Effect of a Randomized Controlled Trial of Housing Vouchers on Adolescent Risky Sexual Behavior Over a 15-Year Period

Abstract

We examined whether a housing voucher intervention influenced adolescent risky sexual behavior (RSB) across 15 years in the Moving to Opportunity Study. Low-income families in public housing that resided in 5 cities were randomized to one of three treatment groups: a housing voucher to move to low-poverty neighborhoods (i.e., < 10% poverty rate), a Sect. 8 voucher but no housing relocation counseling, or a control group that could remain in public housing. Youth and their caregivers completed baseline surveys, as well as two uniform follow-ups: interim (2001–2002; 4–7 years after baseline) and final (2008–2010; 10–15 years after baseline). Approximately 4,600 adolescents (50.5% female) aged 13–20 years participated at the final timepoint. Adolescents reported on their RSB, including condom use, other contraceptive use, early sexual initiation (< 15 years old), and 2+ sexual partners in the past year. We modeled each indicator separately and as part of a composite index. We tested baseline health vulnerabilities as potential effect modifiers. The low-poverty voucher group and the Sect. 8 voucher group were combined due to homogeneity of their effects. Applying intent-to-treat (ITT) regression analyses, we found no significant main effects of voucher receipt (vs. control) on any RSB. However, we found protective effects of voucher receipt on RSB among youth with health problems that limited activity, and youth < 7 at baseline but adverse effects among females, youth > 7 at baseline, and youth who were suspended/expelled from school. Results highlight the importance of understanding how housing interventions differentially influence adolescent health and behaviors.

Introduction

Preventing risky sexual behaviors (RSBs, i.e., risky behaviors related to sexual relationships) represents a key population health priority (US Department of Health & Human Services, 2020). Developmental psychopathology and evolutionary models highlight adolescence as a biological and social transition during which risky behaviors may be more likely (Ellis et al., 2012). More than 40% of sexually active adolescents, for example, reported not using a condom during their last sexual encounter (Szucs et al., 2020). Engaging in RSB in adolescence can have consequences that persist across the life course, such as an increased risk of alcohol use disorder and poorer later-life health (Epstein et al., 2018), infertility (Tsevat et al., 2017), and cancer (Cohen et al., 2019). RSB can also increase the risk of adverse reproductive health outcomes, including sexually transmitted infections (STI; Kugler et al., 2015) and unwanted pregnancies, among other short-term consequences (Newbern et al., 2013). The incidence of STI is increasing in the USA, with more than half of all new STIs each year occurring among adolescents (Bowen et al., 2019). Given that RSB in adolescence increases the risk of STI (among other adverse reproductive outcomes), identifying upstream causes of RSB, as well as points for prevention and intervention, is important. However, few studies have leveraged an experimental design to examine upstream causes of RSB.

Neighborhoods and Risky Sexual Behavior

The neighborhood context operates as a fundamental determinant of reproductive health behaviors and outcomes (Phelan & Link, 2015). As such, interventions or policies that target where a person lives may offer promising solutions to reduce RSB. Neighborhoods influence health and behaviors through many pathways, such as the availability of health-promoting resources and services, social norms, and opportunity structures (Cubbin et al., 2005). Current approaches to understanding these so-called neighborhood effects consider the structural characteristics of neighborhood networks and how individuals interact with them, including characteristics of the residents and neighborhood spaces (Browning et al., 2015).

While the neighborhood context has long been of interest to researchers, the number of articles published each year has accelerated over the past three decades. Yet, work in this area remains methodologically weak; for instance, most studies are observational and cross-sectional, which presents limitations to causal inference (Oakes et al., 2015). Few studies examine interventions, and fewer still, randomized interventions. In addition, little research investigates multiple indicators of RSB simultaneously, which holds relevance, given that behavioral risks for STI and other adverse reproductive outcomes tend to co-occur (Brener & Collins, 1998). Below, we discuss prior research examining associations between the neighborhood context and relevant RSB, highlighting the need for experimental research that focuses on RSB explicitly.

Early Sexual Initiation

Previous observational studies, both cross-sectional and longitudinal, have consistently shown that adverse neighborhood conditions are linked to earlier sexual initiation (Decker et al., 2018). For instance, both greater neighborhood decay and lower collective efficacy predicted a younger age at sexual initiation (Orihuela et al., 2020).

Contraceptive Use

Research often shows that residents of disadvantaged neighborhoods exhibit reduced likelihood of contraceptive use (Decker et al., 2018). However, these associations vary with context. For example, in one study, neighborhood stress was associated with RSB for adolescents in the Northeast USA, but not the Southeast (Kerr et al., 2015). Another study found that neighborhood disadvantage predicted lower condom use among adolescents cross-sectionally, but not over time (Bauermeister et al., 2011).

Multiple Sexual Partners

Numerous studies have demonstrated correlations between residing in poorer neighborhoods and more sexual partners (Carlson et al., 2014; Warner, 2018). Moreover, residing in a neighborhood with greater deterioration was associated with a higher risk of multiple sexual partners (Voisin & Kim, 2018).

Risky Sexual Behavior Indices

Although most research has investigated indicators of RSB among adolescents separately, a few studies have examined RSB indices. One observational study showed that parent perceptions of lower neighborhood safety and quality predicted an index of RSB, including having sex for drugs/money, nonromantic sex, anal sex, anal sex with nonromantic sexual partners, and no condom use (Chen et al., 2010). Another study found that neighborhood poverty, safety, drug activity, and crime each predicted RSB (Green et al., 2019).

Moving to Opportunity Study and Risky Sexual Behaviors

The observational studies described above, while informative, are weak for inferring a causal connection between neighborhood context and RSB. Nearly all extant experimental approaches to the topic—including this analysis—makes use of the Moving to Opportunity Study (MTO). MTO was a randomized controlled trial of housing vouchers among low-income families living in public housing across five large US cities, starting in 1994. Families volunteered and were randomly assigned to receive one of three types of housing assistance: one group received a housing voucher redeemable only in low-poverty neighborhoods (less than 10% of the population below the poverty line) and received housing counseling (“low-poverty neighborhood voucher group”); a second group received a Sect. 8 voucher (the US Department of Housing and Urban Development [HUD] has renamed this to a Housing Choice Voucher) which subsidized housing costs without locational restrictions (“Sect. 8 voucher group”); a third group did not receive a housing voucher, but remained eligible to stay in public housing (“control group”). Follow-up surveys evaluated impacts on families’ earnings, employment, well-being, health, and behaviors. Although researchers have investigated the effect of MTO on adolescent RSB 4–7 years after randomization, the long-term (i.e., 10–15 years after randomization) effects of MTO on these behaviors remain unknown. This represents a critical gap in existing research because the effect of MTO on RSB among those who were in early childhood at randomization has not yet been investigated, as the existing research was conducted among those who were already adolescents at randomization.

Effect Modification by Gender and Age

At the interim survey (4–7 years after randomization), studies found null effects of MTO housing voucher receipt (low-poverty neighborhood vouchers and Sect. 8 vouchers, each compared to the control group) on indices of general risky behaviors (e.g., marijuana, smoking, or alcohol use in the past 30 days and having ever gotten someone pregnant) among adolescents overall (Orr et al., 2003; Schmidt et al., 2017). However, an examination of gender differences revealed that housing voucher receipt predicted fewer risky behaviors (i.e., substance use and sexual behaviors) among adolescent girls and more risky behaviors among adolescent boys (Orr et al., 2003). Other MTO research examining RSB on its own showed that receiving a housing voucher to move to a low-poverty neighborhood increased the likelihood of sexual initiation for girls, but not for boys (Leventhal & Dupéré, 2011), in the older cohort of children when they were ages 12–19 (interim survey). The age of the children at the time of random assignment (i.e., baseline) also played an important role; Schmidt et al. (2017) showed that housing voucher receipt was particularly deleterious for risky behaviors among boys aged 10 years and older at baseline. This effect modification in opposite directions of the MTO voucher exposure on health and risky behaviors by gender (and age) was a strong pattern for the initial cohort of children at the interim survey. Additionally, the developmental timing of relocation with respect to the age of the child may impact the effect of housing voucher receipt, such that the trauma of moving during middle childhood or adolescence and disrupting social groups may have an adverse impact on RSB (Schmidt et al., 2017).

Effect Modification by Children’s Health and Developmental Problems

While not yet explored in relation to RSB, baseline health may also modify effects of housing voucher receipt on sexual risk behaviors in adolescence. Prior MTO work has found that special needs of the child modified the effect of housing voucher receipt on alcohol use (Thyden et al., 2022). Specifically, housing voucher receipt was protective against alcohol use for youth in families with children with special health care or educational needs, which may be explained by greater access to the resources required by these families (Thyden et al., 2022). This contrasts with prior findings that more vulnerable groups may not benefit from an intervention to the same extent as less vulnerable groups (McCormick et al., 2006; Osypuk et al., 2012a, 2012b).

Purpose of the Present Study

We used the MTO Study to estimate how receiving a housing voucher, compared to remaining in public housing, affects sexual health 10–15 years later among adolescents who were children at baseline. We expected that youth from families that received a housing voucher (low-poverty voucher and Sect. 8 groups) when they were children (ages 0 to 11) would exhibit fewer RSB compared to controls. Further, we tested effect modification by baseline socioemotional health vulnerabilities indicating special needs, gender, age, and race. We anticipated that the positive effect of housing voucher receipt on RSB would be stronger for girls and for younger compared to older adolescents. For baseline socioemotional health vulnerabilities, we hypothesized that children with special needs would have stronger effects of housing voucher receipt compared with children without special needs.

Method

Participants

The MTO Study was a randomized controlled trial of housing vouchers sponsored by HUD in five US cities (Baltimore, Boston, Chicago, Los Angeles, and New York). Families were eligible if they were low income, had at least one child under 18 years old in the household, resided in public housing in distressed (i.e., high poverty) areas, and qualified for federal rental assistance. Adults filled out informed consent, and assent was obtained from youth under 18. Baseline surveys were completed by household heads and their children. After the initial enrollment period (1994–1998), there were two uniform follow-ups: the interim time point (2001–2002; 4–7 years after baseline) and final time point (2008–2010; 10–15 years after baseline) where adults and children completed additional surveys.

We used the MTO Tier 1 Restricted Access Data version of the data, which is available to qualified users to analyze within a Federal Statistical Research Data Center (FSRDC). These analyses employ restrictive protections around the data and presentation of results to preserve confidentiality. Therefore, some numbers may be suppressed or rounded. These analyses have undergone strict review prior to release from the FSRDC.

Exposure of Interest: Moving to Opportunity Treatment Assignment

Upon study enrollment, families (N = 4600) were randomly assigned to one of three treatment groups. The MTO low-poverty neighborhood voucher group was offered a housing voucher that would subsidize rent for a neighborhood with a less than 10% poverty rate and provided housing relocation counseling. The Sect. 8 voucher group was offered a standard Sect. 8 voucher, which aligned with the Sect. 8 rules at the time, that could be redeemed in any neighborhood, and did not receive housing relocation counseling. Vouchers for the low-poverty neighborhood and Sect. 8 groups expired after 90 days if not used. Lastly, the control group did not receive a housing voucher but could remain in public housing. We conducted a statistical test of homogeneity to determine whether the two voucher groups could be combined to increase power, which is appropriate if both voucher groups showed similar associations with the outcomes, compared to the control group. We modeled the effect of being offered a housing voucher (intent-to-treat analysis) which preserves the causal effect of randomization.

Assessment

We used outcome data from the final survey of adolescents, in 2008–2010 for this analysis. We excluded 10–12-year-olds because sexual activity is uncommon among those 12 and younger (Finer & Philbin, 2013). We also do not have children in this final survey sample who were adolescents at baseline, as they had aged out of the household by 2008 and were not surveyed at the final survey. Thus, the sample represents youth that were 13–20 years old at the time of the final survey assessment (ages 0 to 11 at baseline). The survey was conducted via computer-assisted personal interviewing technology, with youth completing surveys privately.

Measures

Outcome: Risky Sexual Behaviors

We assessed RSBs with four self-reported items, which have been validated and used in nationally representative US surveys (Moore et al., 1999). Items included: no condom use during last sexual intercourse, no contraceptive use during last intercourse, early sexual initiation (i.e., engagement in sexual intercourse before the age of 15; Spriggs & Halpern, 2008), and two or more sexual partners in the past year. Prior to responding to engagement in RSB, all respondents reported whether they had ever had sex (“Have you ever had sexual intercourse, that is, made love, had sex, or gone all the way?”). Those that had not engaged in sexual intercourse were not asked the follow-up questions and were given a zero for our index. The index was the fraction of the four RSB items (range 0–1), consistent with prior research investigating risky behaviors among youth (Schmidt et al., 2017).

Effect Modifiers

Socioemotional Health Vulnerabilities

We defined socioemotional health vulnerabilities at baseline, including (1) problems that made it difficult to get to school or play active games, (2) behavior or emotional problems in the past 2 years, (3) learning problems in the past 2 years, (4) calls from the school about behavior in the past 2 years, (5) current health problems that required special medication or equipment, and (6) school suspension or expulsion in the past 2 years. Each health vulnerability was coded dichotomously as “no” (0) or “yes” (1).

Demographic Characteristics

We modeled gender, as defined at baseline as male or female at birth, as an effect modifier. Age in years at baseline was calculated from date of birth. Race/ethnicity was defined as Black versus Not Black, given that the sample includes predominantly Black and Hispanic children, with few other racial/ethnic identities.

Covariates

We included site (city, using New York City as referent) and household size (2, 3, 4, and 5 + people) as covariates in all models. While adjustment for covariates is not strictly necessary in a randomized trial design, we adjusted for any covariates that were imbalanced at baseline.

Data Analysis

We examined descriptive statistics for all included variables. Intent-to-treat (ITT) analyses, which include everyone randomly assigned to treatment regardless of if they took up treatment. ITT analyses demonstrate the average effect of being randomly assigned to treatment, compared to controls, on our primary outcome: the fraction of risky behaviors in which participants engaged. This main effect model was modeled via linear regression. We also separately modeled the odds of engaging in any of the four RSB comprising the index using logistic regression.

We estimated the two-way interactions between housing voucher receipt and each effect modifier (e.g., gender), with a significant interaction indicating statistical difference in treatment effects on the outcome. We followed up significant interactions with post hoc probing via simple slope analyses (Aiken et al., 1991), such as presenting the housing voucher receipt associations with RSB, stratified by gender. We adjusted all regression models for demographics (gender, age, household size and race), site, and any variables that were imbalanced at baseline. We weighted the analyses to account for (1) changing random assignment ratios, (2) attrition, (3) sampling of children within households. We imputed missing data for the baseline covariates using Multivariate Imputation by Chained Equations. Analyses were conducted using STATA version 16.1, in the University of Minnesota FSRDC computing environment.

Results

Descriptive Statistics

Baseline variables were balanced across MTO treatment groups (i.e., housing voucher receipt vs. controls) as expected given a randomized design of this size (see Table 1), except for whether the adult had a GED and whether they had applied for Sect. 8 before, and thus were controlled for. The weighted means of baseline variables overall and by treatment group for adolescents aged 13–20 years old in 2008–2010 are also presented in Table 1.

Table 1.

Descriptives of Moving to Opportunity baseline (1994–1997) covariates and effect modifiers, overall and by treatment group

	All youth (N = 4,600)	Control group (N = 1,500)	Combined voucher group (N = 3,200)	Test for imbalance p-value
Youth demographic characteristics
Age (mean)	5.01	5.10	4.97	0.167
Gender
Male	49.5	51.2	48.6	0.145
Female	50.6	48.8	51.4
Race
Black	66.8	66.3	67.1	0.734
Other	33.2	33.8	32.9
Hispanic	32.4	33.5	30.8	0.474
Youth socioemotional health vulnerabilities
Behavioral or emotional problems past 2 years	2.5	2.7	2.3	0.561
Learning problem past 2 years	5.3	6.0	5.0	0.300
School called about behavior past 2 years	8.6	8.4	8.6	0.821
Current health problems that limit activity	5.3	5.3	5.3	0.974
Current health problems require medicine or equipment	7.8	7.3	8.0	0.473
Suspended or expelled from school past 2 years	1.5	1.4	1.5	0.894
Baseline covariates
MTO study site
Baltimore	13.4	12.7	13.8	0.776
Boston	18.5	19.3	18.2
Chicago	23.1	22.2	23.5
Lost Angeles	23.1	22.6	23.3
New York	21.9	23.3	1.9
Household size
Two	8.6	7.6	9.1	0.199
Three	24.6	25.6	24.1
Four	26.0	24.0	26.9
Five or more	40.9	42.8	40.0
Mother had GED	18.5	22.2	16.8	0.010
Mother graduated from high school	35.0	33.9	35.5	0.479
Mother currently enrolled in school	18.1	18.0	18.1	0.980
Mother never married	66.0	67.2	65.5	0.462
Mother had first child before age 18	29.4	29.0	29.6	0.783
Mother employed	22.3	21.6	22.6	0.618
On AFDC or TANF	81.4	80.8	81.8	0.593
Parent had a car	18.5	16.8	19.3	0.182
Household member had a disability	14.4	15.6	13.9	0.321
No teenagers in household	74.7	76.1	74.1	0.330
Household member a victim of violent crime past 6 months	42.7	41.3	43.4	0.394
Mother felt sure they would find an apartment	49.7	47.2	50.9	0.145
Mother moved more than three times in past five years	10.9	12.0	10.4	0.326
Mother wanted to move to escape drugs/gangs	76.9	77.4	76.6	0.687
Mother wanted to move for better schools	53.2	50.0	54.7	0.055
Mother previously applied for Sect. 8 voucher	38.9	43.5	36.7	0.005
Mother lived in neighborhood for 5 + years	55.7	57.1	55.0	0.398
Mother chats with neighbor at least once/week	52.4	52.7	52.3	0.887
Mother would talk to neighbor if child got in trouble	55.0	55.8	54.6	0.627
Mother had no family in neighborhood	63.6	64.7	63.1	0.525
Mother had no friends in neighborhood	40.1	40.8	39.7	0.653
Mother felt streets were very unsafe at night	51.1	49.0	52.1	0.218
Mother very dissatisfied with neighborhood	48.6	46.2	49.7	0.164

Bold indicates significant at p < .05. Means were weighted to account for survey design and family clustering. Sample sizes are rounded according to FSRDC rounding rules. All results were approved for release by the US Census Bureau, authorization number CBDRB-FY23-CES018–006

Primary Analyses

The low-poverty voucher group and the Sect. 8 voucher group were combined into one “housing voucher” group for all models due to nonsignificant tests of heterogeneity across treatment groups. Results from ITT models of the main effect of housing voucher receipt on the RSB index and the four indicators of RSB are presented in Table 2. Receiving a housing voucher did not predict overall risky sexual behaviors (B = 0.001; 95% CI -0.02, 0.02) or any of its components. Specifically, housing voucher receipt showed null associations with the odds of having 2+ sexual partners in the past year (OR = 0.99; 95% CI 0.83, 1.17), condom use during last sexual encounter (OR = 1.00; 95% CI 0.81, 1.23), other contraceptive use during the last sexual encounter (OR = 1.01; 95% CI 0.86, 1.18), and early sexual initiation (OR = 1.30; 95% CI 0.95, 1.76), compared to controls.

Table 2.

Intent-to-Treat effects of Moving To Opportunity voucher treatment compared to control group on youth risky sexual behaviors at final survey occasion (2008–2010, N = 4,600)

		Control group	Combined voucher group	Combined voucher treatment group vs. control group effect
	Test for imbalance			Treatment effect
	p-value	Mean	Mean	(OR or B)	SE	p-value	LCI	UCI
Ever engaged in sexual activity	.596	0.664	0.654	–	–	–	–	–
Risky sexual behavior index	.783	0.247	0.243	0.001	0.010	.893	−0.017	0.020
Early sexual initiation	.217	0.060	0.075	1.296	0.157	.099	0.952	1.765
2+ partners in the past year	.383	0.364	0.350	0.986	0.085	.866	0.834	1.165
No condom use during last intercourse	.549	0.170	0.168	0.999	0.105	.989	0.813	1.226
No other contraceptive use during last intercourse	.621	0.378	0.373	1.008	0.080	.926	0.861	1.179

B is presented for the model predicting the risky sexual behavior index and OR is presented for the models predicting early sexual initiation, 2+ partners in the past year, no condom use during last intercourse, and no other contraceptive use during last intercourse. We controlled for MTO site, household size, age, race, gender, and the baseline covariates that were imbalanced at baseline (i.e., whether the mother had obtained a GED and whether the mother had applied for a Sect. 8 before) in each model. Model results were weighted for sampling design and household clustering. All results were approved for release by the US Census Bureau, authorization number CBDRB-FY23-CES018–006

Effect Modification by Socioemotional Health Vulnerabilities

The p-values from a series of regressions testing the interactions between housing voucher receipt with child socioemotional health vulnerabilities on RSB are presented in Table 3 (see Supplementary Table 1 for full model results). Whether a child had baseline health problems that limited activity significantly modified the effect of housing voucher receipt on overall RSB (B_interaction = −0.08; 95% CI −0.16, −0.01), such that for youth with problems that limited activity, receiving a housing voucher predicted fewer RSB compared to controls (B = −0.08; 95% CI −0.15, −0.004; see Fig. 1) and lower likelihood of no condom use during last intercourse (OR = 0.42; 95% CI 0.19, 0.94). But for the group of children who did not have such health problems, there was no significant effect of a housing voucher on overall RSB (B = 0.01; 95% CI −0.01, 0.03) nor condom use during last intercourse (OR = 1.05; 95% CI 0.85, 1.30).

Table 3.

Interaction test results: p-values for interactions between MTO voucher treatment vs. control, and baseline effect modifier, on youth risky sexual behavior outcomes (2008–2010)

Effect modifier	Risky sexual behavior index	Early sexual initiation	2+ partners in past year	No condom use during last intercourse	No contraceptive use during last intercourse
Sociodemographic characteristics
Gender	0.625	0.074	0.975	0.900	0.261
Age	0.218	0.392	0.453	0.712	0.064
Race	0.886	0.384	0.700	0.719	0.258
Socioemotional health vulnerabilities
Behavioral or emotional problems past 2 years	0.717	0.281	0.290	0.425	0.800
Learning problem past 2 years	0.694	0.871	0.726	−0.52	0.507
School called about behavior past 2 years	0.177	0.284	0.617	0.278	0.679
Current health problems that limit activity	0.030	0.239	0.183	0.035	0.287
Current health problems require medicine or equipment	0.314	0.482	0.945	0.370	0.186
Suspended or expelled from school past 2 years	0.044	0.201	0.826	0.017	0.017

We controlled for MTO site, household size, age, race, gender, and the baseline covariates that were imbalanced at baseline (i.e., whether the mother had obtained a GED and whether the mother had applied for a Sect. 8 before) in each model. Model results were weighted for sampling design and household clustering. All results were approved for release by the US Census Bureau, authorization number CBDRB-FY23-CES018–006

Fig. 1.

Effect of MTO (1994–1997) on youth risky sexual behaviors (index) (2008–2010) by presence of baseline health problems limiting activity. We controlled for MTO site, household size, age, race, gender, and the baseline covariates that were imbalanced at baseline (i.e., whether the mother had obtained a GED and whether the mother had applied for a Sect. 8 before). Model results were weighted for sampling design and household clustering. All results were approved for release by the U.S. Census Bureau, authorization number CBDRB-FY23-CES018–006

There was also a significant interaction effect of a housing voucher on overall RSB by school suspension or expulsion in the past 2 years at baseline (B_interaction = 0.22; 95% CI 0.01, 0.43). For youth that were suspended or expelled in the past 2 years at baseline, we found a harmful effect of housing voucher treatment versus controls on the RSB index (i.e., greater RSB; B = 0.22; 95% CI 0.004, 0.43, see Fig. 2). When examining the RSB indicators separately, we found a harmful effect of housing vouchers, with increased odds of no condom use during last intercourse (OR_interaction = 10.18; 95% CI 1.49, 69.76) and increased odds of not using other contraceptives during last intercourse (OR_interaction = 6.16; 95% CI 1.36, 27.85) among suspended or expelled youth in housing voucher groups. However, for the group of children that were not suspended or expelled, there was no significant effect of housing vouchers on condom use during last intercourse (OR = 0.96; 95% CI 0.78, 1.18) nor other contraceptive use during last intercourse (OR = 0.98; 95% CI 0.84, 1.14).

Fig. 2.

Effect of MTO treatment (1994–1997) on youth risky sexual behaviors (index) (2008–2010) by whether youth had been suspended or expelled from school in the past 2 years at baseline. We controlled for MTO site, household size, age, race, gender, and the baseline covariates that were imbalanced at baseline (i.e., whether the mother had obtained a GED and whether the mother had applied for a Sect. 8 before). Model results were weighted for sampling design and household clustering. All results were approved for release by the U.S. Census Bureau, authorization number CBDRB-FY23-CES018–006

We did not detect effect modification of housing voucher receipt on the examined outcomes for any other baseline socioemotional health vulnerabilities (i.e., behavior or emotional problems, learning problems, calls from the school about behavior, and health problems that required special medication or equipment).

Effect Modification by Demographic Characteristics

Table 3 presents p-values from interactions between housing voucher receipt and demographic characteristics on RSB (see Supplementary Table 1 for full model results). Interestingly, effect modification of housing voucher receipt by age-approached significance in the model predicting likelihood of contraceptive use during last intercourse (OR_interaction = 1.06; 95% CI 1.00, 1.13). As shown in Fig. 3, youth in the voucher group that were between 0 and 7 years of age at baseline were less likely to report not using contraception during last intercourse (i.e., a protective effect of housing voucher receipt on using contraception for younger kids) compared to controls (OR’s from 0.25 to 0.81; see Supplementary Table 1 for 95% CI’s). However, there was a harmful effect of housing voucher receipt on contraceptive use for youth that were older than 7 at baseline (OR = 1.35; 95% CI 1.15, 1.59). Effect modification of housing voucher receipt by age was not significant for any other outcome. There were no significant interactions between housing voucher receipt and gender for any of the examined outcomes.

Fig. 3.

Effect of MTO (1994–1197) on likelihood of no contraceptive use during past intercourse by age at randomization. We controlled for MTO site, household size, race, gender, and the baseline covariates that were imbalanced at baseline (i.e., whether the mother had obtained a GED and whether the mother had applied for a Sect. 8 before). Model results were weighted for sampling design and household clustering. All results were approved for release by the US Census Bureau, authorization number CBDRB-FY23-CES018–006.

Discussion

In this study, we used the MTO randomized trial to investigate whether receiving a housing subsidy voucher to move to private housing, compared to remaining in public housing, affected risky sexual behaviors among low-income adolescents after 15 years. We also tested modification of this effect by baseline socioemotional health vulnerabilities and sociodemographic characteristics. This study had two primary findings. First, housing voucher receipt did not significantly predict the fraction of RSB (as measured by an index comprising four behaviors) or its component behaviors among youth overall. Second, tests of effect modification revealed variation in housing voucher receipt effects. We found protective effects of housing voucher receipt on RSB among youth with health problems that limited activity, and youth ages 7 or younger at baseline. However, we found adverse effects of housing voucher receipt on RSB among females, youth older than age 7 at baseline, and youth who were suspended or expelled from school at baseline.

One way to interpret our findings assumes that families who used the housing voucher ended up in lower poverty neighborhoods. Indeed, families in the MTO low-poverty neighborhood voucher group lived in 17.1 percentage points lower poverty at 1 year, 9.8 percentage points lower at 5 years, and 4.9 percentage points lower at 10 years post random assignment with slightly attenuated effects in the Sect. 8 voucher group (Ludwig et al., 2011). Both voucher groups also moved to neighborhoods with improved economic, social, and physical conditions (Nguyen et al., 2017) and improved neighborhood opportunities for children (Kim et al., 2023).

Comparisons with Previous Research

Main Effect of Moving to Opportunity Treatment on Risky Sexual Behaviors

Contrary to our expectations, there was no main effect of housing voucher receipt on the RSB index among adolescents. Generally, the neighborhood context is associated with RSB in observational studies (Decker et al., 2018). Within MTO, some prior research suggests that voucher receipt is linked to fewer risky behaviors, including RSB (Schmidt et al., 2017) at the interim survey but not at the final survey (Gennetian et al., 2012). The differences in findings across studies may also be attributed to differences in the operationalization of risky behaviors (i.e., an index of 10 general risky behaviors in Schmidt et al., 2017 and Gennetian et al., 2012, versus an index specific to RSB in the present study).

Effect Modification by Socioemotional Health Vulnerabilities

Health Problems That Limit Activity

Housing voucher receipt predicted fewer RSB for youth with health problems limiting activity at baseline, which was consistent with our hypotheses. This finding aligns with previous MTO research which found that housing voucher receipt reduced the likelihood of ever drinking alcohol only for youth with socioemotional health vulnerabilities at baseline (Thyden et al., 2022). The protective effect of housing voucher receipt may be partial because kids with health issues experienced better accommodations, such as higher-resourced schools, that allowed their health conditions to be less impactful in everyday life.

School Suspension or Expulsion in Past 2 Years

Contrary to the findings among children with health problems, children who had been suspended or expelled from school at baseline experienced a harmful effect of housing voucher receipt on RSB. This aligns with adverse effects of MTO treatment on youth outcomes, including behavioral problems, substance use, mental health symptoms, and asthma (Kling et al., 2007; Orr et al., 2003; Osypuk et al., 2012a, 2012b). Perhaps school suspension or expulsion indicates a unique type of challenge and is not comparable to having a health problem limiting physical activity. For example, a history of school suspension may signal inability or unwillingness to assimilate into new environments and social norms.

Effect Modification by Sociodemographic Characteristics

Gender

We did not detect effect modification by gender on any of the examined RSB outcomes. This contrasts with previous MTO findings from the interim time point, suggesting that housing voucher receipt increases the likelihood of sexual initiation among girls, but not boys (Leventhal & Dupéré, 2011), and with many studies out of MTO showing gender effect modification of its effects on behavioral outcomes. Other qualitative MTO research found girls’ risk of sexual violence was high in baseline neighborhoods and relief or reduction in sexual violence may have generated beneficial effects for girls (e.g., mental health; Osypuk et al., 2012a, 2012b; Popkin et al., 2008), and one may hypothesize that this safety from sexual violence might result in reduced early sexual initiation for girls in the voucher treatment group vs. controls.

Age

The interaction between voucher receipt and age only trended toward significant, which may be due to the MTO study not being designed or powered to detect subgroup effects. However, because of the developmental significance of this finding, coupled with the unique and policy-relevance of the MTO study, we believe that reporting this marginally significant effect can be useful for key stakeholders and future researchers in this field. Our findings coalesce with other studies that have documented harmful effects of housing voucher receipt among the cohort of boys who were older (adolescents) at the time of random assignment/baseline. The boys exhibited harmful effects of the housing voucher on substance use including binge alcohol drinking (Osypuk et al., 2019), smoking (Orr et al., 2003), and mental health outcomes (psychological distress, behavior problems; Osypuk et al., 2012a, 2012b), at the interim survey. Our results also showed that kids who were randomized earlier in childhood exhibited beneficial effects of housing voucher receipt on contraceptive use. This aligns with Chetty et al.’s (2016) finding that housing voucher receipt improved income in early adulthood (compared to controls) if the children were under age 13 at baseline. This suggests there is a sensitive period in development for neighborhood exposures on outcomes, including sexual risk behaviors. For example, it is possible that the increased autonomy that often occurs across childhood may lead youth to spend more time in their neighborhoods, and thus, the effect of residing in an impoverished neighborhood may persist into adolescence.

It is important to note that this intervention was meant to improve families’ lives, and a body of work has identified some of the beneficial effects of the housing voucher treatment. However, we (and others) have documented that housing vouchers sometimes lead to adverse effects. Implications that flow from this include that it is necessary to supplement this relatively simple intervention, that occurred only in the housing sector, with supports outside of the housing sector, to ensure that all families benefit from housing mobility and housing interventions. We must learn what we can from this study, to improve housing programs. For example, more recent housing mobility programs include financial counseling, driver’s education assistance, landlord recruitment, and handling maintenance requests (Poverty & Race Research Council, 2022).

Strengths and Limitations

The primary strength of this study is the randomized controlled trial design, where the random assignment controls confounding (both measured and unmeasured), leading to strong causal inference (Aschengrau & Seage, 2013). This design makes MTO one of the most powerful studies of how the neighborhood context, a fundamental cause of population health (Link & Phelan 1995), shapes adolescent health. This study also had long-term follow-up, which enables us to see how an exposure that occurred across different time points in a child’s early life affected their adolescent sexual risk about 10 years later.

However, the findings of our study should be interpreted in the context of some limitations. Second, the MTO study was not originally powered to test effect modification. We did combine the two voucher groups, which increased power to detect effects. However, we were often underpowered with trending significance rather than interactions below traditional significance thresholds. Even though these tests were underpowered, exploring heterogeneity of effects is important for unpacking upstream causal effects. This is particularly important since such effects may be complicated across time and space and difficult to hypothesize a priori (Glymour et al., 2013). Third, as is common in the literature on RSB, several items pertaining to sexual behavior were most applicable for those in heterosexual relationships (i.e., condom use during last intercourse and other contraceptive use during last intercourse) and may not generalize to those in non-heterosexual relationships. Given the rise in publicly identifying as a sexual and/or gender minority, it is important to incorporate more diverse definitions of sexual activity (Matthews et al., 2014). Fourth, given that MTO was originally designed to test the effect of housing vouchers on economic outcomes, the measurement of special needs of a child at baseline is crude. While clinical, diagnostic, or legal definitions might better operationalize a special needs child, such measures were not collected in MTO at baseline (e.g., having an individual education plan, or whether a doctor diagnosed a child with autism). Lastly, we focused on intention-to-treat models and did not model treatment adherence (i.e., accounting for whether families used the voucher). Approximately half of those offered a voucher (46% of low-poverty neighborhood group; 59% of Sect. 8 voucher group) did not use it, so the ITT effects in this case will estimate a watered-down effect. However, prior MTO research that has adjusted for non-adherence has found that housing voucher receipt effects are roughly twice as large, with a similar pattern of effects (Ludwig et al., 2011).

Implications

Housing vouchers remain one of the primary affordable housing policies for low-income families in the USA (National Low Income Housing Coalition, 2022). As such, findings from the MTO study are informative for housing policy. From a developmental perspective, the findings of our study suggest that there are sensitive/critical periods for reducing the risk of RSB for youth residing in disadvantaged neighborhoods. To confer the greatest benefit on reducing the risk of RSB, housing choice voucher policies may prioritize assisting families with young children or families with children who have health-related special needs.

Conclusions

This study deepens our understanding of the benefits and risks of housing relocation programs on sexual behavior, illuminating several demographic and contextual factors to be considered when designing and implementing housing policy. Moreover, our findings highlight the importance of neighborhood environments in shaping adolescent health. Although housing choice voucher programs are a promising approach to aid low-income families, the demand far outweighs the resources allocated for these programs (Ellen, 2020). In addition to expanding funding for these programs at the federal and state level, there is a need to address the structural conditions that give rise to poor and inequitable neighborhood conditions, to improve population health.

Data availability

The data and code for this manuscript are in the Federal Statistical Research Data Center and are not available for public use.