Strong African American Families Program

Study 1

Study 1 examined the first cohort of a larger study that examined two cohorts.

Summary

Brody et al. (2004) and five other reports used a cluster randomized trial to examine 332 African American mothers and their 11-year-old children residing in eight rural county units in Georgia. The sample consisted of the first cohort of an ongoing project (see Study 2). The eight county units were randomly assigned to either the control or intervention conditions. Families in both groups completed a posttest (approximately three months after intervention end) and a 29-month follow-up with measures of parenting, youth protective factors, alcohol use, and sexual risk behavior.

Brody et al. (2004) and five other reports found that relative to the control group, the intervention group had significantly:

• more regulated, communicative parenting at posttest
• higher levels of youth protective factors at posttest
• lower levels of youth risk behavior at posttest
• fewer new alcohol users and slower growth in alcohol use at long-term follow-up

Evaluation Methodology

Design:

Recruitment: The study recruited African American mothers and their 11-year-old children from nine rural, high-poverty counties in Georgia, but two small contiguous counties were combined into one county unit. Schools in these eight units provided lists of 11-year-old students. The sample consisted of the first cohort of an ongoing project (see Study 2). A total of 521 families were selected randomly and 332 (64%) families were eligible and agreed to participate (note, however, that Brody, Murry, Kogan et al. [2006] list slightly different numbers in the CONSORT diagram).

Assignment: In a cluster randomized design, the eight county units were randomly assigned to the intervention condition (n = four schools, 182 families) or the control condition (n = four schools, 150 families). While the intervention families participated in the seven-week prevention program, the control families received three leaflets on adolescent development, stress management, and the importance of exercise. Consent followed random assignment of the counties, but Brody et al. (2004, p. 905) stated that "Families were initially recruited into the overall project but were informed that they could be assigned to an intervention condition." The statement may imply that participants were unaware of assignment when they consented and completed the pretest.

Assessments/Attrition: Assessments occurred at posttest (about seven months after pretest or three months after the program end) and at 29 months after pretest (about 24-25 months after the program end). Brody et al. (2004, 2005) and Brody, Murry, Gerrard et al. (2006) reported completion rates of 96% for both the pretest and posttest; Brody, Murry, Kogan et al. (2006) reported a completion rate of 92% at 29 months; Gerrard et al. (2006) used the 281 participants (85%) who completed the pretest, posttest, and 29-month follow-up; and Murry et al. (2007) used the 284 participants (86%) who provided data at all time points.

Sample: As reported in Brody et al. (2004), the families who completed the posttest had an average of 2.7 children. In 53.6% of these families, the target child was a girl. Of the mothers in the families, 33.1% were single, 23.0% were married and living with their husbands, 33.9% were married but separated from their husbands, and 7% were living with partners to whom they were not married. Of the two-parent families, 93.0% included both of the target child's biological parents. The mothers' mean age was 38.1 years, and the fathers' mean age was 39.4 years. A majority of the mothers, 78.7%, had completed high school. The families' median household income was $1,655.00 per month.

Measures:

The outcome measures came from parent and youth reports. African American university students and community members who were unaware of group assignment served as field researchers to collect data.

Regulated, Communicative Behaviors of Primary Caretaker: Four indicators of the primary caregivers' behaviors obtained from parent reports were combined into an index: involved-vigilant parenting, adaptive racial socialization, communication about sex, and clear communication of expectations about alcohol use. Alphas exceeded .70.

Youth Parenting Assessment: Youths reported on three variables: receipt of nurturant-involved parenting, racial socialization, and clear household rules from their primary caregivers. Alphas ranged from .61 to .85.

Youth Protective Factors: The measures consisted of youth-reported variables that were combined into an index: future-oriented goals, resistance efficacy, negative images of drinkers, acceptance of parental alcohol influences, and negative attitudes toward alcohol use and sexual activity. Alphas ranged from .62 to .81.

Youth Self-Control: Mothers assessed their child's self-control using scales for self-control (alphas = .79-.83) and lack of self-control (alphas = .59-.70). This parent-rated measure may not be independent.

Risk Behaviors: Risk behavior initiation was evaluated using the sum of three dichotomous items: ever drank alcohol, ever used marijuana, or ever had sexual intercourse.

Youth Alcohol Use: Brody, Murry, Kogan et al. (2006) used a youth-reported alcohol composite index measuring ever consumed an entire alcoholic drink, consumed an entire alcoholic drink during the past month, and ever drunk three or more alcoholic drinks at one time (binge drinking). The mean alpha across the three waves of data collection was .70.

Sexual Risk Behavior: Murry et al. (2007) used a nine-item measure of sexual risk (e.g., used a condom during last sex) that had an alpha of .67.

Mediators (Prototype Images and Similarity): Gerrard et al. (2006) measured judgements (popular, selfish, smart, cool, unattractive, and dull) of images of alcohol-drinking youths, parental monitoring of child behavior with parent reports on knowing "where your child is," frequency of parental involvement and inductive discipline, willingness to drink as shown in responses to hypothetical scenarios, and intentions to use alcohol in the next year.

Analysis: Most analyses used structural equation modeling, which controlled for measurement error and examined change over time with latent growth models (and thereby included the baseline outcomes). Exceptions included the analysis of alcohol use initiation, which examined differences in proportions.

The analyses did not adjust for clustering within the eight county units that were used for randomization because there were too few clusters for multilevel structural equation models. Brody et al. (2004) argued that intraclass correlations of .01 for communicative parenting and .04 for youth protective factors justified not adjusting for clustering. They also replicated the results using multilevel ANCOVA models. In addition, Brody, Murry, Kogan et al. (2006) reported that multilevel tests of county effects did not reveal any statistically significant differences.

Missing Data Strategy: The study used complete case analysis without imputation or FIML.

Intent-to-Treat: The analyses included all families in the intervention condition who completed all assessments regardless of the number of prevention sessions they attended.

Outcomes

Implementation Fidelity:

Brody et al. (2004) reported that 65% of the intervention families took part in five or more sessions, with 44% attending all seven. Coverage of the curriculum components exceeded .80 for the primary caregiver, target child, and family sessions.

Baseline Equivalence:

Using HLM because of the hierarchically nested structure of the data, tests for baseline equivalence found two significant differences across 13 measures (See Table 1, Brody et al., 2004). Scores were higher in the control group than in the prevention group on mothers' reports of communication about sex and youths' reports of negative attitudes toward alcohol and sexual activity. The researchers controlled for these measures in the analyses.

Differential Attrition:

Attrition was minimal (4%) at posttest (Brody et al., 2004, 2005; Brody, Murry, Gerrard et al., 2006). At 29 months, Brody, Murry, Kogan et al. (2006) reported that two-factor multivariate analyses of variance for each baseline measure revealed no significant condition-by-attrition interaction effects. Gerrard et al. (2006) stated that "There were no differences in refusal or attrition rates across groups." Murry et al. (2007) did not address differential attrition.

Posttest:

Posttest (Brody et al., 2004, 2005; Brody, Murry, Gerrard et al., 2006): Intervention parents reported significantly higher regulated communication about race, sex, and alcohol than control group parents (d = .47). Intervention youths reported significantly stronger protective factors (e.g., efficacy, attitudes) (d = .39) and significantly fewer risk behaviors (composite of drinking, marijuana use, and sexual intercourse) than the control youths.

Mediation tests found that changes in regulated, communicative parenting behaviors mediated the effect of treatment condition on changes in the other four outcomes of youth protective factors, youth assessments of parenting, youth risk behaviors, and youth self-control.

Long-Term:

29 Months (Brody, Murry, Kogan et al., 2006): Intervention youths reported significantly lower proportions of new alcohol users and a significantly slower rate of growth in alcohol use than control youths. Mediation tests found that intervention effects on changes in youth alcohol use through 29 months were mediated by improvements in youth protective processes from pretest to posttest.

29 Months (Gerrard et al., 2006): Youths in the intervention group reported significantly smaller increases in drinking between pretest and follow-up than did those in the control group. Mediation tests found that the effect of the intervention on change in alcohol consumption was mediated through decreases in children's willingness to drink (by making their images of drinkers less favorable) and intentions to drink.

29 Months (Murry et al., 2007): The intervention significantly improved parenting practices (d = .49) but did not affect sexual risk behaviors. Mediation tests found that intervention-induced changes in parenting behaviors were associated indirectly with sexual risk behavior through adolescent self-pride, peer orientation, and sexual intent.

Study 2

Summary

Brody et al. (2008, 2010) and eight other reports used a cluster randomized trial to examine 667 African American mothers and their 11-year-old children residing in eight rural county units in Georgia. The sample expanded upon that in Study 1 to include both the first and second cohorts. The eight county units were randomly assigned to either the control or intervention conditions. Families in both groups completed a posttest and four additional assessments through 65 months with measures of parenting, conduct problems, alcohol use, and sexual risk behavior. Subsamples ranging from 272-517 youth provided blood draws and survey measures at ages 19-25 (8-14 years after baseline).

Brody et al. (2008, 2010) and eight other reports found that intervention youths relative to control youths reported significantly:

• fewer conduct problems at 29 months for a propensity-score matched subsample (n = 482)
• better parenting at posttest
• lower alcohol use initiation through 29 months
• slower growth in alcohol use frequency through 65 months
• lower blood markers of inflammation (an indicator of likely chronic health problems) at eight years for a subsample (n = 272)
• lower blood cotinine levels (an indicator of smoking) at eight years for a subsample (n = 424)

Evaluation Methodology

Design:

Recruitment: The study recruited African American mothers and their 11-year-old children (grade 5) from nine rural, high poverty counties in Georgia, but two small contiguous counties were combined into one county unit. Schools in these eight units provided lists of 11-year-old students. Of the families assessed, 667 (64%) met eligibility requirements and agreed to participate. The sample expanded upon that in Study 1 to include both the first and second cohorts.

When the youths had reached ages 19-20, a subgroup of 500 was randomly selected for continued study, as budget limitations prevented follow-up of all randomized participants.

Assignment: In a cluster randomized design, the eight county units were randomly assigned to the intervention condition (n = four county units, 369 families) or the control condition (n = four county units, 298 families). While the intervention families participated in the seven-week prevention program, control families received three leaflets on adolescent development, stress management, and the importance of exercise. As in Study 1, consent appeared to follow randomization of the counties, but participants may not have been aware of their assignment when consenting.

Some discrepancies in figures appeared across the studies. Miller et al. (2014) reported 372 families in the intervention condition and 295 families in the control condition, which indicates three more families in the intervention group and three fewer in the control group. Murry et al. (2014) reported 371 families in the intervention condition and 299 in the control group, which indicates a total sample of 670.

Assessments/Attrition: Brody et al. (2008) reported on assessments at posttest (about eight months after pretest and five months after the program end) and at long-term follow-up (about 29 months after pretest and 26 months after the program end). However, counting the booster session extends the program length to one year and defines a 17-month follow-up. About 91% completed both the posttest and the long-term follow-up.

Brody et al. (2010) reported on assessments at 9, 18, 29, 53, and 65 months after pretest. About 85% of the participating families completed the pretest, posttest, and four follow-up assessments. Murry et al. (2014) reported 95% completion at posttest and 85% completion at 65 months.

Studies of blood markers done 8-14 years after baseline included young adults at ages 19-25 but began with a randomly selected subsample of 500. After attrition, the studies ended up analyzing 272 at age 19 (Miller et al., 2014), 399 at age 20 (Brody et al., 2016), 424 at age 20 (Chen et al., 2017), 388 at age 20 (Beach et al., 2018), and 390 at age 25 (Brody et al., 2017).

Brody et al. (2019) provided somewhat different figures. They stated, "At age 19, a reduced sample of 520 young adults was randomly selected," and 517 "provided data on drug use and BMI during at least 1 year from ages 19 to 25." Further, "70.0% provided data at all four waves spanning ages 19-25, an additional 19.4% provided data at three waves, and the remaining 10.6% provided data at one or two waves. The data analyses were conducted with a sample of 517 young adults."

Brody et al. (2020) examined 381 individuals from the subsample of 500 (76%) who had both teacher reports of self-control at pretest and participant assessments at ages 19, 20, 21, and 25 years.

Sample: As presented in Brody et al. (2008), the families who completed the posttest had an average of 2.7 children. In 52.7% of these families, the target child was a girl. Of the mothers in the families, 54.0% were single, 36.2% were married and living with their husbands, 2.2% were married but separated from their husbands, and 7.1% were living with partners to whom they were not married. The mothers' mean age was 37.7 years, and the fathers' mean age was 39.8 years. A majority of the mothers, 80.0%, had completed high school. For the smaller sample in Brody et al. (2016), about 46% of the families lived below federal poverty standards.

Measures:

The outcome measures through the 65-month follow-up came from parent and youth reports. As noted in Brody et al. (2010), African American university students and community members who were unaware of group assignment served as field researchers to collect data. The outcome measures for young adults (ages 19-25) came from blood draws or youth self-reports.

Parenting: Parents completed four measures that were combined into a single index of regulated, communicative parenting. The four measures included communication about drugs, alcohol, and sex; consistent child management practices; positive affect in the parent-child relationship; and racial socialization. Alphas ranged from .69-.88. In addition, Miller et al. (2014) used a measure of parenting that combined measures of the frequency of nurturant-involved parenting and the frequency of harsh-inconsistent parenting. Chen et al. (2017) used a measure of parenting that combined measures of emotional support, youth management, and consistent non-harsh discipline (alphas = .81-.84). Brody et al. (2020) referred to measures of harsh parenting and supportive parenting.

Conduct Problems: Brody et al. (2008) examined a measure that counted the past year frequency of engaging in disruptive behaviors involving theft, truancy, and suspension from school.

Alcohol Use Frequency: Brody et al. (2010) measured if a youth ever drank alcohol and how many times over the past month a youth drank alcohol.

HIV-Related Risk Behavior: Murry et al. (2014) used a measure that summed three dichotomous items: ever had sex, sex in the last month, and use of a condom.

Blood Markers of Inflammation: Miller et al. (2014) reported on an outcome that combined six blood markers of inflammation (alpha = .93). Low-grade inflammation contributes to multiple health problems, including obesity, insulin resistance, high blood pressure, the early stages of coronary heart disease, and psychiatric conditions like depression, posttraumatic stress disorder, and substance abuse.

Blood Cotinine Marker of Smoking: Chen et al. (2017) used a measure of cotinine, a byproduct of nicotine exposure and an indicator of smoking.

Blood Marker of Epigenetic Aging: Brody et al. (2016) used a DNA measure to identify the disparity between an individual's biological and chronological ages. A high score indicates the undesirable outcome of premature or accelerated aging.

Blood Marker of Prediabetes: Brody et al. (2017) measured fasting glucose levels to identify participants with prediabetes.

Substance Use: Beach et al. (2018) examined a measure that combined alcohol use and marijuana use.

Drug Use: Brody et al. (2019) examined a measure that combined past-month cigarette use, excessive drinking, and marijuana use. This composite variable was averaged across ages 19-25 and log transformed to reduce skewness.

BMI: Brody et al. (2019) used each participant's weight and height to calculate the body mass index. Adjusting for pregnancy status did not affect the influence of the measure in the analyses.

Income and Poverty: Brody et al. (2020) measured self-reported average monthly gross personal income and poverty status based on federal poverty guidelines (0 = above 150% poverty, 1 = below 150% and above 100% poverty, and 2 = below 100% poverty).

Analysis:

The analyses mostly used structural equation models with controls for baseline outcomes but also used several types of regression models. The studies of blood markers (Miller et al., 2014; Brody et al., 2016, 2017; Chen et al., 2017; Beach et al., 2018) lacked baseline measures, as did the study of young adult BMI, drug use, and income (Brody et al., 2019, 2020). None of the analyses adjusted for clustering within counties - the unit of assignment. Brody et al. (2008) stated that "Tests of county effects on conduct problems, parenting factors, and youth protective factors [using HLM models] revealed no significant effects based on county membership. Accordingly, the study hypotheses were analyzed at the individual level." In examining a measure of neighborhood SES as a predictor, Brody et al. (2019) estimated models that accounted for non-independence of observations among participants from the same neighborhood.

Brody et al. (2008) adjusted for differences at baseline with a propensity-matched subsample of the randomized sample (n = 482 rather than 667). The propensity score model used "predictors of condition assignment plus key demographic variables to model the distribution of the treatment indicator variable."

Missing Data Strategy: The 10 reports appeared to use complete case analysis without imputation or FIML.

Intent-to-Treat: The 10 reports appeared to include all families in the intervention condition who completed the assessments regardless of the number of prevention sessions they actually attended. The analyses of outcomes at ages 19-26 were based on a randomly selected subsample to reduce follow-up costs, but they also used all participants with complete data regardless of program attendance.

Outcomes

Implementation Fidelity:

Brody et al. (2008) reported that mean attendance was 4.7 of seven total sessions and that mean adherence to intervention components was 90%.

Baseline Equivalence:

Of nine tests examined in Brody et al. (2008), three indicated significant baseline differences for the full randomized sample: per capita income, conduct problems, and youth protective processes. After selecting a propensity-matched subsample (n = 482), significant baseline differences remained for conduct problems and youth protective processes, which were controlled for in the analyses. The other nine studies examined baseline equivalence for their respective analysis samples, as described below.

 Differential Attrition:

Brody et al. (2008) stated only that "No significant experimental condition x attrition interaction effects emerged." Brody et al. (2010) offered more details, stating: "A two-factor multivariate analysis of variance was conducted with the intervention and control groups at each wave of data collection for demographic characteristics (per capita income, maternal education and age, number of children in the household) and outcome measures (alcohol use initiation, frequency of past-month alcohol use). No significant Group x Attrition interaction effects emerged for any variable at any assessment." Murry et al. (2014) simply reported that all but one of the pretest measures were equivalent and that "No differential attrition effects were detected."

The studies of young adults that began with a randomly selected subsample of 500 had varied amounts of attrition and varied tests for differential attrition:

• Miller et al. (2014) conducted six tests for baseline equivalence on the analysis sample and found no significant condition differences. However, the intervention group families in the subsample were more actively involved in the program than the families not in the subsample.
• Brody et al. (2016) compared means for four measures between those providing blood samples at age 20 with those not providing them. Tests revealed no differences (Table 2).
• Brody et al. (2017) noted that the attrition rates in the control and intervention conditions did not differ significantly. Also, a two-factor multivariate analysis of variance for participants who did or did not provide blood samples at age 25 found no significant main effects of condition nor significant interaction effects on eight baseline measures (Table 1).
• Chen et al. (2017) used multivariate analysis of variance with condition and attrition status as factors. No significant main effects or interaction effects emerged for six measures of demographics, smoking, or parenting. Tests for baseline equivalence across conditions for those with cotinine data also found no significant differences.
• Beach et al. (2018) did not report an attrition analysis.
• Brody et al. (2019) reported on tests of baseline equivalence for four variables using the analysis sample. One significant difference showed that intervention families were more disadvantaged than control families. Figures from the CONSORT diagram indicated completion rates of 81% for the intervention group and 73% for the control group.
• Brody et al. (2020) used a two-factor multivariate analysis of variance to compare "participants with or without missing data by intervention group [and] no significant interaction effects emerged for any baseline variables." However, the authors also reported that "Families assigned to the intervention condition experienced more socioeconomic disadvantage than did those assigned to the control condition." Further, the analytic sample had a significantly higher percentage of female participants than the original sample (59.3% vs. 52.8%), and teachers reported that members of the analytic sample displayed significantly higher levels of self-control at baseline than did the original cohort. Overall, the CONSORT diagram indicates completion rates of 65% for the intervention group and 58% for the control group.

Posttest and Long-Term:

29 Months (Brody et al., 2008): Results for the propensity-matched subsample (n = 482) found that intervention youths reported significantly fewer conduct problems than control youths (see Table 2, Model 1). Moderation tests found stronger intervention effects for youths with low levels of self-control and numerous affiliations with deviance-prone peers than for their counterparts. Moderation-mediation tests found that youth protective factors partially mediated intervention effects on conduct problems for youth with lower self-control at baseline, while parenting partially mediated intervention effects on conduct problems for youth who associated with more deviance-prone peers.

65 Months (Brody et al., 2010): The results showed that the intervention youths reported a significantly slower rate of increase in alcohol use frequency than the control youths over the 65-month follow-up period. Mediation tests found that the program significantly reduced alcohol initiation at 29 months, which in turn significantly reduced the growth in alcohol use frequency from baseline to 65 months. 

65 Months (Murry et al., 2014): The results focused only on mediation. The intervention group reported significantly improved parenting, which in turn improved youth perceptions of norms and expectations, which reduced youth opportunities for risk behavior, and which reduced youth risk behavior. However, the analysis did not test for the significance of indirect effects of the program on the sole behavioral outcome of risk behavior.

8 Years (Miller et al., 2014): Intervention participants had significantly lower blood markers of inflammation than control participants (d = -.90). Mediation tests found that the intervention effect was brought about in part by improved parenting at posttest. Moderation tests found stronger effects of the intervention on parenting and inflammation for youths from disadvantaged families.

9 Years (Brody et al., 2016): The study did not examine main effects of the intervention but found moderated and mediated effects. For moderation, the intervention had stronger effects on epigenetic aging for youths whose parents were depressed at baseline than for other youths. For moderated mediation, the intervention reduced harsh parenting among parents with baseline depression, which reduced epigenetic aging.

9 Years (Chen et al., 2017): Intervention participants had significantly lower levels of cotinine in the blood than control participants at age 20 (d = -.20). They also had significantly higher levels of supportive parenting at age 16 (d = .25). Mediation tests found that the intervention effect on cotinine levels was brought about by more supportive parenting at age 16.

9 Years (Beach et al., 2018): Results for a subsample with genotype information (n = 388) showed that intervention youths reported significantly lower substance use (i.e., alcohol and marijuana) initiation at age 13 than the control group. Moderation tests at nine years found that the intervention improved the genetic biomarkers related to regulation but only for those with a particular genetic allele.

14 Years (Brody et al., 2017): Results for the subsample with blood draws at age 25 (n = 390) examined the presence of prediabetes. There was no significant main effect, but tests found a moderation effect. The intervention had a stronger beneficial effect on prediabetes at age 25 for participants with parents having had adverse childhood experiences.

8-14 Years (Brody et al., 2019): Results across ages 19-25 years indicated no significant main effects of the intervention on BMI or drug use. Moderation tests for both neighborhood disadvantage at ages 11-16 and gender found that the intervention 1) significantly reduced the BMIs of women who lived in disadvantaged neighborhoods and 2) significantly reduced drug use among men who lived in disadvantaged neighborhoods. Moderation-mediation tests further showed that for the subgroup of men who had lived in a disadvantaged neighborhood, the effect of the intervention on drug use was mediated by improved parenting.

14 Years (Brody et al., 2020): Results at age 25 indicated no significant main effects of the intervention on employment income or poverty status. Moderation tests for self-control at age 11 found that the intervention significantly reduced both employment income and poverty status for young adults who had low levels of teacher-rated self-control at baseline (i.e., age 11). Moderation-mediation tests further showed that for the subgroup of participants with low baseline self-control, the effect of the intervention on employment income was mediated by enhanced supportive parenting.

Study 3

Summary

Futris and Kogan (2014) and Kogan et al. (2016) used a randomized controlled trial to examine 465 African American fifth- and sixth-grade students and their primary caregivers residing in eight rural counties in Georgia. Families were randomly assigned to either the intervention condition or a waitlist control group. Families in both groups completed a pretest and posttest (approximately three months after intervention end) with measures of parenting, youth self-regulatory behavior, and risk behavior vulnerability.

Futris and Kogan (2014) and Kogan et al. (2016) found that relative to the control group, the intervention group had significantly greater improvements in:

• youth risk behavior vulnerability
• youth self-regulatory processes
• parenting practices

Evaluation Methodology

Design:

Recruitment: Participants were recruited over the span of five years from school lists of African American fifth- and sixth- grade students in eight impoverished rural counties in Georgia. A total of 465 families were recruited during this time. Eligibility requirements for families included the presence of a youth 11 or 12 years of age at pretest who self-identified as African American; caregiver self-identification as African American was not required.

Assignment: Families were randomly assigned to either the intervention group (n=242) or a wait-list control group (n=223) that would receive the program after a 12-month waiting period.

Attrition: Assessments occurred at baseline and posttest, three months after program completion. At posttest there was an overall retention rate of 89.5%.

Sample:

Child participants ranged from 10-13 years in age, averaging 11.49 years at baseline, and were evenly split by gender (51% male). Primary caregivers were predominantly female (95.9%) and ranged in age from 22-81 years, with a baseline average of 37.11 years. Most caregivers were a biological parent of the participant child (89.9%) and most lived in a single-parent household (61.6). More than half were below the poverty line (57.4%).

Measures:

Assessments occurred at baseline and posttest (six months after pretest and three months after program completion).

Intervention-targeted parenting was assessed with both parent and youth reports on a nine-item involved/vigilant parenting scale, which assessed caregivers' use of monitoring, consistent discipline, and positive parenting practices. Internal consistency was relatively low (α=.62-.63). Parents and youth also completed a nine-item scale assessing parental discussion with youth regarding expectations about risky behavior such as alcohol use (α=.94-.95). Parents also reported on relationship harmony and distress (e.g., "your child is easy to get along with") using the Interaction Behavior Questionnaire (α=.90).

Intervention-targeted self-regulatory processes were assessed with youth self-reports on a four-item anger control scale (α=.84), a seven-item scale indexing behavioral self-control (α=.70), and a resistance efficacy measure that presents hypothetical scenarios where youth are offered alcohol, cigarettes, or marijuana to examine behavioral intentions (α=.72-.74).

Risk behavior vulnerability was assessed with the Tolerance for Deviance Scale (α=.86-.87), as well as a 13-item inventory on intentions to use substances and engage in sexual activity (α=.86-.88) and a 12-item measure assessing affiliations with risk-taking peers (α=.80-.81).

Parent-reported youth self-control was assessed by asking "how often does your child…work toward a goal, think ahead of time about the consequences of his/her actions."

Parent depressive symptoms were assessed with a 20-item depressive symptoms measure. Reliability was not reported.

Analysis:

The effects of the program at posttest were evaluated with structural equation models that controlled for baseline levels of risk behavior vulnerability, intervention-targeted parenting, and youth self-regulatory processes. Complier-average causal effect modeling was used to identify and compare latent groups of compliers within both conditions. Multilevel analysis was used to account for clustering by county.

Intent-to-Treat: All available data were used in analyses.

Outcomes

Implementation Fidelity:

All sessions were recorded and adherence checks were performed on six sessions (28%) for each seven-week program. Mean session coverage was 84.5% and did not vary significantly by site or implementation year.

Baseline Equivalence:

Families assigned to the intervention group were experiencing slightly (but significantly) more economic hardship than control group participants. Otherwise, the groups were equivalent on demographic and outcome measures.

Differential Attrition:

Tests revealed no significant differences between program completers and attriters, though baseline-by-condition attrition was not examined.

Posttest:

At posttest, compared to control participants, intervention youth showed significantly greater improvements in youth risk behavior vulnerability, youth self-regulatory processes, and parenting practices.

Long-Term:

Not examined.

Study 4

Summary

Kogan et al. (2019) used a randomized controlled trial to examine 472 families with an African American youth aged 11 or 12 years in seven rural Georgia counties. Families were randomized to either the intervention group (53%) or a control group (47%). Participants completed four assessments: baseline, posttest (10 months post-baseline), initial follow-up (22 months post-baseline) and final follow-up (34 months post-baseline). The main outcome measure was the frequency of alcohol use.

Kogan et al. (2019) found that relative to the control group, the intervention group had a significantly: 

• slower growth in alcohol use frequency through 34 months

Evaluation Methodology

Kogan et al. (2019) reported main effects and Kogan et al. (2023) reported moderation effects that tested whether the intervention attenuated the association of racial discrimination with adverse mental health outcomes among Black adolescents.

Design:

Recruitment: The sample of families was recruited from seven rural counties in Georgia. School districts provided lists of African American fifth-grade students, whose parents were contacted in random order to discuss participation. The sole eligibility requirement for families was the presence of a youth 11 or 12 years of age at pretest who self-identified as African American. Recruitment began on June 7, 2013, and data collection concluded on December 10, 2017. Of the 825 families screened for eligibility, 625 met the eligibility requirement and 472 enrolled (76%).

Assignment: After the baseline assessment, participants were assigned randomly, with an oversample of about 5% for the treatment group, to the intervention group (n = 252, 53%) or the control group (n = 220, 47%).

Assessments/Attrition: The four assessments occurred at baseline, posttest (10 months post-baseline), initial follow-up (16 months after program end or 22 months post-baseline), and final follow-up (28 months after program end or 34 months post-baseline). Retention from baseline to the final follow-up was 91.7%.

Sample:

Youths had a mean age of 11.61 years at baseline, and 50.8% were girls. The mean age of caregivers was 37.2 years. Of the caregivers, 18.2% had less than a high school education, 26.5% had completed high school or obtained a GED, and the remaining 55.3% had at least some college education. Most participating families, 64.4%, had family incomes below the federal poverty threshold.

Measures:

African American field researchers made home visits to collect data using audio computer-assisted self-interviews on laptop computers. In Kogan et al. (2019), the main outcome, the frequency with which the youths drank alcohol during the past three months, was measured on a scale from 0 (none) to 6 (> 30 times).

Analysis:

The analysis used growth curve models with Poisson distributions to represent changes in alcohol use over all four assessments. In addition, a Complier Average Causal Effect (CACE) model estimated intervention effects for those attending five or more sessions, those attending six or more sessions, and those attending all seven sessions. To estimate the CACE models, the analysis included youths' sex, age, tolerance for deviance, and academic orientation, as well as the family socioeconomic risk index and youth reports of protective parenting. The study reported one-tailed tests (excepting tests for baseline equivalence).

Kogan et al. (2023) conducted exploratory post hoc moderation analyses (group assignment × discrimination) with baseline adolescent gender and family socioeconomic disadvantage included as covariates.

Missing Data Method: The models used full information maximum likelihood estimation that included all available data. The authors noted that the attrition analysis, as described below, suggested that the data met the requirement of missing at random.

Intent-to-Treat: The intent-to-treat analysis used all available data with full information maximum likelihood (FIML) estimation. The CACE analysis, however, examined subsets of intervention participants based on high program dosage.

Outcomes

Implementation Fidelity:

Mean attendance was 4.5 of seven total sessions. Approximately 33% of those assigned to the intervention group attended all sessions, and approximately 20% attended no sessions. Adherence assessments indicated that coverage of the prevention curriculum exceeded 80% of manualized activities.

Baseline Equivalence:

Tests for six baseline sociodemographic and outcome measures using the full randomized sample (Table 1 of Kogan et al., 2019) found no significant differences.

Differential Attrition:

The attrition rates were 10% for the intervention group and 6% for the control group. The authors stated that attrition at the last follow-up "was not associated with experimental condition or baseline demographics, including youth gender, age, and family socioeconomic risk (e.g., parental education, family poverty status)."

Posttest and Long-Term:

The intent-to-treat analysis found that the linear increase in alcohol use over the study period was significantly slower for the intervention group than the control group (p = .02 for the one-tailed test remains significant at p < .05 for a two-tailed test).

The dose-response analysis using CACE models revealed a significant medium-sized intervention effect on growth in alcohol use (d = .51) when participants attended five of the seven offered sessions and a very large effect (d = 1.51) when participants attended all seven sessions.

In exploratory moderation analyses using two-tailed tests, Kogan et al. (2023) found that study condition significantly moderated the association between Black adolescents' experiences of racial discrimination and depressive symptoms 28 months after program end (p = .005). For the control group, racial discrimination was associated with increases in depressive symptoms (p < .001), whereas there was no association between racial discrimination and depressive symptoms for the intervention group (p = .09).