A Comprehensive Meta-Analysis of Triple P-Positive Parenting Program Using Hierarchical Linear Modeling: Effectiveness and Moderating Variables

Abstract

A meta-analysis encompassing all studies evaluating the impact of the Triple P-Positive Parenting Program on parent and child outcome measures was conducted in an effort to identify variables that moderate the program’s effectiveness. Hierarchical linear models (HLM) with three levels of data were employed to analyze effect sizes. The results (N = 55 studies) indicate that Triple P causes positive changes in parenting skills, child problem behavior and parental well-being in the small to moderate range, varying as a function of the intensity of the intervention. The most salient findings of variables moderating the interventions’ impact were larger effects found on parent report as compared to observational measures and more improvement associated with more intensive formats and initially more distressed families. Sample characteristics (e.g., child’s age, being a boy) and methodological features (e.g., study quality) exhibited different degrees of predictive power. The analysis clearly identified several strengths of the Triple P system, most importantly its ability to effect meaningful improvement in parents and children. Some limitations pertain to the small evidence-base of certain formats of Triple P and the lack of follow-up data beyond 3 years after the intervention. Many of the present findings may be relevant to other evidence-based parenting programs.

Appendix Formulae

Between-groups effect size:

$$ s_{\rm pooled}=\sqrt{\frac{(n_{\rm T}-1)s_T^2+(n_{\rm C}-1)s_{\rm C}^2}{n_{\rm T} + n_{\rm C} - 2}}\quad g_{\rm Hedges} = \frac{x_{\rm T}-x_{\rm c}}{s_{\rm pooled}} $$

$$ d=g_{\rm Hedges}\left(1-\frac{3}{4(n_{\rm T} + n_{\rm C}-2)-1}\right) \quad s_{\rm d}^2=\frac{n_{\rm T} + n_{\rm C}}{n_{\rm T} n_{\rm C}}+\frac{d^2}{2(n_{\rm T} + n_{\rm C})} $$

where s _T, s _C = post-intervention standard deviation of intervention and control group, respectively; n _T, n _C = post-intervention sample size; and x _T, x _C = post-intervention mean outcome score.

Within-group effect size:

$$ s_{\rm Diff}=\sqrt{s_{\rm Pre}^2 + s_{\rm Post}^2 - 2r_{{\rm Pre, Post}} s_{\rm Pre}s_{\rm Post}} \quad d_{\rm Within} = \frac{x_{\rm Pre}-x_{\rm Post}}{s_{\rm Diff}}$$

$$ d= d_{\rm Within}\left(1- \frac{3}{4(n_{{\rm Pre}-1)-1}}\right) \quad s_{\rm d}^2 = \frac{1}{n} + \frac{d^2}{2(n_{\rm Pre}-1)} $$

Correlation estimation between pre-post/pre-follow-up time points:

$$ t=\sqrt{F} $$

$$ {\rm SD}_{\rm ChangeScore}^2 = \frac{n\left({\rm Mean}_{\rm post}-{\rm Mean}_{\rm Pre}\right)^2}{t^2} $$

$$ r=\frac{\left({\rm SD}_{{\rm Pre}}^2 + {\rm SD}_{\rm Post}^2 - {\rm SD}_{\rm ChangeScore}^2\right)}{2 {\rm SD}_{\rm Pre}{\rm SD}_{\rm Post}} $$

Based on studies: 2, 3, 6, 7, 8, 13, 26, 39, 40, 60

Hierarchical Linear Model:

• Unconditional model:

  • Level 1: Within-studies

    $$ d_{jk}=\delta_{jk}+e_{jk} $$

  • Level 2: Between-intervention groups (nested within studies)

    $$ \delta_{jk}=\beta_{0k}+r_{jk} $$

  • Level 3: between-studies

    $$ \beta_{0k}=\gamma_{oo}+u_{0k} $$

• Conditional model:

  • Level 1: within-studies

    $$ d_{jk}=\delta_{jk}+e_{jk} $$

  • Level 2: Between-intervention groups (nested within studies)

    $$ \delta_{jk}=\beta_{0k}+\sum\limits_{p}\beta_pW_{pk}+r_{jk} $$

  • Level 3: between-studies

    $$ \beta_{0k}=\gamma_{oo}+\sum\limits_{q}\gamma_qx_{qk}+u_{0k} $$

    Intervention groups j = 1,…J within studies k = 1,…K e _jk  = sampling error associated with d _jk ; e _jk ∼N(0;V _jk ) r _jk  = random error at Level 2; r _jk ∼ N(0;τ_δ) u _ok  = random error at Level 3; u _ok ∼ N(0;τ_β)