Copyright 2008 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2008
To examine the long-term effects of a universal intervention in elementary schools in promoting positive functioning in school, work, and community, and preventing mental health problems, risky sexual behavior, substance misuse, and crime at ages 24 and 27 years.
Nonrandomized controlled trial.
Fifteen public elementary schools serving diverse neighborhoods including high-crime neighborhoods in Seattle, Washington.
Sex-balanced and multiracial/multiethnic sample of 598 participants at ages 24 and 27 years (93% of the original sample in these conditions).
Teacher training in classroom instruction and management, child social and emotional skill development, and parent workshops.
Main Outcome Measures
Self-reports of functioning in school, work, and community and of mental health, sexual behavior, substance use, and crime, and court records.
A significant multivariate intervention effect across all 16 primary outcome indices was found. Specific effects included significantly better educational and economic attainment, mental health, and sexual health by age 27 years (all P < .05). Hypothesized effects on substance use and crime were not found at ages 24 or 27 years.
A universal intervention for urban elementary schoolchildren, which focused on classroom management and instruction, children's social competence, and parenting practices, positively affected mental health, sexual health, and educational and economic achievement 15 years after the intervention ended.
Poverty, unemployment, and neighborhood disorganization are persistent problems in US cities.1,2 Crime, drug use, teen pregnancy, mental health problems, and high rates of school dropout plague many urban children and families.3- 7 Public schools, available to all children in the United States beginning at age 5 or 6 years, are a potentially powerful setting for preventive intervention. We examined the effects of a 3-component preventive intervention provided in public schools during the elementary grades on outcomes at ages 24 and 27 years, 15 years after the intervention ended. The objective of the intervention was to improve the skills of teachers, parents, and children to increase positive functioning in school and decrease problems related to mental health, risky sexual behavior, substance use, and criminal behavior.
The mid-20s are important years for the adoption of adult roles. Engagement in education or occupational roles is an important predictor of future adult functioning.8- 10 Civic engagement is also likely to increase during this period.11,12 However, the mid-20s are also years of relatively high vulnerability to mental health problems,13,14 sexual risk-taking,15,16 and continued risk of substance use and crime.17- 22
Little is known about the long-term effects of universal intervention in public elementary schools on these outcomes. Kellam et al23 observed students from schools serving predominantly African American children from poor to lower middle-class families who had been exposed to a classroom-based behavior management program in the first and second grades. By ages 19 to 21, male subjects, particularly those who had demonstrated more aggressive or disruptive behavior in the first grade, reported significantly reduced rates of regular cigarette smoking, fewer drug and alcohol abuse or dependence disorders, and less antisocial personality disorder.23 We are aware of no other studies of universal interventions in the elementary grades that have investigated long-term effects on indicators of adult functioning.
The Seattle Social Development Project (SSDP) intervention was guided theoretically by the social development model.24,25 We sought to identify and develop methods of management and instruction that could be used by public school teachers and adult caretakers to set children on a positive developmental course by promoting opportunities for children's active involvement in classroom and family, developing children's skills for participation, and encouraging reinforcement from parents and teachers for children's effort and accomplishment. Two intervention conditions were examined: a full-intervention condition implemented throughout grades 1 through 6 and a late-intervention condition implemented only in grades 5 and 6.26
Studies of the SSDP intervention have found significant effects in childhood and adolescence across outcomes.27- 32 By age 21 years, the full-intervention group, compared with the control group, exhibited significantly better outcomes for education, employment, and mental health, as well as reduced crime, sexual risk behavior and disease, and early pregnancy.33,34 Some effects were found to be moderated by sex, race/ethnicity, or childhood poverty.30,31,33,34 Herein we examined the effects of the SSDP intervention at ages 24 and 27 years, 12 to 15 years after the intervention ended. The sex-balanced and racially/ethnically diverse sample enabled investigation of possible moderators of intervention effects.
The Figure shows the overall design of the study. Beginning in fall 1981, the intervention was initiated among first-grade students in classrooms randomly assigned to condition in 8 public schools serving high-crime areas in Seattle, Washington. Three hundred seventy-seven students who remained in or entered the 8 schools during grades 1 through 3 were observed prospectively to fifth grade. The study was then expanded to include 676 fifth-grade students in 10 additional schools, and all parents were asked for consent for their child to participate in the longitudinal follow-up study. Of the population of 1053 fifth-grade students in the 18 schools, parents of 808 children (77%) consented.
Schools were assigned nonrandomly to the intervention or control condition in fall 1985. Thereafter, all fifth-grade students in each school received the intervention according to their school's intervention assignment. This resulted in a nonrandomized controlled trial with 4 conditions. The full-intervention group consisted of those who received at least 1 semester of intervention in grades 1 through 4 and at least 1 semester of intervention in grades 5 and 6, with a mean of 4.13 years of intervention exposure. The late-intervention group consisted of those who received the intervention during grades 5 and 6 only, with a mean of 1.65 years of exposure. The control group received no intervention. A fourth group was offered parent training only during grades 5 and 6 and is not discussed in this article. Twenty-four participants could not be classified into any of these groups because they left participating schools before attending for at least 1 semester. After excluding these 24 participants, all analyses were based on intervention assignment. All phases of the study were approved by the Human Subjects Review Committee at the University of Washington, Seattle. Participants were informed about the nature of the interviews and provided consent before participation in the study at ages 24 and 27 years.
Retention for analysis of the full-intervention, late-intervention, and control groups averaged 93% at both ages 24 and 27 years (Figure). There were equal numbers of female and male participants at each age. Racial/ethnic identification was 46% European American, 26% African American, 22% Asian American, and 6% Native American. As children, 56% of participants were eligible for the federal school lunch or breakfast program at some point in the fifth, sixth, or seventh grade, indicating low income status.
At both ages 24 and 27 years, the overall distribution of participants in the intervention conditions did not differ significantly for those lost to attrition vs the analysis sample (χ22 = 2.16, P = .34 at age 24 years; and χ22 = 1.41, P = .50 at age 27 years); in addition, among those retained in the analysis sample, at both ages 24 and 27 years, the distribution of participants in the intervention conditions did not differ for sex (χ22 = 0.38, P = .83; and χ22 = 0.59, P = .74, respectively), race/ethnicity (white vs other: χ22 = 0.13, P = .94; and χ22 = 0.02, P = .99, respectively), or childhood poverty (χ22 = 0.23, P = .89; and χ22 = 0.33, P = .85, respectively).
Given the requirement that students who received the full intervention attended project schools at some point in grades 1 through 4 and in grades 5 and 6, whereas some students in the control group were added to the study at grade 5, it is important to rule out differences in residential stability, a potential threat to internal validity. Analyses comparing the full-intervention and control groups found no significant differences in mean number of years living in Seattle by grade 6 (F = 0.61, P = .44 at age 24 years; and F = 1.83, P = .18 at age 27 years), mean number of residences lived in from age 5 to 14 years (F = 1.57, P = .21; and F = 1.56, P = .21, respectively), percentage of single-parent homes during grade 5 (χ2 = 0.11, P = .74; and χ2 = 0.02, P = .89, respectively), or living in a disorganized neighborhood at age 16 years (eg, high crime or rundown housing: χ2 = 0.47, P = .49; and χ2 = 0.13, P = .72, respectively). Differential school or teacher receptivity to intervention is also an unlikely threat to internal validity. Teachers in 6 of 8 participating schools during grades 1 through 4 were randomly assigned to either intervention or control classrooms. At fifth grade, newly eligible schools were matched demographically to early experimental schools, and each agreed to serve as a control or late-intervention school depending on assignment. In addition, during the course of the intervention, the Seattle school district used mandatory busing to achieve racial/ethnic equality in schools, which substantially reduced the risk that outcomes observed reflected contextual or neighborhood differences, school demographic differences, or parent school-selection effects in the populations attending different schools.
An exception to the pattern of condition equivalence was the percentage of surveyed participants who reported at age 24 years that their mothers were aged 19 years or younger when they were born. Nine percent of the full-intervention condition compared with 21% of the control condition reported that their mothers were teenagers when they were born (χ2 = 8.56, P < .01). Having a teenaged mother was included as a covariate in all outcome analyses in this study.
The intervention is described elsewhere.32- 35 Each year during grades 1 through 6, teachers in the intervention conditions received 5 days of in-service training in instructional methods.27,36- 40 In addition, first-grade teachers received instruction in the use of a cognitive and social skills training curriculum,41,42 and during grade 6, a study consultant provided students with training in refusal skills.43 When children were in grades 1 through 3, parents were offered a 7-session curriculum in child behavior management skills35 and a 4-session curriculum in skills for supporting their children's academic development.44 During grades 5 and 6, parents of participants in intervention conditions were offered a 5-session curriculum designed to strengthen skills to reduce their children's risk of problem behaviors.45 Forty-three percent of parents of children eligible for the full intervention attended at least 1 parenting class during grades 1 through 3,35 and 29% of parents of children in eligible intervention conditions attended at least 1 class during grades 5 or 6, which indicates that the parent-training component had less reach than the teacher training and child social skills training components.
Primary outcome indices of success in each life domain were analyzed. The indices provide a limited set of comparisons for multivariate statistical tests of intervention effects. Analyses of primary outcome indices were followed up, where warranted, by analyses of specific behaviors, attitudes, and events to provide descriptive detail to the findings and to enable comparison with previous reports.33,34 Measures were participant self-reports of events in the last year at ages 24 and 27 years unless otherwise noted, and court records from age 10 to 24 years. Extreme values for open-ended numeric responses were limited to an appropriate maximum to limit the effects of outliers.
For school and work functioning, a median socioeconomic status (SES) attainment index was assessed by creating a dichotomous measure of those scoring at or above the median in completed education (attaining at least a high school diploma by age 24 or 27 years) or household income (at least $44 000 at age 24 years and at least $45 000 at age 27 years).40 Specific SES measures included cumulative education completed46 and income (divided by 1000 for analysis). Other measures included the degree to which students were integrated at school (eg, time in class and participation, coded 0 [poorly integrated] to 4 [well integrated]; mean reliability coefficient α at ages 24 and 27 years, α = 0.37)33 and the degree of responsibility on the job for those who were employed (2 items coded 0 [low responsibility] to 4 [high responsibility]; mean r = .36).33 (For purposes of comparison, measures at age 21 years were replicated as closely as possible. As a result, some reliability coefficients for measures corresponding to those examined at age 21 years are low.) Constructive engagement summed the average number of hours per week engaged in school or work (divided by 10 for analysis),33,47 and constructive self-efficacy concerned perceived opportunities for attaining a good education and a good job (coded 0 [low engagement] to 4 [high engagement]; α = .78).33 Two additional descriptive measures, for which we did not expect intervention effects, were student status in the last year and employment status in the last month, ranging from nonstudent or unemployed (coded 0) to full-time student or employed (coded 4).33 To assess community involvement, a civic engagement index summed the mean number of hours per month of involvement in community groups and volunteer activities.47
Mental health problems were assessed with the Diagnostic Interview Schedule46,48- 52 to measure Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition) (DSM-IV) criteria.53 A disorder criterion index (coded 0-37; α = .95) summed across the total number of DSM-IV criteria met in the last year for a generalized anxiety disorder (GAD) criterion count (coded 0-6; α = .88), a social phobia criterion count (coded 0-5; α = .92), a posttraumatic stress disorder criterion count (coded 0-17; α = .98), and a major depressive episode criterion count (coded 0-9; α = .98). Specific criterion count measures for each of these disorders were also analyzed, including a separate measure of suicidal thoughts (coded 0-3; α = .67).33 In addition, a dichotomous disorder diagnosis index was computed to indicate those meeting criteria for DSM-IV diagnosis for any of the measured disorders (GAD, social phobia, posttraumatic stress disorder, or major depressive episode).33
The lifetime sexually transmitted disease (STD) index was a dichotomous measure based on reports of having ever been diagnosed as having an STD.34 Specific measures of sexual risk behavior included the number of lifetime sex partners and, among those not in an exclusive relationship, the number of past-year sex partners and condom use in the last year (coded as percentage of time).34 We also report descriptive measures of having ever been or gotten someone pregnant and ever having or fathering a baby, for which we did not expect intervention effects. Lifetime measures were constructed accounting for affirmative responses in previous-year surveys.
A substance abuse and dependence criterion index was computed as the sum of DSM-IV criteria met for alcohol and illicit drug abuse and dependence disorders (coded 0-22; α = .86), and a substance abuse and dependence diagnosis index indicated those meeting criteria for an abuse or dependence diagnosis.46 Specific behaviors related to problem substance use included high variety of substance use (identifying those above the 90th percentile in number of different substances used in the last year) and the extent of substance use interference with life (coded 0 for no use to 4 for “very much”).33 Any past-year substance use (tobacco, alcohol, or illicit drugs) is reported for descriptive purposes.33
A past-year crime index was computed from self-reports of criminal acts other than driving violations or illicit drug use.33 Specific crime measures included high variety of crime, identifying those above the 90th percentile in number of different types of crimes committed, and measures of having sold drugs and having been arrested. In addition, official state and federal crime files were matched to SSDP participants (including survey nonrespondents) to assess the presence of a past-year court charge or a lifetime court charge for any noncriminal, misdemeanor, or felony charge through age 24 years.33
Because of the study design, the unit of intervention assignment consisted of the series of classrooms to which some students were assigned in grades 1 through 4 and the condition assignments of schools attended by all participants in grades 5 and 6. Of the 643 participants assigned to the control, late-intervention, and full-intervention conditions, more than 169 different classroom or school sequences were identified, consistent with the unit of intervention assignment. On average, only 3.80 participants experienced the same units of intervention within conditions. For this reason and to be consistent with previous reports, analyses were conducted at the individual level.
A multivariate analysis of covariance was conducted to assess overall intervention effects across multiple dependent variables, controlling for teenaged mother at birth. All 16 primary outcome indices (8 each at ages 24 and 27 years) were included in this analysis. Results showed a significant overall difference between the full-intervention and control groups with listwise deletion (Wilks Λ, F[16, 276], 1.98; P = .01). This multivariate analysis of covariance was replicated across 5 data sets for which imputation procedures were used to account for missing data on some outcomes.54 Each analysis confirmed an overall group difference (Wilks Λ, F[16, 319], 1.84-1.94; P = .03-.02). These results indicate a significant multivariate effect and provide overall control for type I error rate in the significant univariate findings.55 Multivariate analyses of covariance comparing the late-intervention group with the control group across the 16 outcomes were not significant (Wilks Λ, F[16, 423], 0.69-0.75; P = .81-.74).
Participants in the full-intervention group were significantly more likely than those in the control group to be at or above the median in SES (educational attainment or household income) by age 27 years (93% vs 84% in the control group) (Table). The differences in SES attainment were similar at age 24 years, though not significant. Specific comparisons in this domain found that those in the full-intervention groups were marginally more likely to have continued their education beyond high school, with 34% having completed an associate degree by age 27 years compared with 22% in the control group (P < .06). Nonsignificant trends in completion of a bachelor's degree and in household and earned income also favored the full-intervention group.
The civic engagement index indicated marginally more community involvement and volunteerism in the full-intervention group at ages 24 and 27 years compared with the control group. Both intervention groups reported significantly more participation in community groups at age 24 years compared with the control group.
Across outcomes related to school, work, and community at ages 24 and 27 years for which we had directional hypotheses, 27 of 28 outcomes analyzed showed directional patterns between the full-intervention and control conditions consistent with prediction, and both primary outcome indices showed positive full-intervention effects significant beyond P < .08 by age 27 years. Twenty-two of these 28 outcomes demonstrated a pattern consistent with a dose effect, with the late-treatment group falling between the full-treatment and control groups.
Analyses found significantly fewer symptoms of mental health disorders on the disorder criterion index in the full-treatment condition compared with the control group at both ages 24 and 27 years. Analyses also found significantly lower prevalence of those meeting criteria for at least 1 of 4 DSM-IV diagnoses on the disorder diagnosis index in the full-treatment group compared with the control group at age 27 years and a nonsignificant trend toward lower prevalence at age 24 years. Measures of specific disorder criterion counts indicated at least marginally significant (P < .10) reductions in the number of GAD, social phobia, and posttraumatic stress disorder criteria met at age 24 years and of GAD, posttraumatic stress disorder, and major depressive episode criteria met, as well as significantly fewer suicidal thoughts, at age 27 years. Across all 22 mental health outcomes examined, reported problems were lower in magnitude in the full-intervention group compared with the control group. Participants in the late-intervention group reported levels between or equal to those of the other groups for all but 3 of the 22 mental health outcomes, consistent with a dose effect.
The lifetime STD index showed significantly lower prevalence of having ever been diagnosed with an STD at ages 24 and 27 years for the full-intervention group compared with the control group. No intervention effects on specific sexual behaviors were found.
No intervention effects on either of the substance abuse and dependence indices were found in young adulthood, and no significant effects were observed for the specific measures of high variety of substance use or substance use interference with life. A marginally significant trend (P < .09), counter to hypotheses, for the last-year crime index at age 27 years indicated a higher prevalence of having committed a crime among participants in the full-intervention group compared with the control group. This difference stemmed primarily from nonsignificant but higher rates in the full-intervention group of minor theft (having taken anything worth less than $50; 9% vs 4% in the control group) and drug selling (8% vs 3%, respectively) at age 27 years. This was the only finding counter to hypotheses that approached significance across all of the outcomes examined in the Table. No significant intervention effects were found for any specific measures related to crime.
Adding interaction terms to the regression models revealed 2 significant effects of the full intervention specific to male subjects. Male subjects in the full-treatment group were significantly more likely to meet median education or income criteria at age 24 years as assessed using the SES attainment index compared with their counterparts in the control group (97% [SD, 0.17] vs 85% [SD, 0.36], respectively; P = .03). In addition, male subjects in the full-treatment group were significantly more constructively engaged at age 24 years compared with male subjects in the control group (mean [SD] 29.9 [13.9] h/wk of engagement in school or work vs 24.4 [16.4] h/wk, respectively; P = .02). Neither of these outcomes showed significant intervention differences for female subjects.
Two significant intervention effects were also found specific to participants from a background of childhood poverty. Those in the full-treatment group from a background of childhood poverty reported a mean (SD) of almost 9.5 (13.2) hours per month of involvement in community groups at age 27 years compared with 4.9 (9.2) hours per month for those in the control group (P = .004). Participants from childhood poverty in the full-intervention group were also significantly less likely to meet sufficient criteria for a diagnosis of GAD at age 24 years compared with those in the control group (3% [SD, 0.16] vs 14% [SD, 0.35], respectively; P = .02).
After controlling for childhood poverty, race/ethnicity also significantly interacted with the intervention for some outcomes. Responsibility on the job at age 24 years was significantly higher among white participants in the full-treatment group compared with whites in the control group (mean = 2.78 [SD, 1.10] and mean = 2.34 [SD, 1.06], respectively; P = .02). African Americans in the full-intervention condition compared with their control counterparts reported significantly higher household income at age 27 years (mean = $55 594 [SD, $45 966] and mean = $35 288 [SD = $35 410], respectively; P = .046), fewer GAD criteria met at age 27 years (mean = 0.94 [SD, 1.39] and mean = 2.51 [SD, 1.99], respectively; P < .001), fewer past-year sex partners at age 27 years (mean = 1.39 [SD, 1.65] and mean = 2.89 [SD, 2.78], respectively; P = .03), and reduced prevalence of lifetime STD diagnosis at age 24 years (12% [SD, 0.33] in the full-intervention condition and 55% [SD, 0.50] in the control condition; P = .001) and at age 27 years (16% [SD, 0.37] in the full-intervention condition and 61% [SD, 0.49] in the control condition; P = .001). These results suggest that the significant main effects of the full intervention on STD diagnoses at ages 24 and 27 years were primarily because of the reductions in STD diagnoses among African Americans in the full-intervention group.
The SSDP intervention in the elementary grades showed a significant overall effect of the full intervention in a multivariate analysis of 16 outcome indices at ages 24 and 27 years, 12 and 15 years after the intervention ended. Summary indices revealed significantly better SES, mental health, and sexual health by age 27 years in those assigned to the SSDP full-intervention condition compared with those in the control group. In contrast, effects of the intervention on crime and substance use were not found at ages 24 and 27 years.
In line with previous analyses,32,33 the rank of means and prevalence rates for the late-intervention group compared with the full-intervention and control groups suggests a dose effect for the domains most influenced by the intervention. That is, although differences were not significant, the late-intervention group reported better outcomes than the control group across the preponderance of measures for which there were directional hypotheses related to school, work, mental health, or community, but worse outcomes compared with the full-intervention group across these same measures.
Our findings indicate that a theory-based intervention that improved parenting practices, children's social competence, and classroom management and instruction during the elementary grades influenced some, though not all, indices of adult functioning in individuals in their mid- to late 20s. The elementary grade intervention was associated with greater accomplishment and engagement in school, work, and community and fewer mental health problems by ages 24 and 27 years. Effects of the SSDP intervention on mental health outcomes are particularly noteworthy at this stage of young adulthood, given the potentially debilitating consequences of depression and anxiety.56- 59
Findings should be considered in the context of the design and analysis approach, which was based on condition assignment rather than attendance or implementation, using controlled 2-tailed statistical tests and an omnibus test of multiple outcomes. Strengths of this study include the ability to detect relatively small effect sizes, comprehensive assessment capturing multiple outcome domains, delivery of the intervention package universally to all students in intervention classrooms, and the ability to investigate possible moderating effects of sex, race/ethnicity, and childhood poverty. As in previous reports,33 there was little evidence of serious threats to internal validity. In addition, procedures for assigning classrooms (in the early experiment) and schools to conditions likely guarded against possible effects of differential school or teacher receptivity to intervention, and mandatory busing provided further protection against confounding neighborhood or school demographic differences or parent school-selection effects. An exception to the general pattern of group equivalence at the start of the longitudinal study was a significantly higher percentage of participants in the control group who reported having had a teenaged mother at birth. This difference was controlled for statistically throughout all analyses.
Limitations should be noted. The study was quasi-experimental and geographically limited. It relied heavily on self-reported data from study participants. Effects of the intervention on school, work, and mental health were observed, but no significant effects on substance use or crime were observed at these ages.
Intervention effects reported herein indicate that universal intervention during the elementary grades to improve the management and instructional skills of urban public elementary schoolteachers, strengthen parenting practices in multiracial/multiethnic urban families, and ensure that children have the skills to participate in the social and academic life of elementary school can positively affect attainment, functioning, and mental health in young adulthood.
Correspondence: J. David Hawkins, PhD, Social Development Research Group, School of Social Work, University of Washington, 9725 Third Ave NE, Ste 401, Seattle, WA 98115 (firstname.lastname@example.org).
Accepted for Publication: April 23, 2008.
Author Contributions: Dr Hawkins had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Hawkins, Kosterman, Catalano, and Hill. Acquisition of data: Hawkins and Hill. Analysis and interpretation of data: Kosterman, Catalano, Hill, and Abbott. Drafting of the manuscript: Kosterman. Critical revision of the manuscript for important intellectual content: Hawkins, Kosterman, Catalano, Hill, and Abbott. Statistical analysis: Kosterman and Abbott. Obtained funding: Hawkins, Kosterman, Catalano, and Hill. Administrative, technical, and material support: Hill. Study supervision: Hawkins.
Financial Disclosure: Dr Catalano receives reimbursement for being a member of the board of Channing Bete Co, distributor of Supporting School Success and Guiding Good Choices.
Funding/Support: This study was supported by grants 1R01DA09679-11 and 9R01DA021426-08 from the National Institute on Drug Abuse, R24MH56587-06 from the National Institute of Mental Health, and 21548 from the Robert Wood Johnson Foundation.
Additional Information:Supporting School Success and Guiding Good Choices were tested in the study that produced the data set used in this article.
Hawkins JD, Kosterman R, Catalano RF, Hill KG, Abbott RD. Effects of Social Development Intervention in Childhood 15 Years Later. Arch Pediatr Adolesc Med. 2008;162(12):1133-1141. doi:10.1001/archpedi.162.12.1133