Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Dec 1.
Published in final edited form as: Psychol Sch. 2019 Dec 19;57(12):1830–1844. doi: 10.1002/pits.22331

Childhood homelessness, resilience, and adolescent mental health: A prospective, person-centered approach

Janette E Herbers 1, J J Cutuli 2, Joanna N Keane 3, Jake A Leonard 4
PMCID: PMC7792983  NIHMSID: NIHMS1063124  PMID: 33424042

Abstract

School districts and other service providers are increasingly aware of the substantial mental health needs of students experiencing family homelessness. Past findings are mixed regarding whether homelessness conveys unique risk beyond the risks associated with extreme poverty. With prospective longitudinal data on homelessness experiences across childhood, we utilized latent profile analysis as a person-centered approach to conceptualizing mental health outcomes in adolescence for 3,778 youth. We considered literal family homelessness as well as families living doubled-up, and we employed propensity score matching to identify a comparison group of non-homeless students balanced across a range of covariates to address systematic bias. Results indicated that students who experienced literal homelessness during childhood were significantly less likely to demonstrate profiles of resilience in mental health functioning. We considered our approach and findings in light of challenges and opportunities particularly relevant to the school context.

Keywords: family homelessness, mental health, person-centered analysis

Rates of homelessness among U.S. students have been increasing for the past 3 decades, with recent estimates indicating that 1.3 million students experienced homelessness during the 2016–2017 school year (National Center for Homeless Education, 2019; Sulkowski, 2016). In addition to concerns about achievement disparities, school districts and other service providers are increasingly aware of the substantial mental health needs of these students (Cutuli, Treglia, & Herbers, 2019; Sulkowski & Michael, 2014). Compared to other subgroups of students at risk, we know relatively little about the varying profiles of resilience or maladaptation of these highly mobile children and youth (Masten, Fiat, Labella, & Strack, 2015). In the current study, we utilized rich longitudinal data and advanced statistical methods to examine the implications of children’s experiences with different forms of family homelessness for adolescent mental health. We considered our approach and findings in light of challenges and opportunities particularly relevant to the school context.

We conceptualize family homelessness in line with a resilience framework derived from developmental-task theory (e.g., Masten & Tellegen, 2012; see also Cutuli & Herbers, 2014). In brief, factors at multiple levels of the individual and their context contribute to developmental changes over time, usually resulting in adaptive outcomes defined by criteria for competent functioning across multiple domains. These contributions are complex, involving the action and interaction of numerous factors in concert with the individual’s developmental history. Resilience in development is a special application of this general account: experiences of risk and adversity threaten developmental competence, though protective or promotive factors can prevent or offset their impact, resulting in competent functioning.

On average, students who experience family homelessness show more problems than non-homeless peers on assessments of academic achievement, school attendance, attainment, and behavioral and emotional functioning (Cowan, 2017; Cutuli & Herbers, 2014; 2018; Grant, Gracy, Goldsmith, Shapiro, & Redlener, 2013). Chronic risk factors associated with extreme poverty play a key role in these disparities, and children who experience homelessness are typically considered to fall at the high end of a poverty-related risk continuum (Buckner, 2008; Masten et al., 2015). Furthermore, acute adversities associated with episodes of homelessness and residential mobility may exacerbate these chronic stressors (Cutuli & Herbers, 2014). However, the extent to which homelessness itself conveys unique risk above and beyond extreme poverty and associated adversities remains unclear (Cutuli & Herbers, 2018; Bassuk, Richard, & Tzertsvadze, 2015; Buckner, 2008; Park et al., 2012).

In a recent meta-analysis, Bassuk et al. (2015) summarized results of 12 studies that reported on prevalence of mental health issues among homeless children compared to low-income-housed children. Results indicated a significant effect size for differences among school-age (7 to 11 years old) children, and no significant differences among preschool-aged children (3 to 5 years old). In addition to documenting the paucity of studies on the topic, the authors described a number of shortcomings associated with the extant literature. Findings were “limited by small samples and lack of comparability across studies primarily due to differences in demographics, definitions, and duration of homelessness; different measures; and lack of comparison groups” (Bassuk et al, 2015, p. 94). Furthermore, they noted that most findings were based on cross-sectional studies that cannot inform causality or longer-term associations between homelessness experiences and mental health. Bassuk et al. thus called for larger studies with more sophisticated research designs, including longitudinal and person-centered approaches to describing pathways of risk and resilience (2015).

Person-centered Approaches for Understanding Impacts of Homelessness

As is true more generally in the field of risk and resilience, most studies of homelessness have relied on variable-centered approaches (Cutuli & Herbers, 2019; Bassuk et al., 2015; Herbers, Cutuli, Jacobs, Tabachnick, & Kichline, 2019; Huntington, Buckner, & Bassuk, 2008; Laursen & Hoff, 2006). When researchers focus on quantitative differences in individual variables as outcomes of interest, underlying assumptions of this approach can obscure meaningful differences in the functioning of the whole persons involved. More specifically, variable-centered approaches assume that statistical relations among certain risk factors and certain outcomes are constant for all individuals within a sample. In contrast, person-centered approaches seek to identify meaningful subgroups with different configurations of such relations within broader populations (Masten, 2014; von Eye, Bergman, & Hsieh, 2015). By considering whole persons as the units of analysis, person-centered methods can reveal complex patterns of individual differences in adaptive functioning that can shed more nuanced light on the developmental processes at play (Bergman & Magnusson, 1997).

Recent interest in applying statistical methods such as latent class analysis has yielded a wealth of studies utilizing sophisticated person-centered approaches to studying risk (Jobe-Shields, Andrews, Parra, & Williams, 2015; Lanza & Cooper, 2016). Fewer researchers have capitalized on the opportunities that such methods provide for person-centered analyses of outcomes, particularly resilient versus maladaptive profiles of functioning (see for examples: Herbers et al., 2019; King, Lembke, & Reinke, 2016; Yates & Grey, 2012). When seeking to better understand whether family homelessness conveys unique risk, a person-centered approach to mental health may reveal differences in functioning at the level of the individual that would not be apparent by examining variables such as internalizing or externalizing problems in isolation (Bassuk et al., 2015). Understanding how experiences of family homelessness may contribute to different profiles of mental health functioning among students has great potential to inform school- and community-based intervention efforts (Huntington et al., 2008; Canfield, Nolan, Harley, Hardy, & Elliott, 2016).

Definitions and Comparison Groups

Besides its reliance on mostly variable-centered studies, our understanding of possible unique impacts of homelessness on mental health outcomes is limited also by differing definitions of family homelessness and identifying appropriate comparison groups in relation to available samples (Bassuk et al., 2015). Most studies are based on either large administrative datasets or small convenience samples of children residing in family shelters. Administrative data can illustrate the big picture of risk over time, but typically lacks precise measurement of individual differences in experiences, developmental processes, or relevant outcomes (Cutuli et al., 2013; Park et al., 2012). Furthermore, many studies of childhood homelessness utilize administrative data from education systems, which generally define homelessness as situations where the child or family lacks a fixed, regular, and adequate nighttime residence (e.g., Canfield et al, 2016). This “broad” definition of homelessness tends to include families who stay doubled-up with friends or relatives because of financial reasons as well as those in emergency shelters and other homeless situations. In contrast, other studies focus on “literal homeless” situations, defined as staying in a shelter, hotel, vehicle, or other place not intended for human habitation (HUD, 2019).

Convenience samples of children in shelters can better delineate unfolding developmental processes at the individual and family level, but they often lack statistical power for detecting smaller but potentially important effects, and they usually consider only cross-sectional or very short-term longitudinal associations (Bassuk et al., 2015). Furthermore, convenience samples recruited from shelters represent only those children whose families reside in emergency housing and are literally homeless, excluding children whose families double-up with relatives or friends or are otherwise included in the broad definition (Grant et al., 2013). Implications of findings from both administrative data and convenience samples can be limited if there are no non-homeless participants for comparison, or if the comparison groups differ systematically from the homelessness groups on other key factors such as average income level, demographic make-up, or other experiences of risk and adversity (Cutuli & Herbers, 2018; Bassuk et al., 2015; Marcal, 2018; McCoy & Raver, 2014).

When studies do include a non-homeless comparison group, very few use advanced quasi-experimental techniques to better isolate the impact of homelessness relative to other factors that tend to accompany homelessness, like deep poverty and concomitant risks. For example, studies consistently find lower academic achievement for homeless students compared to non-homeless students who qualify for free and/or reduced-price meals (130% or 185% of federal poverty guidelines, respectively; e.g., Cutuli et al., 2013; Herbers et al., 2012). However, families in emergency shelter, for example, tend to experience deep poverty (below 50% of federal poverty guidelines; see Gubits et al., 2016), making it unclear if any differences in functioning are due to homelessness or due to differences in income. Advanced quasi-experimental techniques, like propensity score matching, offer a means to construct more appropriate comparison groups for students experiencing homelessness (King, 2016; Leech, 2012; Marcal, 2018) and, thus, better isolate the impact of homelessness on developmental outcomes. Indeed, differences in school functioning for homeless students were less consistent when a propensity score approach was used in another study of education records (Cutuli & Herbers, 2018). We are not aware of any study that has utilized propensity score matching to test for impacts of family homelessness on mental health outcomes.

The Current Study

In the current study, we capitalized on a large sample with prospective, longitudinal data from the Fragile Families and Child Well-Being Study (FFCWS), and we employed sophisticated statistical methods to address many of the limitations described above. In the FFCWS, housing instability was assessed across five time-points from ages 1 to 15 years, including questions about staying in shelters, hotels, vehicles, or other forms of literal homelessness (Grant et al., 2013) as well as living doubled-up without paying rent, more consistent with the McKinney-Vento definition of homelessness employed by the U. S. Department of Education (42 U.S.C. § 11434a(2)). We utilized propensity score matching based on numerous covariates to identify a non-homelessness comparison group accounting for systematic bias. Next, we conducted a latent profile analysis to identify person-centered profiles of individual differences in adolescent mental health. Predicting mental health profiles based on different forms of homelessness, we expected to find that youth with any experiences of family homelessness would be less likely to demonstrate profiles of resilience in mental health functioning compared to carefully matched, non-homeless peers.

Method

Participants and Procedure

Participants were involved in the Fragile Families and Child Well-Being Study (FFCWS; see Reichman, Teitler, Garfinkel, & McLanahan, 2001), a longitudinal study following nearly 5,000 children from 20 large U.S. cities from birth until age 15 years. The original cohort was oversampled for low-income mothers, most of whom were unmarried at the initial interviews conducted in hospitals following the focal child’s birth in 1998–2000. Although fathers participated in interviews as well, we relied on reports from mothers or primary caregivers for consistency. FFCWS team conducted follow-up telephone interviews when children were ages 1, 3, 5, 9, and 15 years with good retention rates (89%, 86%, 85%, 74%, and 73% respectively). A randomly selected subsample completed in-home interviews at ages 3, 5, 9, and 15 years. The current study included data from assessments conducted at birth and ages 1, 3, 5, 9, and 15 years for the 3,778 participants who had data available for at least one indicator of mental health functioning at age 15 or complete data on predictor variables for propensity score matching. Data for the analyses compared to the original sample appear to be missing completely at random, supported in part by Little’s MCAR test, χ2(1) = .034, p = .853.

Homelessness.

At each of age 1, 3, 5, 9, and 15 assessments, we coded literal homelessness based on whether the primary caregiver described their residence as having lived in a homeless shelter, in a hotel, on the street, in an abandoned building, in an automobile, or any other place not meant for regular housing, either currently or within the past 12 months (9.0% of the sample). Doubled-up participants were those whose primary caregivers indicated that the family had lived with family or friends without paying rent, either currently or within the past 12 months, at any of the assessments, but had never been literally homeless (28.9% of the sample). Participants who met the criteria of literal homelessness, doubled-up, or both at any of the assessments were included in the broad homelessness group (36.5%).

Matching variables.

In order to obtain propensity scores and balance the sample to eliminate systematic differences, we utilized eleven covariates that have been established in the literature as predictors or correlates of family homelessness. Covariates based on demographics reported by the mothers at the baseline or age 1 assessments included child African American ethnicity, child Hispanic ethnicity, child gender, mother’s age when the child was born, mother not born in the U.S., mother’s level of education, and mother being less than 18 years of age when her first child was born (Cutuli & Herbers, 2014; Haskett & Armstrong, 2019; Herbers et al., 2019). From the baseline assessment, we also included the covariate of whether the child was considered low birth weight (weighing less than 2,500 grams at birth; Shaw, Herbers, & Cutuli, 2019).

We utilized four additional covariates constructed by combining information collected across the longitudinal time-points: family poverty ratio, maternal depression, harsh parenting, and exposure to intimate partner violence (Cutuli et al., 2017; Haskett & Armstrong, 2019; Park et al., 2015). Family poverty ratio was an average across the baseline and ages 1, 3, 5, 9, and 15 assessments of ratio of total household income to the official poverty threshold of the previous year established by the U.S. Census Bureau. Maternal depression was based on mother or primary caregiver response to the Composite International Diagnostic Interview Short Form (CIDI-SF; Kessler, Andrews, Mroczek, Ustun, & Wittchen, 1998) during the age 1, 3, 5, 9, and 15 surveys. We considered “depressed” those primary caregivers who met criteria of the instrument for likely depression diagnosis at any of the five time points. Domestic violence was based on primary caregivers’ responses to several questions at the birth, age 1, 3, 5, 9, and 15 surveys. Primary caregivers responded to multiple questions that asked whether and how often they were hit, kicked, hurt, pushed, grabbed, shoved, or involved in physical fights with the child’s father or their current partner. We coded domestic violence “present” if any of these items were endorsed at any of the surveys. Finally, harsh parenting was based on the Conflict Tactics Scales, Parent Child version (CTS-PC; Straus, Hamby, Finkelhor, Moore, & Runyan, 1998) administered during the in-home portions of the age 3 and 5 assessments and by phone interview at age 9 and age 15. Fifteen self-report items referred to psychological aggression, physical assault, and neglect of children. Following recommendations from prior work (Berger, McDaniel, & Paxson, 2005), we coded harsh parenting risk as “present” for those belonging to the sample’s top 10th percentile of any of the CTS-PC subscale scores at any of the surveys.

Measures

Profiles of mental health.

We derived the categorical latent variable for profiles of mental health at age 15 based on five continuous indicators: teen report of internalizing behaviors, teen report of externalizing behaviors, teen report of substance use, parent report of internalizing behaviors, and parent report of externalizing behaviors. We used a z-transformation to standardize the composite score of each indicator for ease of interpretation.

Internalizing (self-report).

Youth participants reported on their own internalizing behaviors in response to 6 items drawn from the Brief Symptoms Inventory 18 (BSI-18: Derogatis & Savitz, 2000) regarding anxiety and 5 items from the Center for Epidemiologic Studies Depression Scale (CES-D: Radloff, 1977) regarding symptoms of depression. These measures have been validated for use with culturally diverse samples of youth (Handal, Gist, Gilner, & Searight, 1993; Perreira, Deeb-Sossa, Harris, & Bollen, 2005). All 11 items were modified to inquire whether, over the past four weeks, the teens strongly agreed, somewhat agreed, somewhat disagreed, or strongly disagreed with statements such as “feeling tense or keyed up,” “feeling fearful,” “I feel sad,” and “I feel I cannot shake off the blues, even with the help from my family and friends.” Together, the items had strong internal reliability as a measure of internalizing symptoms, α = .85.

Delinquency (self-report).

Youth reported on frequency of delinquent behaviors in response to 13 items adapted from measures used in the National Longitudinal Study of Adolescent Health (Bernat, Oakes, Pettingell, & Resnick, 2012). For each item, youth responded 0 (never), 1 (1 or 2 times), 2 (3 or 4 times), or 3 (5 or more times). Sample items included, “deliberately damaged property that did not belong to you,” and “took part in a group fight.” Items had good internal reliability, α = .74.

Substance use (self-report).

Youth reported on their substance use in response to a series of questions about whether they had ever smoked a whole cigarette, had an alcoholic drink more than 2–3 times when they were not with their parents, had ever tried marijuana, had ever tried any other type of illicit drug, and if they had ever taken prescription drugs that were not prescribed for them. Because the majority of teens (71%) responded no to all of these items, we coded responses as 0 (none), 1 (endorsed using only one category of substances), and 2 (endorsed using two or more categories of substances).

Internalizing (parent report).

For parent report of youth internalizing behavior, we utilized primary caregiver responses to 8 items from the anxious/depressed behavior and withdrawn behavior subscales of the Child Behavior Checklist (CBCL: Achenbach & Rescorla, 2001). The CBCL has strong psychometric properties, including concurrent and discriminant validity for identifying internalizing and externalizing disorders among youth (Seligman, Ollendick, Langley, & Baldacci, 2004). Primary caregivers indicated 0 (never), 1 (sometimes), or 2 (often) to describe frequency of youth behaviors such as “cries a lot,” “is too fearful or anxious,” and “is unhappy, sad, or depressed.” The items had good internal reliability, α = .79.

Externalizing (parent report).

Parent report of youth externalizing behavior was also based on the CBCL, with 11 items from the aggressive behavior subscale and 9 items from the rule-breaking behavior subscale. Primary caregivers indicated 0 (never), 1 (sometimes), or 2 (often) to describe frequency of youth behaviors such as “vandalizes,” “gets in many fights,” and “argues a lot.” The items had good internal reliability, α = .89.

Analyses

Latent profile analysis.

We conducted latent profile analyses using Mplus version 7.4 (Muthén & Muthén, 2012). Missing data were accounted for using full information maximum likelihood (Collins & Lanza, 2010; Muthén & Muthén, 2012). Following recommendations for best practice, we estimated a set of models for the latent variable to determine the best solution for the number of profiles (Masyn, 2013; Nylund-Gibson & Masyn, 2016).

First, we estimated solutions for the latent variable of mental health with different numbers of categorical latent classes, ranging from one through five classes, to assess which solution was best fitting and most parsimonious. We considered the Bayes Information Criterion (BIC), the adjusted Lo-Mendell-Rubin likelihood ratio test (LMR-LRT; Lo, Mendell, & Rubin, 2001), the parametric bootstrapped likelihood ratio test (BLRT; McLachlan & Peel, 2000), and the interpretability of the solution in light of theory (Collins & Lanza, 2010; Masyn, 2013). We examined posterior probabilities of assignment and overall entropy for degree of differentiation among classes. We also compared models in which the variances of indicators were constrained to be equal across profiles to those with variances allowed to vary across profiles, consistent with recommendations (Masyn, 2013).

Propensity score matching.

We conducted a logistic regression to estimate predicted probabilities of ever being literally homeless or doubled-up versus never being homeless or doubled-up based on the following covariates: African American ethnicity, Hispanic ethnicity, gender, maternal age at birth, maternal age when her first child was born, maternal education level, low birth weight, family poverty ratio, maternal depression, harsh parenting, and exposure to intimate partner violence. We selected the nearest neighbor on predicted probability for each student who had a history of homelessness or being doubled-up (Guo & Fraser, 2015; Thoemmes & Kim, 2011).

Predictive models.

Finally, we ran two multinomial logistic regression models to predict mental health profile membership from homelessness categories. First, we compared groups based on the broad definition of homelessness, comparing any students who had been literally homeless or doubled-up to their matched peers. Next, we compared three groups: students who had experienced literal homelessness, students who had lived doubled-up but had never experienced literal homelessness, and matched students who never had lived doubled-up and had never experienced literal homelessness. The [blinded institution] IRB found this study exempt as analyses of a pre-existing, publicly available dataset.

Results

Latent Profiles of Mental Health

Based on the model comparison approach, we determined that allowing estimates of variance to differ across profiles produced considerably better fit compared to models with variances fixed across profiles (for example, ΔBIC = 5,304 for the 4-profile solutions). Overall, we considered the solution with 4 latent profiles to have the best balance of fit and parsimony (see Table 1). Although model fit based on BIC did improve with the addition of a fifth latent profile, the LMR-LRT p-value for the 5-profile model was not statistically significant. Furthermore, the two smallest profiles accounted for only about 1% and 2% of the sample, and those two profiles appeared to be a split of a single profile from the 4-profile solution, with other profiles not meaningfully changed. As such, we judged the 4-profile solution more interpretable and more parsimonious. The posterior probabilities of assignment ranged from .89 to .99, and overall entropy of .87 indicated good class separation.

Table 1.

Model selection statistics for latent profile analysis of adolescent mental health

# Profiles Log-likelihood BIC Entropy LMR-LRT LMR-LRT p-value BLRT
1 −24766 49615 - - - -
2 −20850 41863 0.846 7738 <.0001 <.0001
3 −19640 39525 0.849 2391 0.021 <.0001
4 −19109 38547 0.865 1047 0.031 <.0001
5 −16931 34271 0.977 4304 0.684 <.0001
Open in a new tab

Note. Accepted model is presented in boldface.

We depicted means and standard errors of the five mental health indicators for each latent profile in Figure 1. The first profile included 40.3% of the sample, with low average scores across all five indicators of mental health problems. For descriptive purposes, we labeled this profile “Competent.” Representing 32.6% of the sample, the next profile had significantly higher scores than the first on all indicators except for substance use. We labeled this profile “Primarily Internalizing” to contrast it with the third profile, “Internalizing/Delinquent,” which represented 19.3% of the sample and had significantly higher scores for self-reported delinquency, substance use, and parent-reported externalizing but significantly lower average scores on parent-reported internalizing. The Internalizing/Delinquent profile had elevated scores across all indicators compared to the Competent profile. Finally, we labeled the last profile “Extreme Symptoms.” This profile represented only 4.0% of the sample and had average scores significantly higher than all other profiles across the three self-reported indicators. The Extreme Symptoms profile also had scores significantly higher than the Competent and Primarily Internalizing profiles on parent-reported externalizing, and higher than the Competent and the Internalizing/Delinquent profiles on parent-reported internalizing. The Extreme Symptoms profile had notably elevated scores in self-reported externalizing and in substance use, with an average over three standard deviations above the sample mean for that indicator.

Figure 1.

Figure 1.

Open in a new tab

Average scores on mental health indicators by latent profile. Error bars represent standard errors of the mean. PR = parent report, SR = self-report.

Propensity Match

Out of 1,789 homeless students, 1,404 (78%) students were matched with 1,404 students who had never been homeless or doubled-up. Of these 2,808 students in the matched dataset, 2,654 (94.5%) had age 15 outcome data for inclusion in the predictive models (1,331 homeless, 1,323 non-homeless). Of the 1,331 students in the predictive models who were considered homeless based on the broad definition, 325 had histories of literal homelessness while 1,006 had been doubled-up but not literally homeless.

Predicting Mental Health Profiles Based on Homelessness

We conducted a multinomial logistic regression predicting membership in the four mental health profiles based on the broad definition of homelessness, including propensity scores as a control covariate to account for imprecise matching (Guo & Frazer, 2015). Results indicated no significant differences in profile membership based on the broad definition of homelessness, χ2(3) = 5.09, p = .17. See Table 2.

Table 2.

Model results for multinomial logistic regressions

Primarily Internalizing vs. Competent Internalizing/Delinquent vs. Competent Extreme Symptoms vs. Competent
B (SE) OR 95% CI B (SE) OR 95% CI B (SE) OR 95% CI
Propensity score 2.08 (.25) 8.00** [4.89, 13.11] 2.68 (.30) 14.6** [8.17, 26.1] 3.56 (.59) 35.2** [10.9, 112.6]
Homelessness (broad) 0.19 (.10) 1.21 [0.99, 1.47] 0.20 (.12) 1.22 [0.97, 1.53] −0.07 (.23) 0.93 [0.60, 1.46]
Propensity score 3.51 (.60) 33.4** [7.74, 24.8] 2.63 (.30) 13.9** [7.74, 24.8] 3.51 (.60) 33.4** [10.4, 107.3]
Literal homelessness (vs Non-homeless) 0.60 (.17) 1.83** [1.23, 2.55] 0.58 (.18) 1.78** [1.24, 2.55] 0.31 (.33) 1.37 [0.71, 2.63]
Literal homelessness (vs Doubled-up) 0.52 (.16) 1.69** [1.23, 2.31] 0.48 (.18) 1.61** [1.13, 2.29] 0.48 (.33) 1.61 [0.85, 3.04]
Doubled-up (vs Non-homeless) 0.08 (.11) 1.09 [0.88, 1.34] 0.10 (.12) 1.11 [0.87, 1.41] −0.16 (.24) 0.85 [0.53, 1.36]
Open in a new tab

Next, we conducted a multinomial logistic regression predicting profile membership based on three categories of homelessness: literal homelessness, doubled-up only, and never experienced literal homelessness or lived doubled-up. Percentages of students categorized in each profile by homelessness category are depicted in Figure 2. Homelessness categories significantly predicted mental health profiles, χ2(6) = 17.1, p = .009. Specifically, students who had experienced literal homelessness were significantly more likely than non-homeless students to be categorized in the Primarily Internalizing profile, OR = 1.83, p < .001, or the Internalizing/Delinquent profile, OR = 1.78, p = .002, compared to being in the Competent profile. Students who had experienced literal homelessness also were significantly more likely than doubled-up students to be categorized as Primarily Internalizing, OR = 1.69, p = .001, or Internalizing/Delinquent, OR = 1.61, p = .008. There were no significant differences in the likelihood of being in the Extreme Symptoms profile based on homelessness categories. Furthermore, students who had lived doubled-up but never experienced literal homelessness did not differ significantly in likelihood of any profile membership compared to their matched, non-homeless peers. Model information is presented in Table 2.

Figure 2.

Figure 2.

Open in a new tab

Percentages of students in each mental health profile by homelessness category.

Discussion

Only certain forms of family homelessness across childhood reduced the likelihood that youth would demonstrate resilience with regards to their mental health. Episodes of homelessness that involved staying in shelters, hotels, or other circumstances meeting definitions of literal homelessness contributed to developing mental health problems, though no association was found among adolescents who had experienced only doubled-up forms of family homelessness. We utilized a person-centered approach and a comparison group matched on key covariates to account for systematic bias, thereby strengthening confidence in these results.

Contrary to our expectations, we did not find differences in profile membership based on the broad definition of family homelessness, which included students who had experienced literal homelessness and students whose families had lived doubled-up with relatives or friends. Adolescents with histories of family homelessness based on the literal definition were less likely to show a profile with low scores across all indicators of internalizing, externalizing, and substance use. Though adolescents with histories of literal homelessness were not at elevated risk for the most problematic mental health profile, they did have significantly greater likelihood of falling within two problematic profiles, the Primarily Internalizing profile and the Internalizing/Delinquent profile.

Findings based on our person-centered approach provide a nuanced characterization of adolescent mental health profiles generally and in relation to experiences of homelessness. Generally, the configuration of symptom elevations across the three problematic profiles highlight associations among internalizing behaviors, externalizing behaviors, and substance use. Variable-centered analyses that examine these outcomes separately may tempt us to incorrectly infer that such problems usually present in different individuals or that they should be addressed with separate efforts. Neither appears likely to be true, given the results of the current person-centered analyses. Our findings show that all three problematic profiles had elevated rates of internalizing symptoms, suggesting that there is no meaningful subgroup of youth with conduct symptoms exclusively (Snyder, Young, & Hankin, 2017). The profiles also illuminate noteworthy differences in information gleaned from self- versus parent-report. Specifically, parent-reported measures did not differ significantly between the two most problematic profiles, where all three self-reported measures were significantly different. Such person-centered profiles could help pinpoint meaningful characteristics of adolescents who report particularly high rates of substance use and delinquent behavior, when parents may not be privy to these activities.

Person-centered profiles can help to elucidate individual differences in adaptation specifically with regards to students who experience homelessness. First, literal homelessness clearly represents risk for poor mental health, though about a quarter of literally homeless students showed resilience. In comparison, nearly half of the students who never experienced any sort of homelessness also showed the profile with consistently low levels of each type of symptom. Though present in the minority of cases, it remains a crucial endeavor to understand what enables resilient functioning among youth who have experienced family homelessness, as their pathways of positive adaptation in the face of adversity yield clues for informing preventive interventions (Cutuli & Herbers, 2018; Masten et al., 2015).

Second, there were two different configurations of symptom presentations which were more likely for the literally homeless students compared to those never homeless. The first involved high levels of depression and anxiety (self- and parent-report) plus conduct problems (parent-report), but low levels of substance use and average self-reported delinquency (Primarily Internalizing profile, see Figures 1 and 2). This is the most likely presentation for students experiencing literal homelessness, and they were more likely to show this presentation than either the never-homeless students or the students who had experienced doubled-up forms of homelessness only. It is possible that conduct problems are a secondary feature in this profile as irritability and argumentativeness related to internalizing symptoms may have led to more conflict with their families that impacted parent-reported externalizing symptoms. This most-likely presentation has implications for both practice and public perception: Unlike in many variable-centered accounts (e.g., Perlman, Willard, Herbers, Cutuli, & Eyrich-Garg, 2014), internalizing problems like depression and anxiety symptoms, and not substance use problems, were the predominate feature of this most-common presentation for mid-adolescents who had experienced homelessness.

Substance use and conduct problems were a feature of the second most-likely profile for students who had experienced literal homelessness, though these symptoms were present alongside elevated internalizing symptoms as well (Internalizing/Delinquent profile). This profile was more likely for the literal homelessness group relative to students who had never been homeless and students who had been doubled-up. Literal homelessness appears to contribute to two distinct presentations of symptoms, both involving internalizing problems. The less likely presentation also includes delinquency and substance use. While neither type of homelessness related to the most extreme presentation of very high substance use and both internalizing and externalizing symptoms (Extreme Symptoms profile), this profile is also rare, representing only 4% of the overall sample.

Strengths and Limitations

Findings from the current study add to a limited literature on mental health functioning of students with histories of homelessness in several noteworthy ways. First, surveys conducted longitudinally for the FFCWS included questions about literal homelessness including shelter stays as well as whether children had lived doubled-up, enabling us to consider homelessness in accordance with the educational definition and across all of childhood up to age 15. We utilized propensity score matching to identify a comparison group of non-homeless students who were like students with histories of homelessness on key covariates, allowing us to better assess impacts of homelessness beyond systematic differences in associated risks. With a person-centered approach, we examined profiles of mental health functioning to understand differences in individual students rather than individual variables. This produced a more-nuanced picture of the likely presentations of mental health problems for students as well as the risk associated with homelessness for each presentation.

The study is limited in some important ways. First, mental health symptoms are considered in a relative fashion among a sample in which high-risk families are overrepresented. We cannot consider general norms or clinical cutoffs, for example, to better contextualize the symptom levels in the current study. Second, a strength of the data is its prospective, longitudinal nature and the emphasis on mental health symptoms during the 15-year assessment, when population rates of psychopathology begin to increase. However, relations between homelessness and substance use and other mental health problems may change later in adolescence, an important target for future research. Finally, participants in this study were representative of low income families in large urban areas, so we cannot generalize our findings to families experiencing homelessness in rural areas, where doubled-up situations may be more common.

Implications for the School Context

Federal McKinney-Vento legislation requires school districts to proactively identify and engage students experiencing homelessness to ensure equitable access to a free, appropriate education and other needed services (42 U.S.C. § 11431). The current findings affirm that schools and districts should be prepared to screen for mental health problems and ensure that teens have access to quality care in response, especially those who have experienced homelessness. This response can take many forms, including referrals to school-based professionals, outside providers, or integrated care initiated, if not delivered, in the school context. Sulkowski and Michael (2014; Sulkowski, 2016) have outlined a multi-tiered system of support framework that involves universal, targeted, and intensive mental health assessments and interventions that are focused on the barriers associated with homelessness and the needs of these children and youth. Such approaches appear promising in concept and can utilize the findings of the current study to determine the needed content of each level of support. However, rigorous evaluation and tests of efficacy are needed to determine the best interventions, practices and policies for responding to student homelessness (see Herbers & Cutuli, 2014).

Acknowledgments

Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) of the National Institutes of Health under award numbers R01HD36916, R01HD39135, and R01HD40421, as well as a consortium of private foundations. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Contributor Information

Janette E. Herbers, Villanova University

J. J. Cutuli, Nemours Children’s Health System

Joanna N. Keane, Villanova University

Jake A. Leonard, Villanova University

References

  1. Achenbach TM, & Rescorla LA (2001). Manual for the ASEBA School-Age Forms & Profiles. Burlington, VT: University of Vermont, Research Center for Children, Youth & Families. [Google Scholar]
  2. Bassuk EL, Richard MK, & Tsertsvadze A. (2015). The prevalence of mental illness in homeless children: A systematic review and meta-analysis. Journal of the American Academy of Child & Adolescent Psychiatry, 54, 86–96. doi: 10.1016/j.jaac.2014.11.008 [DOI] [PubMed] [Google Scholar]
  3. Berger LM, McDaniel M, & Paxson C. (2005). Assessing parenting behaviors across racial groups: Implications for the child welfare system. Social Service Review, 79, 653–688. doi: 10.1086/454389 [DOI] [Google Scholar]
  4. Bergman LR, & Magnusson D. (1997). A person-oriented approach in research on developmental psychopathology. Development and Psychopathology, 9, 291–319. doi: 10.1017/s095457949700206x [DOI] [PubMed] [Google Scholar]
  5. Bernat DH, Oakes JM, Pettingell SL, & Resnick M. (2012). Risk and direct protective factors for youth violence: Results from the National Longitudinal Study of Adolescent Health. American Journal of Preventive Medicine, 43(2), S57–S66. doi: 10.1016/j.amepre.2012.04.023 [DOI] [PubMed] [Google Scholar]
  6. Buckner JC (2008). Understanding the impact of homelessness on children: Challenges and future research directions. American Behavioral Scientist, 51, 721–736. doi: 10.1177/0002764207311984 [DOI] [Google Scholar]
  7. Canfield JP, Nolan J, Harley D, Hardy A, & Elliott W. (2016). Using a person-centered approach to examine the impact of homelessness on school absences. Child and Adolescent Social Work Journal, 33, 199–205. doi: 10.1007/s10560-015-0420-6 [DOI] [Google Scholar]
  8. Collins LM, & Lanza ST (2010). Latent class and latent transition analysis for the social, behavioral, and health sciences. Hoboken, NJ: Wiley. [Google Scholar]
  9. Cowen JM (2017). Who are the homeless? Student mobility and achievement in Michigan 2010–2013. Educational Researcher, 46, 33–43. doi: 10.3102/0013189X17694165 [DOI] [Google Scholar]
  10. Cutuli JJ, Ahumada SM, Herbers JE, Lafavor TL, Masten AS, & Oberg CN (2017). Toxic stress and children experiencing family homelessness: Implications for health. Journal of Children and Poverty, 23, 41–55. doi: 10.1080/10796126.2016.1198753 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Cutuli JJ, Desjardins CD, Herbers JE, Long JD, Heistad D, Chan CK, … & Masten AS (2013). Academic achievement trajectories of homeless and highly mobile students: Resilience in the context of chronic and acute risk. Child Development, 84, 841–857. doi: 10.1111/cdev.12013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cutuli JJ, & Herbers JE (2014). Promoting resilience for children who experience family homelessness: Opportunities to encourage developmental competence. Cityscape, 16, 113–140. [Google Scholar]
  13. Cutuli JJ & Herbers JE (2018). Resilience in the context of development: Introduction to the special issue. Journal of Early Adolescence 38, 1205–1214. doi: 10.1177/0272431618757680 [DOI] [Google Scholar]
  14. Cutuli JJ & Herbers JE (2019). The effects of housing and service interventions on academic achievement and school factors for children experiencing homelessness. Child Development, 90, 1664–1683. doi: 10.1111/cdev.13041 [DOI] [PubMed] [Google Scholar]
  15. Cutuli JJ, Treglia D, & Herbers JE (2019). Adolescent homelessness and associated features: Prevalence and risk across eight states. Child Psychiatry and Human Development. Advance online publication. doi: 10.1007/s10578-019-00909-1 [DOI] [PubMed] [Google Scholar]
  16. Department of Housing and Urban Development. Homelessness Definition. Accessed October 10, 2019 https://files.hudexchange.info/resources/documents/HomelessDefinition_RecordkeepingRequirementsandCriteria.pdf
  17. Derogatis LR & Savitz KL (2000) The SCL–90–R and Brief Symptom Inventory (BSI) in primary care. In Maruish ME (Ed), (2000). Handbook of psychological assessment in primary care settings, (pp. 297–334). Mahwah, NJ, US: Lawrence Erlbaum Associates Publishers. [Google Scholar]
  18. Grant R, Gracy D, Goldsmith G, Shapiro A, & Redlener IE (2013). Twenty-five years of child and family homelessness: Where are we now? American Journal of Public Health, 103, e1–e10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Gubits D, Shinn M, Wood M, Bell S, Dastrup S, Solari C, … & Kattel U. (2016). Family options study: 3-year impacts of housing and services interventions for homeless families. Available at SSRN 3055295. [Google Scholar]
  20. Guo SY, & Fraser MW (2015), Propensity Score Analysis: Statistical Methods and Applications, (2nd ed.). Thousand Oaks, CA: Sage Publications. [Google Scholar]
  21. Handal PJ, Gist D, Gilner FH, & Searight HR (1993). Preliminary validity for the Langner Symptom Survey and the Brief Symptom Inventory as mass-screening instruments for adolescent adjustment. Journal of Clinical Child Psychology, 22, 382–386. doi: 10.1207/s15374424jccp2203_9 [DOI] [Google Scholar]
  22. Haskett ME, & Armstrong JM (2019). The experience of family homelessness In Fiese BH, Celano M, Deater-Deckard K, Jouriles EN, & Whisman MA (Eds.), APA handbooks in psychology series. APA handbook of contemporary family psychology: Applications and broad impact of family psychology (pp. 523–538). Washington, DC, US: American Psychological Association. doi: 10.1037/0000100-032 [DOI] [Google Scholar]
  23. Herbers JE & Cutuli JJ (2014). Programs for homeless children and youth: A critical review of evidence In Haskett M, Perlman S, & Cowan B. (Eds.) Supporting Families Experiencing Homelessness: Current Practices and Future Directions. (pp. 187–207). New York: Springer Science+Business Media. [Google Scholar]
  24. Herbers JE, Cutuli JJ, Jacobs EL, Tabachnick AR, & Kichline T. (2019). Early childhood risk and later adaptation: A person-centered approach using latent profiles. Journal of Applied Developmental Psychology, 62, 66–76. doi: 10.1016/j.appdev.2019.01.003 [DOI] [Google Scholar]
  25. Herbers JE, Cutuli JJ, Supkoff LM, Heistad D, Chan C-K, Hinz E, & Masten AS (2012). Early reading skills and academic achievement trajectories of students facing poverty, homelessness, and high residential mobility. Educational Researcher, 41, 366–374. doi: 10.3102/0013189X12445320 [DOI] [Google Scholar]
  26. Huntington N, Buckner JC, & Bassuk EL (2008). Adaptation in homeless children: An empirical examination using cluster analysis. American Behavioral Scientist, 51, 737–755. doi: 10.1177/0002764207311985 [DOI] [Google Scholar]
  27. Jobe-Shields L, Andrews AR, Parra GR, & Williams NA (2015). Person‐centered approaches to understanding early family risk. Journal of Family Theory & Review, 7, 432–451. doi: 10.1111/jftr.12118 [DOI] [Google Scholar]
  28. Kessler RC, Andrews G, Mroczek D, Ustun TB, & Wittchen HU (1998). The World Health Organization composite international diagnostic interview short-form (CIDI-SF). International Journal of Methods in Psychiatric Research, 7, 171–185. doi: 10.1002/mpr.47 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. King C. (2016). Food insecurity and housing instability in vulnerable families. Review of Economics of the Household, 16, 255–273. doi: 10.1007/s11150-016-9335-z [DOI] [Google Scholar]
  30. King KR, Lembke ES, & Reinke WM (2016). Using latent class analysis to identify academic and behavioral risk status in elementary students. School Psychology Quarterly, 1, 43–57. doi: 10.1037/e600182013-001 [DOI] [PubMed] [Google Scholar]
  31. Lanza ST, & Cooper BR (2016). Latent class analysis for developmental research. Child Development Perspectives, 10, 59–64. doi: 10.1111/cdep.12163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Laursen BP, & Hoff E. (2006). Person-centered and variable-centered approaches to longitudinal data. Merrill-Palmer Quarterly, 52, 377–389. doi: 10.1353/mpq.2006.0029 [DOI] [Google Scholar]
  33. Leech TGJ (2012). Subsidized housing, public housing, and adolescent violence and substance use. Youth & Society, 44, 217–235. doi: 10.1177/0044118X10388821 [DOI] [Google Scholar]
  34. Lo Y, Mendell NR, & Rubin DB (2001). Testing the number of components in a normal mixture. Biometrika, 88, 767–778. [Google Scholar]
  35. Marcal KE (2018). The impact of housing instability on child maltreatment: A causal investigation. Journal of Family Social Work, 21, 331–347. doi: 10.1080/10522158.2018.1469563 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Masten AS (2014). Ordinary magic: Resilience in development. New York: Guilford Publications. [Google Scholar]
  37. Masten AS, Fiat AE, Labella MH, & Strack RA (2015). Educating homeless and highly mobile students: Implications of research on risk and resilience. School Psychology Review, 44, 315–330. doi: 10.17105/spr-15-0068.1 [DOI] [Google Scholar]
  38. Masten AS, & Tellegen A. (2012). Resilience in developmental psychopathology: Contributions of the project competence longitudinal study. Development and Psychopathology, 24, 345–361. doi: 10.1017/S095457941200003X [DOI] [PubMed] [Google Scholar]
  39. Masyn KE (2013). Latent class analysis and finite mixture modeling In Little T. (Ed.), The Oxford handbook of quantitative methods (Vol. 2, pp. 551–611). Oxford, UK: Oxford University Press. [Google Scholar]
  40. McCoy DC, & Raver CC (2014). Household instability and self-regulation among poor children. Journal of Children and Poverty, 20, 131–152. doi: 10.1080/10796126.2014.976185 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. McLachlan G, & Peel D. (2000). Finite mixture models. New York, NY: Wiley. [Google Scholar]
  42. Muthén LK, & Muthén BO (2012). Mplus statistical analysis with latent variables: Users guide (7th ed.). Los Angeles, CA: Muthén & Muthén. [Google Scholar]
  43. National Center for Homeless Education. (2019). Federal data summary school year 2014–15 and 2016–17: Education for homeless children and youth. Retrieved from https://nche.ed.gov/wp-content/uploads/2019/02/Federal-Data-Summary-SY-14.15-to16.17-Final-Published-2.12.19.pdf
  44. Nylund-Gibson K, & Masyn KE (2016). Covariates and mixture modeling: Results of a simulation study exploring the impact of misspecified effects on class enumeration. Structural Equation Modeling, 23, 782–797. doi: 10.1080/10705511.2016.1221313 [DOI] [Google Scholar]
  45. Park JM, Metraux S, Culhane DP, & Mandell DS (2012). Homelessness and children’s use of mental health services: A population-based study. Children and Youth Services Review, 34, 261–265. doi: 10.1016/j.childyouth.2011.10.022 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Perlman S, Willard J, Herbers JE, Cutuli JJ, & Eyrich-Garg KM (2014). Youth homelessness: Prevalence and mental health correlates. Journal of the Society for Social Work & Research, 5, 361–377. doi: 10.1086/677757 [DOI] [Google Scholar]
  47. Perreira KM, Deeb-Sossa N, Harris KM, & Bollen K. (2005). What are we measuring? An evaluation of the CES-D across race/ethnicity and immigrant generation. Social Forces, 83, 1567–1601. doi: 10.1353/sof.2005.0077 [DOI] [Google Scholar]
  48. Radloff LS (1977). The CES-D Scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385–401. doi: 10.1177/014662167700100306 [DOI] [Google Scholar]
  49. Reichman NE, Teitler JO, Garfinkel I, & McLanahan SS (2001). Fragile families: Sample and design. Children and Youth Services Review, 23, 303–326. [Google Scholar]
  50. Seligman L, Ollendick T, Langley A, & Baldacci B. (2004). The utility of measures of child and adolescent anxiety: A meta-analytic review of the Revised Children’s Manifest Anxiety Scale, the State-Trait Anxiety Inventory for Children, and the Child Behavior Checklist. Journal of Clinical Child and Adolescent Psychology, 33, 557–565. doi: 10.1207/s15374424jccp3303_13 [DOI] [PubMed] [Google Scholar]
  51. Shaw S, Herbers JE, & Cutuli JJ (2019). Medical and psychosocial risk profiles for low birthweight and preterm birth. Women’s Health Issues, 29, 400–406. doi: 10.1016/j.whi.2019.06.005 [DOI] [PubMed] [Google Scholar]
  52. Snyder HR, Young JF, & Hankin BL (2017). Strong homotypic continuity in common psychopathology-, internalizing-, and externalizing-specific factors over time in adolescents. Clinical Psychological Science, 5, 98–110. doi: 10.1177/2167702616651076 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Straus MA, Hamby SL, Finkelhor D, Moore DW, & Runyan D. (1998). Identification of child maltreatment with the parent-child conflict tactics scales: Development and psychometric data for a national sample of American parents. Child Abuse & Neglect, 22, 249–270. doi: 10.1016/s0145-2134(97)00174-9 [DOI] [PubMed] [Google Scholar]
  54. Sulkowski ML (2016). The student homeless crisis and the role of school psychology: Missed opportunities, room for improvement, and future directions. Psychology in the Schools, 53, 760–771. doi: 10.1002/pits.21936 [DOI] [Google Scholar]
  55. Sulkowski ML, & Michael K. (2014). Meeting the mental health needs of homeless students in schools: A Multi-Tiered System of Support framework. Children and Youth Services Review, 44, 145–151. doi: 10.1016/j.childyouth.2014.06.014 [DOI] [Google Scholar]
  56. Thoemmes FJ, & Kim ES (2011). A systematic review of propensity score methods in the social sciences. Multivariate Behavioral Research, 46, 90–118. doi: 10.1080/00273171.2011.540475 [DOI] [PubMed] [Google Scholar]
  57. von Eye A, Bergman LR, & Hsieh C. (2015). Person-oriented methodological approaches In Overton WF & Molenaar PC (Eds.), Theory and method: Vol. 1. Handbook of child psychology and developmental science (7th ed., pp. 789–842). Hoboken, NJ: Wiley. [Google Scholar]
  58. Yates TM, & Grey IK (2012). Adapting to aging out: Profiles of risk and resilience among emancipated foster youth. Development and Psychopathology, 24, 475–492. doi: 10.1017/s0954579412000107 [DOI] [PubMed] [Google Scholar]

RESOURCES