Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2019 Feb 1.
Published in final edited form as: Prev Sci. 2018 Feb;19(Suppl 1):27–37. doi: 10.1007/s11121-016-0632-9

Attendance Patterns and Links to Non-Response on Child Report of Internalizing among Mexican Americans Randomized to a Universal Preventive Intervention

Anne M Mauricio a, Jenn-Yun Tein a, Nancy A Gonzales a, Roger E Millsap a, Larry E Dumka b
PMCID: PMC4956608  NIHMSID: NIHMS753150  PMID: 26786469

Abstract

We examined attendance trajectory profiles among 335 Mexican American families participating in an 11-week universal intervention to explore if heterogeneity in attendance and thus dosage was associated with intervention response, defined as pre-to-2 year post (T2) reductions in child report of internalizing symptoms. We estimated trajectories accounting for the influence of baseline covariates, selected based on the Health Belief Model (HBM) and Latino family research, to understand covariate associations with trajectories. Results supported six attendance trajectory groups: Non-Attenders (NA), Early Dropouts-Low Internalizing (EDOLI), Early Dropouts-High Internalizing (EDO-HI), Mid-Program Dropouts (MPDO), Sustained Attenders-Low Internalizing (SA-LI), and Sustained Attenders-High Internalizing (SA-HI). All groups except EDO-HI showed significant pre-to-post change on child report of internalizing; however, trajectory groups reflecting more attendance did not have greater pre-to-post change. Nonetheless, child report of internalizing differentiated two subgroups of sustained attenders and two subgroups of early dropouts. These results suggest heterogeneity among families with similar patterns of attendance and highlight the importance of modeling this heterogeneity. Although life stress was a barrier to participation, there was minimal support for the HBM. Cultural influences, acculturation and familism, played a more prominent role in distinguishing trajectories. As expected, the EDO-HI group was less acculturated than both sustained attender groups and reported weaker familism values than the SA-HI group. However, unexpectedly, the SA-LI group had lower familism than the EDO-LI group. The results suggest that the influence of culture on participation is nuanced and may depend on child symptomatology.

Keywords: Universal intervention, Mexican-American, attendance patterns, internalizing, non-response

The prevalence of internalizing disorders during adolescence is as high as 14% and lifetime prevalence can be as high as 40% (Merikangas et. al., 2010). The estimated cost to treat these disorders is $12 billion annually (Wang, Simon, & Kessler, 2003). Although family-centered interventions prevent escalation in internalizing symptoms and the onset of internalizing disorders in adolescence, not all youth respond equally to these interventions (NRC/IOM, 2009). Intervention attendance can vary extensively (Baker, Arnold, & Meagher, 2011), influencing dosage or the amount of program received, and thus intervention response. Heterogeneity in attendance is particularly likely in universal prevention programs, as participants will vary in their baseline level of risk, perceived need for an intervention, and motivation to attend.

The current study modeled heterogeneity in attendance among Mexican American (MA) families randomly assigned to the intervention condition in the Bridges to High School (Bridges) efficacy trial (Gonzales et al., 2012). Bridges is a school-based universal intervention for middle school students and their parents that strengthens youth and family competencies as a pathway to the prevention of later mental health and substance use problems, including adolescents’ internalizing symptoms. In intent-to-treat analyses, Bridges showed effects on key outcomes (e.g., adolescent substance use) and on program mediators relevant to internalizing symptom reduction (e.g., parenting skills, coping efficacy); however, there was no observed intervention response on child report of internalizing symptoms (Gonzales et al., 2012). Although families attended five sessions on average, there was much variation in attendance (SD = 3.44), and approximately 17% attending zero sessions and 20% attending only one or two. Consistent with evidence that participant attendance predicts child outcomes in prevention trials (Reid, Webster-Stratton, & Baydar, 2004), we explored if heterogeneity in attendance related to participants’ “nonresponse” on child report of internalizing. Research examining MA participation in preventive interventions that target child internalizing is important because MA children are disproportionately exposed to poverty and corresponding risk factors (Pew Hispanic Center, 2006) that increase vulnerability to internalizing disorders (U.S. Dept. of Ed., 2000).

The primary goal of this study was to identify qualitatively distinct attendance trajectory profiles and examine if these profiles differed in intervention response (or “non-response), operationalized as pre-to-post change on child reported internalizing symptoms. In addition to the pattern of nonparticipation (i.e., zero attendance) observed in our data (i.e., 17%), we hypothesized three distinct patterns of participation-sustained attendance, variable attendance, and early dropout, consistent with prior research, (e.g., Baker et al., 2011; Coatsworth, Duncan, Pantin, & Szapocznik, 2006; Kazdin & Mazurick, 1994). We also hypothesized that the magnitude of pre-to-post change on child self-report of internalizing would differ across trajectories, with change greatest for families with sustained participation and variable attenders reporting greater pre-to-post change than early dropouts and non-attenders.

Person-Centered Methods to Examine Patterns and Correlates of Attendance

A secondary goal of this study was to examine how parent, child, family, and cultural variables functioned to differentiate attendance trajectories. Prior studies examining correlates of participation have produced variable results, with the valence of associations between correlates and participation often varying from study to study (McKay & Bannon, 2004). One explanation is that prior studies did not examine the multiple ways that correlates might operate across time to explain attendance patterns. For example, although families of higher risk youth might be motivated to participate initially due to their perception of greater need, they may not be able to sustain attendance over time due to stress associated with their child’s difficulties.

Studies examining correlates of participation have generally operationalized attendance as a dichotomous outcome (i.e., attended or did not attend) or as the number of sessions attended. Because these methods do not account for variation in participation across program duration or for family disengagement from a preventive intervention at varying points, they limit our understanding of nuanced associations between attendance and covariates of participation. Although some studies have modeled patterns of participation (e.g., Coatsworth et al., 2006; Baker et. al., 2011; Gorman-Smith, Tolan, Henry, & Leventhal, 2002), these studies did not account for change in participation across time, while estimating the influence of baseline covariates on patterns of attendance. The current study used growth mixture modeling (GMM; Muthén, 2002) to address these limitations. GMM is a model- or probability-based technique that uses repeated observed outcomes to model distinct patterns of growth that reflect unobserved heterogeneity in the population and can simultaneously account for the influence of baseline covariates and distal outcomes on trajectories. GMM assumes multiple trajectories with unique intercepts and slopes and classifies persons into trajectory groups based on posterior probabilities (Muthén, 2004). We used GMM to estimate latent attendance trajectory profiles, while simultaneously accounting for the influence of parent, child, family and cultural covariates as well as child self-report of internalizing, our distal outcome. Figure 1 illustrates our model.

Figure 1.

Figure 1

Open in a new tab

An illustration of GMM. Model accounts for parent, child, family, and cultural covariates at baseline and mean levels of our targeted outcome, child report of child internalizing symptomatology, at pretest (T1) and two-years post intervention (T2). S1–S11 represents the binary variables for Session 1 to Session 11.

Parent, Child, Family and Cultural Covariates of Attendance Trajectories

We integrated the Health Belief Model (HBM, Janz & Becker, 1984; Rosenstock, Irwin, Stretcher, & Becker, 1988) and Latino family research on intervention participation (e.g., Carpentier al., 2007) to select baseline covariates. The HBM is a health behavior change model grounded in psychological theory. We focused on three HBM domains that account for participation in health-related programs: (a) perceived need (based on perceived problem susceptibility and severity and potential benefits), (b) self-efficacy, and (c) barriers to action. We included baseline variables relevant to each domain and proposed a cultural extension of the HBM specifically relevant to our Latino population. We focused on parent perceptions for each domain, because we expected parents would drive family participation (Carpentier et al., 2007).

Parents’ perceived need: Baseline child functioning and parenting skills

According to the HBM, people are more likely to perceive a need for and engage in a preventive health behavior if they believe the behavior will mitigate a health risk they for which they are vulnerable (Janz & Becker 1984). Consistent with the HBM, parents who perceive their child is maladjusted may be more likely to participate in a preventive intervention aimed to improve child outcomes (Spoth & Redmond, 1995). In this study, we included parent reports of child internalizing and externalizing symptoms as indicators of parent perceptions of baseline functioning. Because Bridges aimed to improve academic engagement and success (Gonzales, Dumka, Mauricio, & German, 2007), we also included pre-intervention GPA as an indicator of child baseline functioning, as we expected parents would perceive their child’s poor academic performance as indicative of a need for Bridges. We hypothesized that more child internalizing and externalizing symptoms as reported by parents and a lower GPA would be associated with patterns of sustained and variable attendance, rather than with early dropout and non-attendance.

A primary aim of Bridges was to strengthen parenting skills that promote academic engagement and reduce risk for mental health problems during the transition from elementary to middle school (Gonzales et al., 2007). From a developmental perspective, parental monitoring is critical in early adolescence and middle school, when non-adherence to family and social norms and vulnerability to peer pressure increases (Dishion & McMahon, 1998). Consistent with the HBM and research showing that parents who rate their parenting quality as poor have higher rates of intervention participation (e.g., Gorman-Smith et al., 2002; Perrino et al., 2001), we hypothesized that parents self-reporting less baseline monitoring would perceive a greater need for Bridges and be sustained or variable attenders, rather than early dropouts and non-attenders.

Parenting self-efficacy

According to the HBM, self-efficacy or one’s belief in their ability to effect change is a primary driver of health behavior change (Rosenstock et al., 1988). In the context of the HBM, parenting self-efficacy is the expectation that one can be effective as a parent and that effective parenting will lead to positive child outcomes. In this study, we examined effects of parenting efficacy at baseline to examine the link between the HBM dimension of self-efficacy and attendance trajectories. We also included parent involvement in their child’s education (e.g., communicating with teachers) as an indicator of parent efficacy. Parent involvement in their child’s education indicates parent beliefs that advocating for their child will make a difference in their educational outcomes. Consistent with the HBM, we hypothesized that parents reporting more parenting efficacy and involvement in their child’s education would be sustained attenders, rather than early dropouts and non-attenders.

Participation barriers: Income, family structure, life stress, parent depression

The HBM posits that participation barriers counter factors that motivate participation (Janz & Becker, 1984). Among families who engage, barriers may influence whether or not they dropout as well as timing of dropout. Barriers may also discriminate families who engage from families that never engage (Reyno & McGrath 2006). Income and family structure are two socio-demographic factors that may deter participation. Due to income-related barriers (e.g., reliable transportation) and childcare needs, parents with less-income and more children may participate sporadically or drop out (Coatsworth et al., 2006). We examined the effects of income and family structure, operationalized as number of children in the home, on participation. Parent depression and life stressors can make it challenging to engage in activities beyond daily necessities; in this regard, parent depression and life stress are also barriers to participation (Kazdin, Holland, Crowley, & Breton, 1997). We hypothesized more barriers would be associated with less participation, such that fewer barriers would differentiate: 1) sustained attenders from all other groups, 2) variable attenders from early dropouts and non-attenders, 3) and early dropouts from non-attenders.

Cultural influences: Acculturation and familism

Research on engagement of Latino populations has identified cultural dimensions that link to health beliefs and behaviors. For example, in variable centered analyses, we found less acculturated families attended more Bridges’ sessions (Carpentier et al., 2007). Less acculturated families are also likely to be more recent immigrants for whom education and upward mobility are intertwined (Hill & Torres, 2010), and thus may perceive a greater benefit and be more motivated to engage in a program promoting school success, such as Bridges. Relevant to the HBM, the value placed by an individual on a goal and the match between the program and the values of the individuals or groups that the program is attempting to engage can motivate participation (Kumpfer, Alvarado, Smith, & Bellamy, 2002). Because Bridges is a family-centered intervention and recruitment materials highlighted its emphasis on family as a source of strength to help children succeed, we included parent report of familism values in our cultural extension of predictors relevant to the HBM. We hypothesized that parents endorsing strong familism values would show sustained participation consistent with the strong value they place on the family as a source of support.

Current Study and Hypotheses

The current study uses GMM to estimate multiple and distinct latent attendance trajectory profiles among families in a universal intervention. GMM can model multiple patterns of heterogeneity in participation across program duration and account for families disengaging from an intervention at varying points, while accounting for the influence of parent, child, family and cultural covariates and a distal outcome. In this way, the current study makes a unique contribution to the literature in that it may increase understanding of the nuanced associations between patterns of attendance and variables linked to participation, while also exploring how patterns of participation link to change in an intervention-targeted outcome. The goals of this study were to identify distinct attendance trajectory profiles to explore if heterogeneity in attendance patterns and thus dosage was associated with “nonresponse” or failure to show change on internalizing and to examine if parent, child, family, and cultural covariates would further differentiate attendance trajectory profiles. Figure 1 illustrates our model. We hypothesized four trajectories-sustained attendance, variable attendance, early dropout, and nonattendance. We also hypothesized that pre-to-post change on child self-report of internalizing would be greatest among families with sustained attendance and that variable attenders would report greater pre-to-post change than early dropouts and non-attenders. Based on the HBM and Latino family research, we also hypothesized the following.

  1. Perceived need for Bridges, as indicated by parent report of more child problem behaviors and less parental monitoring, would increase the likelihood of membership in the sustained and variable attendance groups, compared with the non-attendance and early dropout groups.

  2. Self-efficacy, defined as parenting self-efficacy and parent involvement in their child’s education, would be associated with increased likelihood of membership in the sustained attendance group, compared with the non-attenders and early dropout groups.

  3. Participation barriers (i.e., less income, more children, higher levels of parent depression, and more stressful events) would be associated with decreased likelihood of memberships in: a) sustained attendance, in comparison to all other groups, b) variable attendance in comparison to early dropout and non-attendance, and c) early dropout compared to non-non-attendance.

  4. Stronger familism values and lower levels of acculturation would increase the likelihood of membership in the sustained attendance group, in comparison to all other groups.

Method

Participants

Participants in this study were female parents or caregivers (parents) and adolescents from families randomly assigned to the intervention condition for the randomized control trial (RCT) of the Bridges to High School program (Bridges; Gonzales et al., 2012; Gonzales et al., 2014). We randomly selected families for the RCT from 7th grade rosters of five, large, urban middle schools; 598 families completed baseline interviews. Thirty-eight of the 598 families were from a school that we excluded from the RCT analyses; however, families from this school randomized to the intervention condition were included in this study. Including the 38 families, 353 and 189 of the 598 families were randomized to the intervention and control conditions, respectively; 55 were unable to be randomized (i.e., unable to contact, left participating school); 1 participant deceased. Of the 353 families randomized to the intervention, 335 had a participating primary female caregiver and were included in this study. See Figure 2 for a CONSORT flowchart of Bridges recruitment and randomization. Parents were biological and adoptive mothers (93%), stepmothers (2%), and other relatives (e.g., grandmothers, 5%). At baseline, parents’ ages ranged from 23 to 70 (M = 37.50; SD =6.71); 59% were born in Mexico and 41% in the U.S. or another country. Fifty percent of adolescents were females; 78% were born in the U.S. and 22% in Mexico. At baseline, adolescents’ ages ranged from 11 to 14 (M = 12.33; SD =.57).

Figure 2.

Figure 2

Open in a new tab

CONSORT Diagram: Illustration of recruitment, enrollment, and randomization in the Bridges to High School randomized control trial (RCT) and RCT families represented in this study’s sample.

1 The numbers in these cells vary slightly from Gonzales et al., 2012 and Gonzales et al., 2014 because they include 38 families that are from a school that was excluded from the RCT. The 38 families are distributed in the following way: 12 = Unable to Randomize; 11 = Control; 15 =Intervention condition; 14 of the 15 families in the intervention condition had a participating primary female caregiver and were included in this study.

Intervention: Bridges to High School

Bridges is a family-focused, culturally competent universal intervention that builds youth and family competencies to reduce internalizing and other problem behaviors and increase academic engagement following the middle school transition (Gonzales et al., 2012; Gonzales et al., 2014). Bridges’ design was guided by an ecodevelopmental framework (Szapocznik & Coatsworth, 1999) that recognizes adolescents need to adapt to multiple contexts, including families, peers, and schools, as well as cultural factors (e.g., acculturation). Bridges has nine weekly group sessions and two home visits, for 11 sessions. Sessions one (S1) and six (S6) were in the family’s home; providers met individually with parents and adolescents to identify goals (S1) and to individualize program skills practice (S6). All other sessions were in the evening at the child’s school. The nine afterschool sessions included parenting and adolescent components, delivered separately but simultaneously, followed by a conjoint family session. Adolescent groups used active learning methods to teach self-regulation, strengthen adolescent coping resources, and promote positive engagement in school and prosocial activities. The parenting groups used active learning to increase effective parenting practices (e.g., monitoring), decrease parent-adolescent conflict, and promote school engagement. The family sessions aimed to strengthen family cohesion, skills practice, and shared values about the importance of education.

The program was in English and Spanish and families chose the language of participation. Across three cohorts, there were 23 intervention groups (English, n= 11; Spanish, n = 12); each group was led by two group leaders (GL) and there were 16 different parent GLs and 20 different adolescent GLs across all cohorts. Sixty-nine percent of the GLs were Latino/a (predominantly MA), 65% were bilingual, and all had experience working with MA families. GLs had varied professional (e.g., social service providers, teachers) and educational backgrounds (58% masters, 42% bachelor’s degrees). GLs received a comprehensive program manual, 45 hours of pre-service training, 3 hours of weekly training, and 2 hours of weekly supervision during the intervention. Using intent-to-treat analyses (Gonzales et al., 2012), Bridges showed effects on adolescent report of substance use and externalizing problems, teacher report of adolescent internalizing and externalizing problems, father report of adolescent externalizing problems, and school GPA. Baseline risk moderated all effects, and intervention-targeted parent and adolescent mediators mediated all program effects. Intervention group language also moderated program effects on adolescent and teacher reports of externalizing.

Procedures

We collected data via in-home computer administered interviews after informed consent. Participants completed interviews at baseline/pretest, immediate posttest, 1-year post-intervention, and 2-years post-intervention. The current study used baseline/pretest (T1) and 2-years post-intervention (T2) data. Providers documented session attendance immediately after each session. The appropriate Institutional Review Board approved all study procedures.

Measures

Parents’ perceived need: baseline child functioning and parenting skills

Baseline child functioning

Child internalizing and externalizing symptoms and child’s GPA at baseline were indicators of child functioning. We assessed parent report of child internalizing (M = 9.28; SD = 6.80; α = .86) and externalizing (M = 8.01; SD = 6.91; α = .90) using the Child Behavior Checklist (CBCL; Achenbach, 1991). GPA was school-reported letter grades for Language, Math, Social Studies, and Science for the 1st semester of 7th grade, which we transformed into numbers from 13 (A+) to 1 (F) and averaged (M = 7.13; SD = 2.70; α = n/a).

Baseline parenting skills

Parent perceptions of parenting skills at baseline was parent self-report of monitoring, which we assessed with an adaptation of Small and Kerns’ (1993) Parental Monitoring scale that had seven items on a 5-point scale (M = 4.36; SD = 0.65; α = .77).

Parenting self-efficacy

Parenting self-efficacy was self-report of parenting efficacy and involvement in their child’s education. The Multicultural Inventory of Parenting Self-Efficacy, with ten items on a 5-point scale (M = 4.24; SD = 0.55; α = .88; Dumka et al., 2002), assessed parenting efficacy. Six items we developed for Bridges and four items from Gottfried, Fleming and Gottfried’s (1994) Extrinsic Motivation subscale of the Parent Motivational Practices Scale assessed parent involvement in their child’s education. Parents reported if they had done each practice in the past month; we summed positive responses (M = 5.25; SD = 2.48; α = n/a).

Participation barriers: Baseline income, family structure, life stress, parent depression

Income included salaries, child support, and state and federal assistance (M = $35,577; SD = $20,349). Family structure was number of children in the home (M = 3.21; SD = 1.56). The Center for Epidemiologic Studies Depression Scale (Radloff, 1977) with twenty items on a 4-point scale assessed depression (M = 13.91; SD = 9.58; α = .87). The Critical Events subscale of the Barriers to Treatment Participation Scale assessed life stress (Kazdin et al., 1997). Parents reported exposure to eighteen stressors in the past twelve months (M = 2.50; SD = 2.37; α = n/a).

Cultural influences: Acculturation and familism values

We assessed acculturation using the Anglo Orientation subscale of the Acculturation Rating Scale for Mexican Americans-II (Cuellar, Arnold, Maldonado, 1995); it has thirteen items on a 5-point scale (M = 2.97; SD = 1.17; α = .95). We used the familism subscale of the Mexican American Cultural Values Scale (Knight et al., 2010); it has fourteen items on a 5-point scale (M = 4.42; SD = 0.43; α = .78).

Outcomes

The binary measure attendance (1 = yes; 0 = no) was parent attendance at each of the 11 sessions. We used parent rather than child attendance because participation was driven by parent attendance (Carpentier et. al., 2007). The distal outcome was child self-report of internalizing problems (M = 8.26; SD = 7.75; α = .89) two years post-intervention (T2) assessed using the internalizing subscale of the Youth Self Report (YSR; Achenbach, 1991). We included child self-report of internalizing problems at pretest (T1; M = 14.15; SD = 9.08; α = .89) to evaluate reductions in pre-to-post internalizing across trajectory groups. Based on adolescent report at baseline, 29% (n = 84) had T-scores in the clinical range (T-scores ≥ 63) and another 8% (n = 22) of adolescents had T scores in the borderline clinical range (60 ≤ T-scores < 63).

Data Analytic Method

We used GMM (Muthén & Muthén, 1998–2013) with binary indicators of growth factors to estimate attendance trajectories across 11 sessions. As outlined by Muthén (2004), our model estimated trajectories accounting for the influence of baseline (T1) parent, child, family, and cultural covariates as well as T1 and T2 levels of child report of internalizing, hypothesized as conceptually important for defining latent classes and for model specification (see Figure 1).

We used a chi-square difference test to determine whether a linear or nonlinear (i.e., quadratic) growth function best fit the underlying growth process in our data (see Schaeffer, 2003), such that we compared a one-class unconditional growth model with only a linear function to a one-class unconditional growth model with linear and quadratic functions. Next, we estimated models with varying numbers of trajectory groups (i.e., classes) beginning with a one-class model that successively increased by one. In every model, we included a "Non-Attenders" class by constraining the growth factors of this class to zero to represent the 43 families that did not attend any session. We used MPLUS 7.11 (Muthén & Muthén, 1998–2013) and full information maximum likelihood estimation to adjust for missing data (Enders & Bandalos, 2001). We adjusted standard errors to account for participant clustering within intervention groups (n=23), which averaged 15 parents, even though intraclass correlations (ICCs) revealed insignificant intervention-group effects on Bridges outcomes; there were also no school or cohort effects (i.e., ICCs < .05; Gonzales et al., 2012).

GMM uses logistic regression to test associations between covariates and trajectory group membership (Muthén & Muthén, 1998–2013). We used the Mplus MODEL CONSTRAINT command to test the significance of pre-to-post differences on YSR for each trajectory group and between-group differences on the magnitude of this change. We used the false discovery rate (FDR), which controls for the expected proportion of false-positives among all significant hypotheses, to adjust for multiple tests when testing trajectory group differences on pre-to-post change on YSR and associations between covariates and trajectory group membership, for which we based the FDR for each covariate on the number of hypothesized contrasts for that covariate. We interpreted effects as reliable if the FDR adjusted p ≤ 10%.

To avoid convergence to a local optimal solution, we generated 1,000 sets of random starting values for the initial stage and specified 500 optimizations for the final stage of maximum likelihood optimizations; we also replicated the model using a random seed that had resulted in the highest log likelihood (Asparhouv & Muthén, 2012). We used multiple indicators of model fit to determine the best solution: 1) Log likelihood (LL) values, 2) the sample-adjusted Bayesian Information Criterion (Sclove, 1987; SABIC), and 3) the adjusted Lo Mendell Rubin Likelihood Ratio Test (LMR LRT; Lo, Mendell, and Rubin, 2001). Consistent with Muthén’s (2004) recommendation that both substantive and statistical considerations are critical for model selection, we used substantive usefulness of the classes to inform model selection.

Results

Identification of Attendance Trajectories

Comparing one-class unconditional growth models with only intercept + linear vs. intercept + linear + quadratic revealed that the intercept + linear + quadratic growth function provided the best fit to the data [Δχ2(1) = 23.91, p < .001] and was selected as our base model for the GMM analyses. Table 1 shows the fit statistics for k-class solutions from the GMM results. Although the LMR-LRT was not significant for the 3–6 class solutions, we chose the 6-class solution as the best fit because the LL and SABIC showed appreciable decreases and there were conceptually meaningful differences between the classes. That is, in the 4-class solution, four distinct patterns emerged: 1) Non-Attendance, 2) Sustained Attendance, 3) Mid-Program Dropout (i.e., variable attenders), and 4) Early Dropout. In the 5-class solution, two classes with sustained attendance emerged; one had low and the other had high T1 and T2 scores on child self-report of internalizing (i.e., YSR). In the 6-class solution, two early dropout classes emerged; one had high and the other low YSR scores. We interpreted the distinction between the two sustained attendance and two early dropout classes as meaningful (See Figure 3). We labeled the six classes: 1) Non-Attenders (NA; 12%), 2) Early Dropouts-Low Internalizing (EDO-LI; 14%), 3) Early Dropouts-High Internalizing (EDO-HI; 5%), 4) Mid-Program Dropouts (MPDO; 13%), 5) Sustained Attenders-Low Internalizing (SA-LI; 32%), and 6) Sustained Attenders-High Internalizing (SA-HI; 24%). The families who attended no sessions were classified accurately mostly; 39 of 43 were in the NA class and the remaining four were in the EDO-HI class, as T1 and T2 YSR scores for children in these four families were very high, consistent with the EDO-HI class. Posterior class probabilities for the NA, EDO-LI, EDO-HI, MPDO, SA-LI, and SA-HI were .99, .93, .96, .94, .85 and .87, respectively.

Table 1.

Fit Indices for growth mixture models estimating attendance trajectories of Primary Female Parent s or Caregivers attending the Bridges to High School program for 1–6 class solutions.

No of
Classes		 LL SABIC Quadratic
Adjusted LMR LRT, p
1 −13374.959 26831.821 n/a
2 −4418.454 8897.675 763.637, p < .001
3 −3952.975 8016.915 166.962, p = ns
4 −3890.066 7941.296 26.975, p = ns
5 −3845.812 7905.628 23.385, p = ns
6 −3802.441 7869.085 26.785, p = ns
7 n/a n/a n/a
Open in a new tab

Note. LL = Log Likelihood; SABIC = Sample=adjusted Bayesian Information Criterion; LMR LRT = Lo-Mendell-Rueben Likelihood Ratio Test. We were unable to replicate the LL in the 7 – class solution, despite using 10,000 sets of random starting values. Thirteen percent (n=43) of the 335 female caregivers in this study did not attend any sessions based on observed data; in every k-class solution, except the 1-class solution, we modeled a "Non-attendance" class (i.e., class 1) by constraining growth factors in this class to zero to account for this pattern of intervention nonparticipation.

Figure 3.

Figure 3

Open in a new tab

Graphs for GMM results for 2, 3, 4, 5, and 6-class solutions

Note: NA = Non-Attenders, ET = Early Dropouts, MPDO = Mid-Program Dropouts, SA = Sustained attenders. LI = Low Internalizing, HI = High Internalizing. Horizontal axis = session number, Vertical axis = probability of session attendance. Thirteen percent (n=43) of the 335 female caregivers in this study did not attend any sessions based on observed data; in every k-class solution, except the 1-class solution, we modeled a "Non-attendance" class (i.e., class 1) by constraining growth factors in this class to zero to account for this pattern of non-participation. We employed no constraints in the 1-class solution. In the 6-class solution, 4 of the 43 female caregivers with observed zeros were classified in the EDO-HI class because, although their attendance was consistent with NAs, their scores on predictors of class membership and on T1 and T2 child report of internalizing were more consistent with the EDO-HI class.

Trajectory Subgroup Differences on Pre-to-Post Change in Child Report of Internalizing

Figure 4 shows mean levels of child internalizing (i.e., YSR) at T1 and T2 for each trajectory group. All groups except EDO-HI showed significant pre-to-post change on the YSR: NA: b = −10.12(SE=2.19), FDR-p (pfdr) < .001; EDO-HI: b = 1.26(2.31), ns; EDO-LI: b = −9.91 (1.68), pfdr < .001; MPDO: b = −9.11(0.87) pfdr < .001; SA-LI : b = −5.90(1.05), pfdr < .001; SA-HI: b = −2.96(1.29), pfdr < .05. Pre-to-post change for the EDO-LI and MPDO groups was greater than both SA groups [EDO-LI vs. SA-LI: b = −4.01(1.82), pfdr < .05; EDO-LI vs. SA-HI: b = −6.95(2.11), pfdr < .01; MPDO vs. SA-LI: b = −3.24(1.27), pfdr < .05; MPDO vs SA-HI: b = −6.18(1.62), pfdr < .001)]. The EDO-LI and MPDO groups did not differ from each other [(b = −0.76(2.16)]. The NA group did not differ from the MPDO [b = 0.98(2.33)] or EDO-LI [b = 0.21(2.32)] groups. The two SA groups did not differ from each other [b = 2.94(1.67)].

Figure 4.

Figure 4

Open in a new tab

Mean levels of child self-report of child internalizing at Time 1 and Time 2

Note: NA = Non-Attenders, ET-LI = Early Dropouts-Low Internalizing, ET-HI = Early Dropouts-High Internalizing, MPDO = Mid-Program Dropouts, SA-LI = Sustained Attenders-.Low Internalizing, SA-HI = Sustained Attenders-High Internalizing. Time 1 = Pretest; Time 2 = 2 years post-intervention.

Covariate Associations with Trajectory Subgroup Membership

Table 2, available online, presents logistic coefficients, Odds Ratio, and corresponding adjusted pfdr value for all contrasts and specifies contrasts conducted to test each hypothesis. For hypothesis 1, contrasts tested that higher levels of perceived need would increase the likelihood of membership in the sustained attendance (SA-LI and SA-HI) and variable attendance (MPDO) classes in comparison to early dropout (EDO-LI and EDO-HI) and nonattendance (NA) classes. There were no significant contrasts for hypothesis 1. For hypothesis 2, contrasts tested more self-efficacy would be associated with increased likelihood of membership in the sustained attendance classes (SA-LI and SA-HI) in comparison to the early dropout (EDO-LI and EDOHI) and NA classes. Inconsistent with hypothesis 2, the SA-HI group had less parenting efficacy than did the EDO-HI class. For hypothesis 3, contrasts tested that more participation barriers would be associated with decreased likelihood of membership in the: a) two SA classes in comparison to all other groups, b) MPDO class in comparison to EDO and NA classes, and c) EDO classes compared to NA class. Consistent with hypothesis 3, the EDO-HI group reported more stress than the SA-LI group and the EDO-LI group reported more stress than the NA group. For hypothesis 4, contrasts tested that stronger familism values and lower acculturation would be associated with increased likelihood of membership in the SA classes, in comparison to all other classes. Consistent with hypothesis 4, the SA-HI group had stronger familism values than the EDO-HI and MPDO groups and both SA groups were lower on acculturation than the EDO-LI group. The SA-HI group was also lower on acculturation than the MPDO group. Inconsistent with hypothesis 4, the EDO-LI group had stronger familism than the SA-LI groups; and the EDO-HI group was lower on acculturation than all groups.

Discussion

We used GMM to identify attendance trajectory profiles among MA families randomly assigned to Bridges during the program’s efficacy trial and to examine if profiles differed on parent, child, family, and cultural covariates and on pre-to-post change on child reported internalizing symptoms. Results supported four hypothesized trajectory profile groups-sustained attenders, mid-program dropouts (i.e., variable attenders), early dropouts, and non-attenders-that were further distinguished by child self-report of internalizing. Specifically, both the sustained attenders and early dropouts each separated into two subgroups, one with low internalizing and one with high at pre and post-intervention (see Figure 3, Graphs 4 and 5 and Figure 4), resulting in six trajectory groups: Non-Attenders (NA), Early Dropouts-Low Internalizing (EDO-LI), Early Dropouts-High Internalizing (EDO-HI), Mid-Program Dropouts (MPDO), Sustained Attenders-Low Internalizing (SA-LI), and Sustained Attenders-High Internalizing (SA-HI).

Trajectory Group Differences on Pre-to-Post Change in Child Report of Internalizing

All attendance trajectory groups, except the EDO-HI, showed reductions in internalizing from pretest in 7th grade to the post-test assessment at roughly grade 9. The EDO-HI group showed increasing levels of internalizing, although this change was not significant. However, results did not support our hypothesis that attendance trajectory groups with higher dosage would evidence greater pre-to-post change. In fact, our results showed that pre-to-post differences were greater for some trajectory groups receiving less dosage. For example, change was greater for the NA and the EDO-LI subgroups, compared to both SA groups. These results suggest pre-to-post differences between the trajectory groups on internalizing may reflect pre-existing differences or be due to normal developmental changes. However, child self-report of internalizing did discriminate between the two sustained attender and the two early dropout classes. Child report of internalizing for one of the two sustained attender and for one of the two early dropout groups was high whereas for the other it was low, showing that groups with similar attendance patterns can vary on child internalizing.

Covariate Associations with Trajectory Group Membership

Among families with children reporting high internalizing, those who dropped out early had higher levels of stress than did sustained attenders. In addition, NAs reported more stress than the EDO-LI group. This study offered no other support for the HBM. For example, perceived need, assessed as parent report of child functioning and parenting skills, had no association with attendance trajectories. Also contrary to expectations, parents with more parenting efficacy were more likely to be early dropouts rather than sustained attenders, when these groups had children who reported high internalizing. Perhaps, for these parents, efficacy is more consistent with perceived need. Parents who felt less efficacious perceived a greater need for Bridges and were thus more likely to participate than parents who felt efficacious.

Culture discriminated between attendance trajectory profiles. Families with sustained attendance had stronger familism values than early dropouts, among families with children reporting high internalizing at baseline. This finding supported our hypothesis that Bridges’ emphasis on family as a source of strength to help children succeed motivates participation, at least among families with children self-reporting high internalizing. However, among families with children reporting low baseline internalizing, the influence of familism was contrary to hypotheses; the early dropouts reported stronger familism values than the sustained attenders. Consistent with hypotheses, both SA groups were less acculturated than the EDO-LI group. These results suggest that among families with children reporting low internalizing at baseline, early dropouts are more acculturated and have stronger familism values than sustained attenders. Inconsistent with hypotheses, the least acculturated group, which also had the high child internalizing scores at baseline that increased pre-to-post, dropped out very early. These results suggest the influence of culture on participation is nuanced and may depend on child baseline symptomatology.

Limitations

Attendance assessed in the context of an efficacy trial may limit generalizability to real-world delivery settings. In the Bridges efficacy trial, we used childcare, transportation, and weekly reminder calls to maximize participation. When Bridges goes-to-scale it is unlikely that these strategies will be maintained; as such, attendance patterns in scale-up may also not parallel those that emerged in this study. These participation-incentivizing strategies may have also minimized the effects of logistical participation barriers in our study, obscuring links between participation and these barriers. To ensure generalizability, future research should replicate this study with data from a universal intervention in a real-world delivery setting. Another limitation is that this study’s analyses (i.e., no control group and no random assignment of attendance) precludes interpretation of between-trajectory group differences on pre-to-post change on our outcome as intervention effects, as these differences may reflect pre-existing differences or be attributable to the level of symptoms at baseline.

Conclusions

Our results supported multiple attendance trajectory profiles but profiles reflecting more attendance did not have greater pre-to-post change on child self-report of internalizing as expected. Nonetheless, internalizing did differentiate two sustained attenders subgroups and two dropout subgroups, with the two groups varying on levels (i.e., high vs. low) of internalizing. These results suggest heterogeneity among families with similar attendance patterns and highlight the importance of modeling this heterogeneity in studies that examine correlates of participation. Cultural covariates played a more prominent role than traditional HBM constructs in discriminating between attendance trajectory profiles. Results showed that an extension of HBM that includes cultural constructs is potentially critical to culturally diverse populations and that the influence of cultural variables on participation is complex and may depend on child baseline symptomatology. For example, results supported distinct groups of low-acculturated parents, with some inclined to participate and others much less likely to participate, dependent on child symptomatology. Strategies that use personalized calls and other personal contact methods to engage parents increase participation (McKay et. al., 2004). Understanding attendance trajectory profiles will help identify families vulnerable to dropout who should be targeted for these strategies and inform timing and allocation of resources to incentivize participation families most likely to dropout.

Supplementary Material

11121_2016_632_MOESM1_ESM

Acknowledgments

Funding: Development and evaluation of the Bridges to High School program (Bridges), including the data collected and used in this study, were supported by the National Institute of Mental Health grant R01 MH64707.

Drs. Gonzales and Dumka are the developers of the Bridges program; Dr. Mauricio was involved in implementation of the Bridges program during its efficacy trial; Drs. Tein and Millsap were involved in the evaluation of the efficacy of the Bridges program.

Footnotes

Compliance with Ethical Standards

Informed consent: Informed consent was obtained from all participants in this study and assent was obtained from minors included in the study.

Disclosure of potential conflicts of interest: The authors declare that they have no other conflicts of interest.

Ethical Approval: All study procedures and measures were reviewed and approved by the Arizona State University Institutional Review Board. All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.

References

  1. Achenbach TM. Child behavior checklist/4–18. Burlington: University of Vermont; 1991. [Google Scholar]
  2. Asparouhov T, Muthén B. Using Mplus TECH11 and TECH14 to test the number of latent classes. 2012 [Google Scholar]
  3. Baker CN, Arnold DH, Meagher S. Enrollment and attendance in a parent training prevention program for conduct problems. Prevention Science. 2011;12(2):126–138. doi: 10.1007/s11121-010-0187-0. [DOI] [PubMed] [Google Scholar]
  4. Carpentier FRD, Mauricio AM, Gonzales NA, Millsap RE, Meza CM, Dumka LE, German M, Genalo MT. Engaging Mexican origin families in a school-based preventive intervention. The Journal of Primary Prevention. 2007;28:521–546. doi: 10.1007/s10935-007-0110-z. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Coatsworth JD, Duncan LG, Pantin H, Szapocznik J. Patterns of retention in a preventive intervention with ethnic minority families. Journal of Primary Prevention. 2006;27:171–193. doi: 10.1007/s10935-005-0028-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Cuellar I, Arnold B, Maldonado R. Acculturation rating scale for Mexican Americans-II: A revision of the original ARSMA scale. Hispanic journal of behavioral sciences. 1995;17:275–304. [Google Scholar]
  7. Dishion TJ, McMahon RJ. Parental monitoring and the prevention of child and adolescent problem behavior: A conceptual and empirical formulation. Clinical child and family psychology review. 1998;1(1):61–75. doi: 10.1023/a:1021800432380. [DOI] [PubMed] [Google Scholar]
  8. Dumka LE, Prost J, Barrera M., Jr The parental relationship and adolescent conduct problems in Mexican American and European American families. Journal of Couple & Relationship Therapy. 2002;1:37–57. [Google Scholar]
  9. Enders CK, Bandalos DL. The relative performance of full information maximum likelihood estimation for missing data in structural equation models. Structural Equation Modeling. 2001;8:430–457. [PubMed] [Google Scholar]
  10. Gonzales N, Dumka L, Mauricio A, German M. Building bridges: Strategies to promote academic and psychological resilience for adolescents of Mexican origin. In: Lansford JE, Deater-Deckard K, Bornstein MH, editors. Immigrant Families in Contemporary Society. New York, NY: Guilford Press; 2007. pp. 268–286. 2007. [Google Scholar]
  11. Gonzales NA, Dumka LE, Millsap RE, Gottschall A, McClain DB, Wong JJ, German G, Mauricio AM, Wheeler L, Carpentier FD, Kim SY. Randomized trial of a broad preventive intervention for Mexican American adolescents. Journal of Consulting and Clinical Psychology. 2012;80:1–16. doi: 10.1037/a0026063. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gonzales NA, Wong JJ, Toomey RB, Millsap R, Dumka LE, Mauricio AM. School engagement mediates long term prevention effects for Mexican American adolescents. Prevention Science. 2014;15:929–939. doi: 10.1007/s11121-013-0454-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Gorman-Smith D, Tolan PH, Henry DB, Leventhal A, Schoeny M, Lutovsky K, Quintana E. Predictors of participation in a family-focused preventive intervention for substance use. Psychology of Addictive Behaviors. 2002;16(4S):S55. doi: 10.1037/0893-164x.16.4s.s55. [DOI] [PubMed] [Google Scholar]
  14. Gottfried AE, Fleming JS, Gottfried AW. Role of parental motivational practices in children's academic intrinsic motivation and achievement. Journal of Educational Psychology. 1994;86(1):104. [Google Scholar]
  15. Hill NE, Torres K. Negotiating the American dream: The paradox of aspirations and achievement among Latino students and engagement between their families and schools. Journal of Social Issues. 2010;66(1):95–112. [Google Scholar]
  16. Janz NK, Becker MH. The health belief model: A decade later. Health Education & Behavior. 1984;11:1–47. doi: 10.1177/109019818401100101. [DOI] [PubMed] [Google Scholar]
  17. Kazdin AE, Holland L, Crowley M, Breton S. Barriers to treatment participation scale: Evaluation and validation in the context of child outpatient treatment. Journal of Child Psychology and Psychiatry. 1997;38:1051–1062. doi: 10.1111/j.1469-7610.1997.tb01621.x. [DOI] [PubMed] [Google Scholar]
  18. Kazdin AE, Mazurick JL. Dropping out of child psychotherapy: Distinguishing early and late dropouts over the course of treatment. Journal of Consulting and Clinical Psychology. 1994;62(5):1069. doi: 10.1037//0022-006x.62.5.1069. [DOI] [PubMed] [Google Scholar]
  19. Knight GP, Gonzales NA, Saenz DS, Bonds DD, Germán M, Deardorff J, Roosa MW, Updegraff KA. The Mexican American cultural values scale for adolescents and adults. The Journal of Early Adolescence. 2010;30:444–481. doi: 10.1177/0272431609338178. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kumpfer KL, Alvarado R, Smith P, Bellamy N. Cultural sensitivity and adaptation in family-based prevention interventions. Prevention Science. 2002;3(3):241–246. doi: 10.1023/a:1019902902119. [DOI] [PubMed] [Google Scholar]
  21. Lo Y, Mendell NR, Rubin DB. Testing the number of components in a normal mixture. Biometrika. 2001;88:767–778. [Google Scholar]
  22. McKay MM, Bannon WM., Jr Engaging families in child mental health services. Child and adolescent psychiatric clinics of North America. 2004;13(4):905–921. doi: 10.1016/j.chc.2004.04.001. [DOI] [PubMed] [Google Scholar]
  23. McKay MM, Hibbert R, Hoagwood K, Rodriguez J, Murray L, Legerski J, Fernandez D. Integrating evidence-based engagement interventions into "real world" child mental health settings. Brief Treatment and Crisis Intervention. 2004;4(2):177. [Google Scholar]
  24. Merikangas KR, He JP, Burstein M, Swanson SA, Avenevoli S, Cui L, Swendsen J. Lifetime prevalence of mental disorders in US adolescents: results from the National Comorbidity Survey Replication–Adolescent Supplement (NCS-A) Journal of the American Academy of Child & Adolescent Psychiatry. 2010;49:980–989. doi: 10.1016/j.jaac.2010.05.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Muthén BO. Beyond SEM: General latent variable modeling. Behaviormetrika. 2002;29:81–118. (1; ISSU 51) [Google Scholar]
  26. Muthén B. Latent variable analysis: Growth mixture modeling and related techniques for longitudinal data. In: Kaplan D, editor. Handbook of quantitative methodology for the social sciences. Newbury Park, CA: Sage; 2004. pp. 345–368. [Google Scholar]
  27. Muthén LK, Muthén BO. Mplus User’s Guide. 7th. Los Angeles: CA: Muthén and Muthén; 1998–2013. [Google Scholar]
  28. NRC/IOM. Preventing mental, emotional, and behavioral disorders among young people: Progress and possibilities. Washington, DC: The National Academy Press; 2009. 2009. [PubMed] [Google Scholar]
  29. Perrino T, Coatsworth JD, Briones E, Pantin H, Szapocznik J. Initial engagement in parent-centered preventive interventions: A family systems perspective. Journal of Primary Prevention. 2001;22:21–44. [Google Scholar]
  30. Pew Hispanic Center. Pew Hispanic Center Fact Sheet: Statistical portrait of Hispanics in the United States 2006. Los Angeles: University of Southern California Annenberg School for Communication; 2006. http://pewhispanic.org/files/other/middecade/Table-3.pdf. [Google Scholar]
  31. Radloff LS. The CES-D scale a self-report depression scale for research in the general population. Applied psychological measurement. 1977;1:385–401. [Google Scholar]
  32. Reid MJ, Webster-Stratton C, Baydar N. Halting the development of conduct problems in Head Start children: The effects of parent training. Journal of Clinical Child and Adolescent Psychology. 2004;33(2):279–291. doi: 10.1207/s15374424jccp3302_10. [DOI] [PubMed] [Google Scholar]
  33. Reyno SM, McGrath PJ. Predictors of parent training efficacy for child externalizing behavior problems–a meta-analytic review. Journal of Child Psychology and Psychiatry. 2006;47:99–111. doi: 10.1111/j.1469-7610.2005.01544.x. [DOI] [PubMed] [Google Scholar]
  34. Rosenstock Irwin M, Strecher Victor J, Becker Marshall H. Social learning theory and the health belief model. Health Education & Behavior. 1988;15(2):175–183. doi: 10.1177/109019818801500203. [DOI] [PubMed] [Google Scholar]
  35. Schaeffer CM, Petras H, Ialongo N, Masyn KE, Hubbard S, Poduska J, Kellam S. A comparison of girls' and boys' aggressive-disruptive behavior trajectories across elementary school: prediction to young adult antisocial outcomes. Journal of Consulting and Clinical Psychology. 2006;74(3):500. doi: 10.1037/0022-006X.74.3.500. [DOI] [PubMed] [Google Scholar]
  36. Sclove SL. Application of model-selection criteria to some problems in multivariate analysis. Psychometrika. 1987;52:333–343. [Google Scholar]
  37. Small SA, Kerns D. Unwanted sexual activity among peers during early and middle adolescence: Incidence and risk factors. Journal of Marriage and the Family. 1993:941–952. [Google Scholar]
  38. Spoth R, Redmond C. Parent motivation to enroll in parenting skills programs: A model of family context and health belief predictors. Journal of Family Psychology. 1995;9:294–310. [Google Scholar]
  39. Szapocznik J, Coatsworth JD. An ecodevelopmental framework for organizing the influences on drug abuse: A developmental model of risk and protection. 1999 [Google Scholar]
  40. U.S. Department of Education. Dropout rates in the United States: 1998. Washington DC: U.S. Government Printing Office; 2000. (NCES 2000-022) [Google Scholar]
  41. Wang PS, Simon G, Kessler RC. The economic burden of depression and the cost-effectiveness of treatment. International journal of methods in psychiatric research. 2003;12(1):22–33. doi: 10.1002/mpr.139. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

11121_2016_632_MOESM1_ESM
NIHMS753150-supplement-11121_2016_632_MOESM1_ESM.doc (209KB, doc)

RESOURCES