What Predicts Disconnection Among American Youth?

Young people who are neither in school nor working, often called disconnected youth or opportunity youth, face challenges that can lead to lower lifetime earnings, poorer health, and lower socioeconomic outcomes (Belfield, Levin, and Rosen, 2012; MaCurdy et al., 2006; Hair et al., 2009; Lewis, 2021). Disconnection also generates broader social costs through lost productivity and higher social spending. Effective policy solutions require a clear understanding of the factors that lead some youth to become disconnected.

Existing research points to a variety of potential factors that might influence disconnection in young adulthood, including family environment, mental health, educational attainment, and behavioral factors, among others (Cohen and Wills, 1985; Currie and Thomas, 2001; Furstenberg and Hughes, 1995; Heckman, Stixrud, and Urzua, 2006; Hanushek and Woessmann, 2008). However, data on these measures is limited and researchers are rarely able to follow young people prior to and across spells of disconnection.

In this paper, we use data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) to examine who disconnected youth are and what factors precede disconnection (Harris and Udry, 2022). Add Health is a representative sample of adolescents who were in grades 7 through 12 during the 1994–1995 school year and have been followed through 2018. Our sample of disconnected youth consists of individuals aged 18 to 24 who were not in school and not working at the time of the third wave of the Add Health survey (conducted between 2001 and 2002).⁠^[1] Nearly all (99.7 percent) of our respondents were connected at baseline, given that the initial sampling was restricted to students in middle and high school.⁠^[2]

Overall, our findings highlight both the complexity of the pathways leading to disconnection and the potential for early targeted interventions to alter these trajectories. This paper will be of interest to researchers, policymakers, and practitioners who are developing programs that reconnect youth with education, training, or employment.

Who Is Disconnected?

We first examine the demographic characteristics of those who are disconnected. Overall, 15 percent of youth in our sample are disconnected. Disconnected youth are more likely to be female and twice as likely as connected youth to be a parent (Table 1).⁠^[3] This appears to be driven by women with children: 32 percent of women with children are disconnected compared with 12 percent of women without children. Men with children have slightly higher rates of disconnection than do men without children (18 percent versus 13 percent) but much lower than the rate for women with children.

Disconnected youth have lower levels of education than do connected youth (Table 1): They are three times as likely to not complete high school and 50 percent more likely to have only a high school degree. They are far less likely to have attended college. Similar to other observational studies, we find that disconnected youth are more likely than connected youth to be Black and less likely to be White. Being Hispanic, Native American, Asian or Pacific Islander, or another race or ethnicity does not predict disconnection in this sample. Rates of disconnection are similar among those who were born in the United States and those who were not.⁠^[4]

Demographic Characteristic	Connected	Disconnected	Difference
Female	0.49	0.55	0.06*
Parent	0.15	0.31	0.16***
Less than high school education	0.11	0.33	0.21***
Completed high school	0.29	0.46	0.16***
Some college education or more	0.60	0.22	–0.38***
Hispanic	0.11	0.12	0.01
White	0.77	0.63	–0.14***
Black	0.14	0.28	0.14***
Native American	0.04	0.04	0.01
Asian or Pacific Islander	0.04	0.03	–0.01
Other race	0.06	0.06	0.00
Born in the United States	0.93	0.95	0.02

SOURCE: Features data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) (Harris and Udry, 2022), a representative sample of adolescents who were in grades 7 to 12 during the 1994–1995 school year, and have been followed through 2018.

NOTE: We use individual (person) weights to create statistics that are representative of the U.S. population. Asterisks denote statistically significant differences at the 10 percent (*), 5 percent (**), and 1 percent (***) levels, respectively.

What Early-Life Factors Precede Disconnection?

Beyond examining who is disconnected, we also use the panel nature of the data to examine early-life characteristics that precede—and might explain—disconnection.⁠^[5] Using the baseline wave that surveyed respondents when they were in school (ages 11 through 18), we construct baseline measures of explanatory variables that are thought to influence disconnection later in life (Table 2).⁠^[6]

Domain	Variables
Home environment	Parental education and household income
Academic performance and engagement	English language arts (ELA) and math test scores and a binary indicator for suspension at baseline
Mental health	Composite measure of mental health using the Center for Epidemiologic Studies Depression (CES-D) scale
Substance use	Indicators of substance initiation, including alcohol, cigarette, and drug use
Delinquency	Index of delinquency, based on a set of survey measures
Social support	Index of perceived social support from peers, family, and teachers, based on a set of survey measures

SOURCE: Features information from the National Longitudinal Study of Adolescent to Adult Health (Add Health) (Harris and Udry, 2022).

Home Environment

The data reveal several patterns (Figure 1). First, household and parental factors appear to be important for disconnection, consistent with the literature on parental factors and child socioeconomic outcomes (for examples of reviews, see Haveman and Wolfe, 1995; Duncan and Murnane, 2011). Students from below the top quartile of income and parental education are nearly twice as likely to be disconnected in young adulthood. Finally, females who had ever been pregnant at baseline (by ages 11 through 18) are significantly more likely to be disconnected later in life.

Category	Connected	Connected Confidence interval	Disconnected	Disconnected Confidence Interval
Female	0.49	(0.48–0.51)	0.55	(0.51–0.58)
HH income (Top Quartile)	0.29	(0.27–0.3)	0.16	(0.12–0.19)
Parent Edu (Top Quartile)	0.36	(0.35–0.38)	0.19	(0.16–0.22)
Ever pregnant	0.14	(0.11–0.17)	0.32	(0.25–0.38)
ELA Score (Top Quartile)	0.31	(0.3–0.33)	0.19	(0.16–0.22)
Math Score (Top Quartile)	0.29	(0.27–0.3)	0.18	(0.15–0.21)
Ever suspended	0.23	(0.22–0.25)	0.42	(0.39–0.46)
Depression	0.23	(0.22–0.24)	0.32	(0.28–0.35)
Substance Use: Alcohol	0.54	(0.52–0.55)	0.55	(0.51–0.59)
Substance Use: Cigarettes	0.55	(0.53–0.57)	0.61	(0.57–0.64)
Substance Use: Drugs	0.11	(0.1–0.11)	0.13	(0.12–0.15)
Delinquency	-0.04	(-0.07–-0.01)	0.11	(0.03–0.19)
Social support	0.06	(0.03–0.09)	-0.07	(-0.15–0.01)

SOURCE: Features data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) (Harris and Udry, 2022), a representative sample of adolescents who were in grades 7 to 12 during the 1994–1995 school year, and have been followed through 2018.

NOTE: "Ever pregnant" is the sample of females only. The bars show the mean of each outcome; diagonal lines show the 95-percent confidence intervals (CI) of each mean. We use individual (person) weights to create statistics that are representative of the U.S. population.

School: Academic Performance and Engagement

Students scoring within the top quartile in math and ELA test scores are less likely to become disconnected. This is consistent with a large body of research that suggests that academic performance predicts future labor market success (see Currie and Thomas, 2001; Heckman, Stixrud, and Urzua, 2006; Hanushek and Woessmann, 2008; among others). An indicator for whether a student was ever suspended at baseline ("ever suspended") is significantly correlated with disconnection. This could reflect both behavioral issues and disengagement with school.

Mental Health

Disconnection is correlated with a higher likelihood of being clinically depressed at baseline, as measured by the CES-D scale (Radloff, 1977).⁠^[7] This result masks substantial heterogeneity by gender. Young women who report symptoms of depression are significantly more likely to become disconnected. There are no significant differences in depression rates between disconnected versus connected males. As alternative indicators for mental health, we also examine self-reported suicidal ideation and suicide attempts. Both indicators are also higher for disconnected youth overall. Disconnected females have significantly higher suicidal ideation than connected females; disconnected males have higher (although not significantly different) suicide attempts than connected males. These results suggest that disconnected youth are more likely to have faced mental health challenges of some kind during high school.

Behavioral Risk Factors: Substance Use and Delinquency

A body of research links child conduct problems and substance use to poorer adult outcomes across education, employment, and health (Balsa, Giuliano, and French, 2011; Farrington, 2005; Fergusson, Horwood, and Ridder, 2005). We use a series of variables on substance use during high school (including alcohol, cigarettes, and illegal drugs) to examine how substance initiation correlates with disconnection. We also use a set of survey questions on delinquency to create an index for child conduct issues.⁠^[8] This includes self-reported answers to questions asking whether the child has painted graffiti, damaged property, lied to parents or guardians about activities, stolen items, taken part in violence, or sold drugs at baseline. Disconnected youth have significantly higher reported cigarette and drug use during high school (the time of the baseline survey) than connected youth. Disconnected youth are 10 percent more likely to have ever smoked cigarettes during high school and nearly 20 percent more likely to have ever used drugs.⁠^[9] There is no significant difference in alcohol use between the two groups. This pattern of results is similar across gender. Disconnected youth score significantly higher on the baseline delinquency index, suggesting that behavioral and conduct issues might be an important risk factor for disconnection. This is consistent with the higher likelihood of suspension discussed above.

Social Support

Finally, social support, whether from family, peers, or community, might be one mitigating factor that helps individuals navigate challenges during adolescence and ultimately reduce the risk of disconnection. Strong social ties have been linked to better adult outcomes, including educational attainment, labor market attachment, and psychological well-being (Cohen and Wills, 1985; Furstenberg and Hughes, 1995; Crosnoe and Elder, 2004). We constructed a social support index from a module measuring respondents' reported feelings of being supported and understood by parents, teachers, and friends. Consistent with the literature, higher social support during high school is correlated with a reduction in disconnection in early adulthood.

Suspension for Males and Early Pregnancy for Females Strongly Correlate with Disconnected Status, All Else Being Equal

Our results so far have examined differences in means across disconnected and connected populations. However, these relationships might be picking up spurious correlation with other factors. As our final analysis, we use multivariate regression analysis to examine all factors together, both overall and separately by gender (Figure 2).⁠^[10] This allows us to examine the relationship between each variable and disconnection, holding constant the influence of other factors. "Ever pregnant" and "ever suspended" remain highly positively correlated with disconnection. Examining the data separately by gender, "ever suspended" is only a significant correlate for males.⁠^[11] "Ever pregnant" at baseline is, by definition, only a significant correlate for females.

Suspension and early pregnancy might be mediating factors through which the other covariates affect disconnection. We examine correlates of suspension for males and correlates of early pregnancy for females.⁠^[12] For males, substance use (drugs and, to some extent, alcohol), high scores on the delinquency index, and being Black are positively correlated with suspension. Higher parental education, household income, ELA scores, and math scores are negatively correlated with suspension. These results are consistent with literature that suggests that school discipline, particularly suspension, is disproportionately applied to males and Black students (Okonofua and Eberhardt, 2015; Skiba et al., 2011). These practices are linked to worse academic and longer-term outcomes (Bacher-Hicks, Billings, and Deming, 2024; Perry and Morris, 2014). Alternative methods, such a restorative justice, have been shown to reduce suspensions, but they may need to be paired with academic supports to sustain achievement (Augustine et al., 2018; Gregory et al., 2016). Such measures may (or may not) serve to reduce eventual disconnection. For females, substance use (drugs), depression, and being Black are correlated with early pregnancy. These are purely correlational, but might shed light on risk factors that precede both early pregnancy and suspension and later disconnection.

95% confidence intervals in brackets

• Female = 0.0465 [-0.00783t to 0.101]
• Household Income (Upper Quartile) = -0.0199 [-0.0625 to 0.0228]
• Parent Education (Upper Quartile) = -0.0422 [-0.0902 to 0.00581]
• Parent Married = -0.00722 [-0.0853 to 0.0708]
• Ever Pregnant = 0.231 (p<0.01) [0.0862 to 0.375]
• ELA Score (Upper Quartile) = -0.0271 [0.0821 to 0.0279]
• Math Score (Upper Quartile) = -0.00765 [-0.0560 to 0.0407]
• Ever Suspended = 0.114 (p<0.001) [0.0655 to 0.163]
• Depression (CES-D) = -0.0110 [-0.0673 to 0.0453]
• Substance Use: Alcohol = -0.0388 [-0.0888 to 0.0112]
• Substance Use: Cigarettes = -0.0155 [-0.0718 to 0.0408]
• Substance Use: Drugs = 0.0762 [-0.0492 to 0.202]
• Delinquency Index = -0.0154 [-0.0435 to 0.0128]
• Social Support Index = -0.0244 (p<0.05) [-0.0434 to -0.00547]
• Hispanic = -0.0213 [-0.103 to 0.0608]
• Black = 0.0805 [-0.0230 to 0.184]
• Native American = -0.0388 [-0.142 to 0.0642]
• Asian/Pacific Islander = 0.00660 [-0.0800 to 0.0932]
• Other race = 0.0180 [-0.125 to 0.161]
• Born in the United States = 0.0244 [-0.0869 to 0.136]
• Constant = 0.127 (p<0.05) [0.00295 to 0.250]
• Observations = 1598

SOURCE: Features data from the National Longitudinal Study of Adolescent to Adult Health (Add Health) (Harris and Udry, 2022), a representative sample of adolescents who were in grades 7 to 12 during the 1994–1995 school year, and have been followed through 2018.

NOTE: This figure plots the coefficients from a linear regression of disconnection on the set of covariates. "White" is the excluded racial category. "Ever pregnant" is coded as 0 if the respondent was male. The upper and lower bounds show the 95-percent CI for each covariate. We use individual (person) weights to create statistics that are representative of the U.S. population.

The Path Forward

We provide preliminary evidence on the predictive factors that are associated with disconnection, including some of the first analyses using Add Health data to explore these relationships. Even in the years before disconnection emerges, youth who eventually become disconnected are different from their connected peers across several dimensions. They are more likely to report symptoms of depression, experience an early pregnancy, use substances, engage in delinquent activities, have weaker social support structures, and be suspended from school. Early pregnancy and school suspension, in particular, may serve as pathways through which other risk factors translate into later disconnection.

These estimates are descriptive rather than causal, and unobserved factors correlated with our explanatory variables may drive both early-life risks and eventual disconnection. However, this analysis highlights a set of risk factors that are likely to shape longer-run outcomes and provides a foundation for future research.

This work points to several implications for policymakers. First, early identification and prevention of risk factors associated with disconnection is critical. Eventual disconnection is correlated with academic performance, school suspension, early pregnancy, and substance use; this suggests that risks emerge early on. Preventative policies that reduce these risk factors and improve academic performance may reduce the likelihood of later disconnection.

Second, the links between suspension and disconnection suggest that restorative justice programs may not only be effective in reducing suspensions but also in reducing later disconnections. Restorative approaches emphasize repairing harm, relationship-building, and prevention, rather than excluding children from school. Evidence shows that restorative justice practices are effective in reducing suspension rates (Augustine et al., 2018; Gregory et al., 2016). This in turn may reduce the likelihood of youth disconnection by keeping students engaged in school.

Finally, evidence gaps remain. Evaluations of programmatic and policy interventions targeting these explanatory variables might help clarify mechanisms that lead to disconnection. For instance, evaluations of school-based programs that reduce suspensions or evaluations of interventions that delay early pregnancy could shed light on whether addressing these risk factors reduces disconnection. Importantly, future research should explicitly consider disconnection itself as a key outcome, assessing not only whether interventions affect intermediate risk factors but also whether they ultimately reduce the likelihood of youth becoming disconnected.

Appendix. Additional Figures

Acknowledgments

We are grateful for the contributions and support of our colleagues Andrew Hoehn, Heather Schwartz, and Jennifer Kondo. We thank Ben Master and Christine Mulhern for their careful reviews. We are grateful to Monette Velasco, Libby Sweeney, Stephanie Lonsinger, and Mirka Vuollo for their assistance with editing and the publication process.

Notes

1. Employment is measured as working for pay for at least ten hours per week. Therefore, anyone who is working less than ten hours per week or not working for pay (and otherwise not in school) is counted as disconnected by this measure.

   We examine the period of disconnection during Wave 3; at this point, respondents were 18 to 28 years old, with a median age of 22. Waves 1 and 2 were collected using a sample of adolescents who were in school at the time of the survey, thus disconnection during these waves is, by design, close to 0 percent. Respondents in Wave 4 were aged 25 through 33, and thus outside the age range of disconnected youth, although we can eventually use this sample to examine longer run outcomes of disconnection.

   For our analysis, we use the Add Health public-use sample, which is one-third of the size of the full restricted-use sample. Future planned analysis will use the full sample. Return to content ⤴

2. This means that our sample differs from the overall population of disconnected youth and those who were already disconnected at the time of high school; examining out-of-school students might reveal different characteristics that correlate with disconnection. Return to content ⤴
3. Seventeen percent of females are disconnected whereas 14 percent of males are disconnected. Return to content ⤴
4. These results are broadly similar to those found in Wenger and Bonds (2025). The differences likely stem from the different periods (Wenger and Bonds use data from 2019–2023) and perhaps from the different sampling frames (Add Health forms a sample from connected youth in high school). Return to content ⤴
5. We note that these results are exploratory and not a comprehensive set of all possible covariates. Future planned work will use the Add Health restricted data, which have the full sample of respondents and a richer set of data, allowing us to calculate disconnection at a more granular level, examine geographic variation, and examine a more comprehensive set of covariates. Return to content ⤴
6. We also examine a binary indicator for physical disability. Physical disability is correlated with disconnection but the sample of those disabled is extremely small (less than 1 percent) and thus our estimate is very imprecise. In future analyses, we will re-estimate this using the restricted-use sample which has three times the number of observations. Return to content ⤴
7. The CES-D scale typically has 20 questions, with a score of 16 or above indicating clinical depression (Radloff, 1977). The version in the Add Health has 19 questions, and so we code 15 or above as clinical depression. Results are robust to an alternative coding of 16 and above. Return to content ⤴
8. We create a means effect index of all 15 questions in the delinquency module to create our delinquency index. Return to content ⤴
9. Illegal drug use includes significantly higher marijuana, cocaine, and "other" drug use; there was no significant difference in inhalant use. Marijuana use was about 30 percent higher in the disconnected sample (from 25 to 33 percent) and cocaine use was about 60 percent higher (from 3 to 5 percent). Both of these differences are statistically significant. "Other" drug use (other drugs not included in marijuana, cocaine, and inhalants) was 12 percent higher in the disconnected sample (from 8 to 9 percent) but not statistically significant. Return to content ⤴
10. We estimated both a linear probability model and a logistic regression. Results are similar across models and so we show results from the linear model here. Return to content ⤴
11. See Figure A.1. in the appendix for these results. Return to content ⤴
12. See Figures A.2. and A.3. in the appendix for these results. Return to content ⤴

References

• Augustine, Catherine H., John Engberg, Geoffrey E. Grimm, Emma Lee, Elaine Lin Wang, Karen Christianson, and Andrea A. Joseph, Can Restorative Practices Improve School Climate and Curb Suspensions: An Evaluation of the Impact of Restorative Practices in a Mid-Sized Urban School District, RAND Corporation, RR-2840-DOJ, 2018. As of September 30, 2025: https://www.rand.org/pubs/research_reports/RR2840.html
• Bacher-Hicks, Andrew, Stephen B. Billings, and David J. Deming, "The School to Prison Pipeline: Long-Run Impacts of School Suspensions on Adult Crime," American Economic Journal: Economic Policy, Vol. 16, No. 4, November 2024.
• Balsa, Ana I., Laura M. Giuliano, and Michael T. French, "The Effects of Alcohol Use on Academic Achievement in High School," Economics of Education Review, Vol. 30, No. 1, 2011.
• Belfield, Clive R., Henry M. Levin, and Rachel Rosen, The Economic Value of Opportunity Youth, Corporation for National and Community Service, 2012.
• Cohen, Sheldon, and Thomas A. Wills, "Stress, Social Support, and the Buffering Hypothesis," Psychological Bulletin, Vol. 98, No. 2, 1985.
• Crosnoe, Robert, and Glen H. Elder, Jr., "Family Dynamics, Supportive Relationships, and Educational Resilience During Adolescence," Journal of Family Issues, Vol. 25, No. 5, 2004.
• Currie, Janet, and Duncan Thomas, "Early Test Scores, School Quality and SES: Longrun Effects on Wage and Employment Outcomes," in Solomon Polachek, ed., Worker Wellbeing in a Changing Labor Market, Emerald Group Publishing Limited, 2001.
• Duncan, Greg J., and Richard J. Murnane, eds., Whither Opportunity? Rising Inequality, Schools, and Children's Life Chances, Russell Sage Foundation, 2011.
• Farrington, David P., "Childhood Origins of Antisocial Behavior," Clinical Psychology & Psychotherapy, Vol. 12, No. 3, 2005.
• Fergusson, David M., L. John Horwood, and Elizabeth M. Ridder, "Show Me the Child at Seven: The Consequences of Conduct Problems in Childhood for Psychosocial Functioning in Adulthood," Journal of Child Psychology and Psychiatry, Vol. 46, No. 8, 2005.
• Furstenberg, Frank F., Jr., and Mary Elizabeth Hughes, "Social Capital and Successful Development Among At-Risk Youth," Journal of Marriage and the Family, Vol. 57, No. 3, 1995.
• Gregory, Anne, Kathleen Clawson, Alycia Davis, and Jennifer Gerewitz, "The Promise of Restorative Practices to Transform Teacher-Student Relationships and Achieve Equity in School Discipline," Journal of Educational and Psychological Consultation, Vol. 26, No. 4, 2016.
• Hair, Elizabeth C., Kristin A. Moore, Thomson J. Ling, Cameron McPhee-Baker, and Brett V. Brown, "Youth Who Are 'Disconnected' and Those Who Then Reconnect: Assessing the Influence of Family, Programs, Peers and Communities," Child Trends, Vol. 37, July 2009.
• Hanushek, Eric A., and Ludger Woessmann, "The Role of Cognitive Skills in Economic Development," Journal of Economic Literature, Vol. 46, No. 3, 2008.
• Harris, Kathleen Mullan, and J. Richard Udry, "National Longitudinal Study of Adolescent to Adult Health (Add Health), 1994–2018 [Public Use]," dataset, version 25, Carolina Population Center, University of North Carolina–Chapel Hill, Inter-university Consortium for Political and Social Research, August 9, 2022. As of October 7, 2025: https://doi.org/10.3886/ICPSR21600.v25
• Haveman, Robert, and Barbara Wolfe, "The Determinants of Children's Attainments: A Review of Methods and Findings," Journal of Economic Literature, Vol. 33, No. 4, 1995.
• Heckman, James J., Jora Stixrud, and Sergio Urzua, "The Effects of Cognitive and Noncognitive Abilities on Labor Market Outcomes and Social Behavior," Journal of Labor Economics, Vol. 24, No. 3, 2006.
• Lewis, Kristen, A Decade Undone: 2021 Update, Measure of America, July 29, 2021.
• MaCurdy, Thomas, Bryan Keating, Sriniketh Suryasesha Nagavarapu, and David Glick, "Reprint of: Profiling the Plight of Disconnected Youth in America," Journal of Econometrics, Vol. 243, No.1–2, July 2024.
• Okonofua, Jason A., and Jennifer L. Eberhardt, "Two Strikes: Race and the Disciplining of Young Students," Psychological Science, Vol. 26, No. 5, 2015.
• Perry, Brea L., and Edward W. Morris, "Suspending Progress: Collateral Consequences of Exclusionary Punishment in Public Schools," American Sociological Review, Vol. 79, No. 6, 2014.
• Radloff, Lenore Sawyer, "The CES-D Scale: A Self-Report Depression Scale for Research in the General Population," Applied Psychological Measurement, Vol. 1, No. 3, 1997.
• Skiba, Russell J., Robert H. Horner, Choong-Geun Chung, M. Karega Rausch, Seth L. May, and Tary Tobin, "Race is Not Neutral: A National Investigation of African American and Latino Disproportionality in School Discipline," School Psychology Review, Vol. 40, No. 1, 2011.
• Wenger, Jennie W., and Stephanie Bonds, Understanding Disconnection Among American Youth, RAND Corporation, PE-A4207-1, October 2025. As of October 1, 2025: https://www.rand.org/pubs/perspectives/PEA4207-1.html

Related Content

• 

  Expert Insights

  Understanding Disconnection Among American Youth

  Oct 1, 2025

Topics

• Community Health and Well-Being
• Employment and Unemployment
• Life Skills
• School-to-Work Transitions
• United States
• Young Adults