Using Multi-Site Multiple Mediator Instrumental Variables to Explore Mediator Effects of Growth Mindset Intervention in the National Study of Learning Mindsets (NSLM)

Gopalan, Maithreyi; Gopalan, Maithreyi

The National Study of Learning Mindsets (NSLM) is a study of a nationally representative sample of 9^th graders from 65 high schools in the US followed longitudinally after the delivery of a randomized control trial aimed at fostering student growth mindset—the belief that intellectual abilities are not fixed but can be developed. While this short, universal, preventative psychological intervention had modest but consequential effects on outcomes such as the grade point average one-year post-intervention, especially for previously low-achieving students (Yeager et al., 2018), the mechanisms through which the positive effect emerges remains unanswered. For example, how does adopting a belief in the malleability of intellectual abilities improve students’ academic performance? Do they adopt behaviors that aid their learning? Are they resilient and persistent in tasks? Do they simply exert more effort or use other strategies?

Past research (mostly lab studies) and theory posits a key role for adaptive and mastery oriented behaviors as an important mediator. Students who endorse growth mindset display behaviors (referred to as growth mindset behaviors hereafter)—such as taking on more “mastery-oriented” tasks, interpret struggles as opportunities to learn, and maintain a positive affective response to setbacks (see Yeager & Dweck, 2019 for a review). Theoretically, the hypothesis is that students who interpret challenges as learning opportunities rather than indications of fixed ability may choose to take on more challenging tasks, be more effortful and persistent, imbue a challenge-seeking ethos thereby experiencing future positive academic outcomes. While theory and past lab studies have measured these mediators only through self-reported measures, we operationalize growth mindset behavior using a novel performance task—referred to as “Make-a-math worksheet” task—that all students (treatment and control students) completed post-intervention. First, we find that students exposed to the growth mindset intervention display higher growth mindset behavior—they choose more challenging tasks (d = 0.22; p < 0.001). However, there is heterogeneity in this effect across the 65 sites. We exploit this effect variation across sites in a multi-site multiple mediator instrumental variable (MSMM-IV) framework to explore the mediating effects of students’ growth mindset behavior (instrumented by the random assignment to condition) on students’ post-intervention grade point average.

Preliminary results based on MSMM-IV analysis shows that the average causal mediated effect of students’ growth mindset behavior is positive and significant providing empirical evidence for the complete causal pathway between mindsets and achievement for the first time. We also use two-stage-least-squares (TSLS) estimation using randomization condition as the instrumental variable to estimate the average mediator effect given that the mediating effects vary across the sites. Finally, we use site-by-randomized condition indicators as multiple instruments to estimate the average effect of the mediator—growth mindset behavior, by further incorporating the heterogeneity across sites on students' average academic outcomes for all students.

Association for Public Policy Analysis & Management

Panel Paper: Using Multi-Site Multiple Mediator Instrumental Variables to Explore Mediator Effects of Growth Mindset Intervention in the National Study of Learning Mindsets (NSLM)