Notes on Attendance Patterns in SFP 10-14 (Latent Class Analysis)

Study overview

• Topic: Predictors of attendance patterns in a universal family-based preventive intervention (SFP 10-14).
• Objective: Identify latent classes of attendance across seven program sessions and examine individual, family, and sociodemographic predictors of class membership to inform retention strategies.
• Program studied: Strengthening Families Program: For Parents and Youth Ages 10-14 (SFP 10-14).
• Design context: Data from a community-randomized PROSPER trial across 28 rural/semi-rural districts in PA and IA; intervention condition offered SFP 10-14 after Fall Grade 6 assessment.
• Analytic sample: Families where adolescent and parent attended at least one SFP 10-14 session. Final analytic sample: N = 957.
• Sessions: Seven weekly two-hour sessions with dinner and childcare; recruitment through multiple channels (classroom presentations, newsletters, conferences, invitations).
• Key aim: Distinguish patterns of attendance (beyond a simple high/low dichotomy) and identify predictors for each pattern to target retention efforts.

Context and theoretical framing

• Attendance is multi-faceted and may be influenced by:
  • Individual factors (adolescent behaviors, attitudes, and peer context).
  • Family factors (family climate, parental knowledge, family engagement).
  • Sociodemographic factors (income, single-parent status).
• The Health Belief Model (Rosenstock, 1966) is used to frame participation decisions, with four core perceived determinants:
  • Susceptibility of child to problem
  • Severity of problem
  • Perceived benefits of the program
  • Perceived barriers to participation
• Prior literature shows divergent findings: higher initial engagement sometimes linked to more problem behaviors; practical barriers (scheduling, transportation) disproportionately affect low-income or single-parent families.
• Prior attendance trajectory work (Mauricio et al.; St. George et al.) identified common patterns: early dropouts, mid-program dropouts/declining, sustained attenders; our study adds nuance by identifying two early-dropout subgroups.

Sample and measures

• Participants and procedures
  • Data from PROSPER trial; two cohorts of Grade 6 students (beginning 2002–2003) followed yearly through Grade 12.
  • Fall Grade 6 self-report assessment (n ≈ 10,845 students; ~90% of eligible).
  • Intervention condition: n = 5,515 offered SFP 10-14; about 17 ext{%} enrolled.
  • Analytic sample: N = 957 families with at least one attended session.
• Measures by domain
  • Attendance: attendance of one adolescent and one parent together at each of the seven sessions.
  • Individual factors (Fall Grade 6):
  • Conduct problem behaviors; 12 items; scale 1 (Never)–5 (5+ times); reliability
    {alpha} = 0.84.
  • Deviant peer involvement; 3 items; scale 1–5; reliability
    {alpha} = 0.81.
  • Negative school adjustment and bonding; 5 items; scale 1–5; reliability
    {alpha} = 0.57.
  • Family factors:
  • Family climate (Family Environment Scale); 1–5; higher scores indicate more cohesion/organization, less conflict; reliability
    {alpha} = 0.73.
  • Parental knowledge (child monitoring subscale of General Child Management Scale); 4 items; reliability
    {alpha} = 0.74.
  • Adolescent positive engagement in the family (AQRS); 3 items; 1–5; reliability
    {alpha} = 0.90.
  • Sociodemographic factors:
  • Single-parent status (Yes/No).
  • Low income status: proxy for free/reduced-price lunch eligibility (Yes/No).
• Descriptive baseline comparisons
  • Those who attended SFP 10-14 more likely not to be single parents (group differences reported in descriptive analyses).

Analytical approach

• Primary method: Repeated measures latent class analysis (RMLCA) to identify latent attendance patterns across the 7 sessions.
  • RMLCA identifies latent classes based on patterns of categorical changes in attendance across time (at least 3 time points).
  • Model selection criteria used: AIC, BIC, sample-size adjusted BIC (a-BIC), entropy, and bootstrap likelihood ratio test (BLRT); assessed model stability and interpretability.
  • Maximum likelihood estimation with 1,000 initial starts and 500 final starts; missing data handled with full-information ML.
  • Clustering by community was accounted for in standard errors.
• Model fitting results
  • Tested models with 1–6 classes.
  • AIC minimized at the 5-class solution; BIC minimized at the 3-class solution; a-BIC minimized at the 4-class solution.
  • Entropy ranged from 0.70 to 0.93; larger models tended to have lower entropy (in the 0.70s).
  • BLRT became insignificant at the 6-class model, suggesting no additional information beyond 5 classes.
  • Examination of the 5-class solution indicated a non-distinct class (overextraction).
  • Final model chosen: 4-class solution due to being stable, interpretable, and theoretically meaningful.
• Class assignment and predictor analysis
  • Class membership probabilities were estimated; modal class assignment with classification-error correction (R3STEP in Mplus) used to relate predictors to latent class membership.
  • Final predictor analysis included domain-specific predictors that showed significant pairwise differences across classes in separate analyses.
  • Domains considered: individual, family, and sociodemographic factors.
  • All analyses adjusted for clustering at the community level; analysis conducted in Mplus, v8.4.

Results: Latent classes of attendance

• Four distinct classes emerged (prevalence in parentheses):
  • Class 1: First-Session Attenders (7%).
  • High probability of attending Session 1; near-zero probabilities for later sessions (e.g., Session 2 ≈ 0.00, Session 3 ≈ 0.13).
  • Class 2: Early Attenders (9%).
  • Moderate-to-high probability of attending Sessions 1–2; declining probabilities for later sessions.
  • Class 3: Declining-High Attenders (18%).
  • High probabilities for the first four sessions; then moderate to declining probabilities for Sessions 5–7.
  • Class 4: Consistent-High Attenders (66%).
  • High probabilities of attending all seven sessions.
• Visual depiction: Figure 1 shows the four-class trajectory patterns (not reproduced here).
• Summary: These patterns align with prior work identifying early dropouts, mid-program dropouts, and sustained attenders, but with two distinct early-dropout groups in this sample.

Predictors of class membership (domain-specific and final model)

• Individual factors (from initial analyses):
  • Significant pairwise differences for adolescent deviant peer involvement and negative school adjustment/bonding; trend for conduct problems. These were retained in the final model.
• Family factors:
  • Pairwise differences significant for family climate; retained in the final model.
• Sociodemographic factors:
  • Pairwise differences significant for low-income status; retained in final model.
• Final multinomial predictor results (class membership relative to the reference class: Consistent-High Attenders)
  • Key predictors retained in the final model: negative school adjustment and bonding, deviant peer involvement, and low-income status.
  • Deviant peer involvement
  • Higher adolescent deviant peer involvement increased the odds of being Early Attenders versus Consistent-High Attenders and Declining-High Attenders (vs Consistent-High Attenders).
  • In numerical terms, Early Attenders showed a significant positive association with deviant peer involvement (OR ≈ 1.74 per unit increase, p < .05).
  • Negative school adjustment and bonding
  • Higher negative school adjustment and bonding increased the likelihood of being Declining-High Attenders relative to other patterns, and reduced the likelihood of being Early Attenders relative to Consistent-High Attenders.
  • For Declining-High Attenders vs Consistent-High Attenders, the effect was positive (OR > 1; e.g., OR ≈ 1.41 per unit increase in the negative school adjustment/bonding scale).
  • For First-Session Attenders vs Consistent-High Attenders, the effect was less favorable (reflected as OR < 1 for the corresponding comparison).
  • Low-income status
  • Low-income families were more likely to be First-Session Attenders than Consistent-High Attenders (OR ≈ 2.62; p < .05).
• Interpretive takeaways from the final model
  • Adolescent deviant peer involvement is a salient risk marker for early but not sustained attendance (i.e., pushes toward Early Attenders pattern).
  • Negative school adjustment/bonding signals risk for mid-to-late dropout patterns (Declining-High Attenders) and reduces likelihood of consistent engagement during early sessions.
  • Low-income status is linked to an initial attempt to engage (First-Session Attenders), but these families face higher barriers to sustained attendance.

Practical implications and strategies for retention

• Tailored retention approaches by pattern subgroup
  • For Early Attenders (and those high in deviant peer involvement):
  • Provide a warmer welcome and enhanced reinforcement around Session 2, which is often their first or only subsequent session.
  • Consider brief, targeted interventions at outset to address barriers and perceived susceptibility/benefits (Health Belief Model framework).
  • Leverage Consistent-High Attenders as peer mentors or advocates to encourage ongoing attendance.
  • For Declining-High Attenders (those with negative school bonding):
  • Integrate school-related elements into sessions (e.g., invite teachers, align with school-based activities) to bolster school bonding and relevance.
  • Offer additional school-related supports outside of session time to sustain engagement.
  • For First-Session Attenders (low-income or initially engaged despite barriers):
  • Proactively reduce barriers early (e.g., flexible scheduling, childcare, transportation assistance, small incentives like cash or gift cards).
  • Provide brief verbal motivational interventions at program onset to reinforce continued participation.
  • Cross-cutting strategies:
  • Early check-ins with families after Session 1 and before Session 2 to address emergent barriers.
  • Use brief overviews of material from prior sessions to reduce fidelity concerns while bridging gaps for those missing sessions.
  • Consider social-network-based engagement strategies (e.g., involving other parents in the Consistent-High Attenders group).
• Practical relevance and real-world impact
  • Findings highlight that retention efforts must be subgroup-specific rather than one-size-fits-all.
  • The study points to actionable entry points for program implementers to improve overall program impact by increasing retention in subgroups at greater risk for dropout.

Connections to literature and broader implications

• Consistent with prior work identifying three main attendance trajectories (early dropouts, mid-program dropouts, sustained attenders) across family-based programs (Mauricio et al.; St. George et al.).
• Adds nuance by revealing two distinct early-dropout subgroups (First-Session Attenders and Early Attenders) in a large universal preventive program.
• Supports ongoing use of latent-class approaches to understand heterogeneity in attendance and tailor implementation strategies accordingly.
• The Health Belief Model remains a useful framework for understanding and addressing barriers and facilitators to ongoing engagement in preventive interventions.

Reliability, limitations, and methodological notes

• Limitations acknowledged by authors:
  • Inability to run multilevel (community-level) predictors due to statistical power limits.
  • Attendance change processes between sessions were not analyzed (no session-by-session predictors beyond baseline).
  • The negative school adjustment and bonding measure had relatively low reliability (
    {alpha} = 0.57), warranting cautious interpretation.
  • Data are from 2002–2003; SFP 10-14 delivery was similar thereafter, but generalizability to other contexts or time periods may be constrained.
  • Sample composition: rural/semi-rural sites with predominantly White families; generalizability to more diverse populations may be limited.
• Strengths to note:
  • Large sample size for a latent class analysis of attendance patterns (N = 957).
  • Longitudinal design with seven sessions tracked, allowing nuanced pattern detection.
  • Use of robust statistical methods: Repeated measures latent class analysis (RMLCA), three-step predictor analysis (R3STEP), full-information ML for missing data, and clustering adjustment by community.

Key numerical and statistical references (high-level)

• Sample and sessions
  • Analytic sample: N = 957 families; attendance across 7 sessions.
  • Prevalence: First-Session Attenders 7 ext{%}; Early Attenders 9 ext{%}; Declining-High Attenders 18 ext{%}; Consistent-High Attenders 66 ext{%}.
• Model comparison and selection (latent classes)
  • Model fit indices used: AIC, BIC, a-BIC, entropy, BLRT; comparisons across 1–6 classes.
  • Final selection: a 4-class solution chosen for interpretability and stability; 5-class showed a non-distinct class; 6-class did not improve fit (BLRT insignificant).
  • Class-level attendance probabilities (Table 3):
  • First-Session Attenders: Session1 ≈ 0.99; Session2 ≈ 0.00; Session3 ≈ 0.13; Session4 ≈ 0.00; Session5 ≈ 0.01; Session6 ≈ 0.00; Session7 ≈ 0.04.
  • Early Attenders: Session1 ≈ 0.52; Session2 ≈ 0.96; Session3 ≈ 0.31; Session4 ≈ 0.00; Session5 ≈ 0.03; Session6 ≈ 0.05; Session7 ≈ 0.00.
  • Declining-High Attenders: Session1 ≈ 0.80; Session2 ≈ 0.85; Session3 ≈ 0.85; Session4 ≈ 0.77; Session5 ≈ 0.59; Session6 ≈ 0.14; Session7 ≈ 0.43.
  • Consistent-High Attenders: Session1 ≈ 0.90; Session2 ≈ 0.89; Session3 ≈ 0.89; Session4 ≈ 0.90; Session5 ≈ 0.88; Session6 ≈ 0.96; Session7 ≈ 0.94.
• Predictors (final model, relative to Consistent-High Attenders)
  • Deviant peer involvement: Early Attenders vs Consistent-High Attenders (OR ≈ 1.74 per unit increase; p < .05).
  • Negative school adjustment and bonding: Declining-High Attenders vs Consistent-High Attenders (OR > 1 per unit increase; p < .05); Early Attenders vs Consistent-High Attenders less likely with higher negative school adjustment (pattern-dependent effects).
  • Low-income status: First-Session Attenders vs Consistent-High Attenders (OR ≈ 2.62; p < .05).
• Practical takeaway: These numerical patterns highlight subgroup-specific risks and targets for retention strategies in multi-session family-based programs.

Ethical and reporting notes

• Compliance with ethical standards: Authors declare no conflict of interest.
• Data are from an Author Manuscript version of a peer-reviewed article; the final published version may differ in copyediting and content.
• Corroborating sources and references provided span foundational work on preventive interventions, engagement, and latent class methodologies (e.g., Spoth et al., Mauricio et al., St. George et al.).

Summary takeaways for exam preparation

• Four attendance patterns emerged in SFP 10-14 with prevalences of 7 ext{%}, 9 ext{%}, 18 ext{%}, and 66 ext{%} for First-Session, Early, Declining-High, and Consistent-High Attenders, respectively.
• The best-fitting solution was a 4-class model (despite other indices suggesting alternative minima); this model was stable and interpretable.
• Key predictors of class membership (relative to Consistent-High Attenders) were:
  • Adolescent deviant peer involvement → higher odds of Early Attenders.
  • Negative school adjustment and bonding → higher odds of Declining-High Attenders; lower odds of Early Attenders.
  • Low-income status → higher odds of being First-Session Attenders.
• Practical implications emphasize targeted retention efforts, especially around Session 2 for Early Attenders, school-linked engagement for Declining-High Attenders, and early barriers reduction for First-Session Attenders among low-income families.
• Methodologically, the study exemplifies the use of Repeated Measures Latent Class Analysis (RMLCA), three-step predictor analysis (R3STEP), and multi-domain predictors to understand attendance trajectories in preventive interventions.