Estimating Impacts of the Good Behavior Game with Noncompliance on Teacher Efficacy and Burnout: A Complier Average Causal Effect Application

Introduction: Teachers are often the primary implementers of classroom-based preventive interventions in elementary school, and the degree to which they opt to participate in implementing the intervention varies. Most studies take an intent-to-treat (ITT) approach to estimate the effect of being assigned to the treatment condition. In the case of low teacher compliance, the effect of being assigned to treatment may substantially differ from the effects on those who are assigned and participate and ITT effects may be underestimated (Stuart, Perry, Le, & Ialongo, 2008). The complier-average treatment effect (CACE) provides an estimate of the treatment effect accounting for noncompliance. The current study applied the CACE method to estimate the impacts of a universal, behavioral management and social-emotional learning intervention in elementary schools on teachers’ self-efficacy and burnout.

Method: Data come from 350 teachers in 27 schools involved in a one-year RCT testing the Good Behavior Game (GBG; Embry et al., 2003) versus an integration of GBG with the PATHS (Greenberg & Kusché, 2011) social-emotional curriculum. The teacher-reported outcomes were measured at four time points across the school year and include two measures of efficacy -- behavioral management (Main & Hammond, 2008) and social-emotional learning (Domitrovitch & Poduska, 2008) – and three measures of burnout (emotional exhaustion, depersonalization, personal accomplishment) on the Maslach Burnout Inventory (Maslach et al., 1997). Compliance was defined as teachers who were in the 75^th percentile and above on 2 implementation measures: number of games played and minutes of games played across the school year. We used a maximum likelihood approach to CACE estimation in which latent growth mixture modeling (Mplus 7; Muthén & Muthén, 1998-2012) was used to estimate class membership in the control group from observed class membership in the treatment group (Little & Rubin, 1987).

Results: Briefly, ITT estimates of the effects of the integrated GBG with PATHS component (compared to the control condition) on the slopes of behavioral management and social emotional learning efficacy were .07 (p<.01) and .10 (p<.001), respectively. CACE estimates were .25 (p<.001.) and -.32 (n.s.). The slope estimates for emotional exhaustion, depersonalization, and personal accomplishment were -.02 (n.s.), .02 (n.s.), and .08 (p<.05), respectively. The CACE estimates were .36 (p<.05), .16 (n.s.), and -.11 (n.s.).

 

Discussion: Results suggest that the estimates for compliers were generally stronger than the ITT estimates. The assumptions underlying CACE estimation will be discussed, along with the added value of CACE analysis for understanding impacts of universal, school-based interventions.