What We Are Learning About Fade-Out of Intervention Effects: A Commentary

When designing intervention research that has a longterm goal, fade-out is an important consideration. Bailey, Duncan, Cunha, Foorman, and Yeager (2020; this issue) offer several important takeaways for such interventions, beginning from the initial plan to later longitudinal analyses of treatment effects. For example, researchers would be well advised to consider the contextual influences, such as whether the treatment is in a low-income urban school district undergoing pending gentrification plans at the onset of the intervention, which could change the demographic characteristics of the targeted student population. Gentrification of a neighborhood may have profound implications for the initial sample selection, instrumentation, and measurement. The authors also suggest that intervention designers with long-term goals need to request additional support for subsequent data-collection efforts. We assume this would include such factors as obtaining overpowered initial treatment and control samples, identifying stable contextual conditions (e.g., neighborhood, student and teacher mobility), and a clear temporal vision of subsequent treatment outcomes, all of which are likely to affect the sample balance necessary for evaluating the impact of the intervention over time. Although Bailey and colleagues are comprehensive in their focus on fade-out and possible remediation of its effect, we argue that the dominance of the psychological perspective on education interventions and their purposes tend to overlook other research designs in which problems of fade-out can more easily be adjusted (e.g., quasiexperiments with generalizable longitudinal samples that include nested interventions) or other naturally occurring treatment effects (e.g., use of online instruction during a pandemic). The authors focus on interventions designed to enhance psychological traits or skill-based tools and bring in other research in economics and sociology that they perceive as complementary to their perspective. Our review highlights some additional problems of designing interventions involving randomized controlled trials (RCTs) that specifically focus on avoiding fade-out and recognize the complexity of measures required to understand persisting effects of an intervention on either psychological traits or skill-based tools. In addition, we put forward several measurement issues that arise when considering postintervention analyses for RCTs or quasiexperiments.