowding we observe. Also, the PSID sample has limited racial and ethnic diversity, particularly with respect to Hispanics and Asians. Future work should examine whether the effects of crowding vary across different racial, ethnic, and immigrant groups. Finally, descriptively, we know that 15- to 18-year-olds who live in crowded homes are different from 15- to 18-year-olds who do not. Even though we control for a variety of variables to reduce bias in the crowding estimates, new research might use data files with more detailed information on these high school–aged youth to determine whether factors that are unaccounted for here might explain the differences we observe.
Given that education is a primary engine of cumulating inequality across multiple domains of the life course, our results suggest that crowding may be an important contributor to inequality that is independent of the effects of severe deprivation and compounded disadvantage measured by a family’s per capita family income and their housing cost burden. Household spatial resources matter for children’s educational attainment over and above family economic resources. Furthermore, our results suggest that policymakers who are interested in reducing crowding through housing policies or other social programs should pay particular attention to crowding during children’s high school years. Public housing officials also might consider allocating more space to families with high school–aged children, given that this could potentially affect their human capital accumulation and future economic well-being.
This study investigates one set of outcomes related to educational attainment. Future research might also ask about the relationship between childhood crowding and other important life course outcomes, such as economic well-being, transitions to prison and the military, physical and mental health, and family formation and dissolution. Such research should consider the full range of exposure to childhood crowding because it is possible that the effects of exposure to crowding are different at specific developmental stages or cumulative exposure matters for other outcomes (i.e., crowding during the high school years matters for educational outcomes, but crowding at other ages or in variable amounts may matter for other outcomes). Given that crowding can also affect adults in households, which may affect parenting and the consequences of children’s actions for adults, the linked-lives principle of the life course perspective suggests that examination of potential intergenerational influences of crowding is also critically important. Much work remains to understand the consequences of crowding and its impact on the lives of children and adults in the United States.
We thank Emily Cardon, Maddy Hamlin, and Mary Stottele for their research assistance on this project.
Deep poverty is usually defined as income below half the federal poverty threshold (Shaefer and Edin 2013).
In 2009, the poverty threshold for a family of three with one adult and two children was $17,285 (U.S. Census Bureau n.d.).
Each interviewed family in the PSID was assigned a head of household. The head was at least 16 years old and had the primary financial responsibility within the family. In nearly all cases, the head was male. Although the children born into the PSID could have various relationships with the head (e.g., a grandchild, a niece/nephew), the vast majority were the head’s children. For instance, among those born in 1968, 81 % were children of the head, and 13 % were grandchildren, nieces, or nephews. For the remaining 6 %, the relationship status was not reported. For simplicity, we use the terms head of household and parent interchangeably.
The PSID question asks the respondent to report the number of rooms in the household for the “family.” Some respondents may adjust their count of rooms to take into account the presence of nonfamily members in the household. However, we have no means to ascertain the extent to which that is the case.
The PSID recorded the Census Needs Standard using a schedule reported by the U.S. Census Bureau based on the size of the family and number of children. This schedule is available online (http://www.census.gov/hhes/www/poverty/data/threshld/index.html).
More details on the official definition can be found online (http://www.census.gov/hhes/www/poverty/about/overview/measure.html).
The PSID started collecting information on the “Spanish or Hispanic descent” of the heads of household in 1985. We created a variable for Hispanic that overlaps with the racial categories (i.e., among Hispanic respondents, some are white, some are African American, and the rest fall into the “other race” category).
Most housing cost burden measures include, in addition to annual mortgage payments or rent, mortgage interest, property taxes, the cost of utilities, and housing (or rental) insurance (see, e.g., Newman and Holupka 2014a). In selected years, the PSID measured some of these variables, but only the mortgage or rental payment was consistently measured throughout the period of our study.
We ran several preliminary logistic regression models to determine whether model choice changed our findings, and found that the results were substantively identical.
Another potential modeling option that would reduce omitted variable bias is a family fixed-effects model. In such a model, one compares the crowding experienced between (among) siblings to reduce bias in the crowding coefficients. Factors that are common among siblings, such as parental supervision, are removed from the estimation process and cannot bias coefficient estimates. These models are predicated on variation in crowding between siblings. Unfortunately, ever-crowding does not vary much between siblings. For example, only 14 % of the cases had variation in crowding (measured as a HCR > 1) between ages 15 and 18. Even a continuous measure of crowding showed very little variation. Given this low level of variation, we chose not to use a family fixed-effects model.
In supplemental analyses, we estimated these models separating the respondents into four quartiles based on their INTR and found no evidence of heterogeneous effects by SES.
In supplemental analyses, we estimated models to determine whether crowding mattered at specific ages in the high school years. Our results suggest that the effects are constant from ages 15 to 18 and do not surface before age 15.