Association for Public Policy Analysis & Management

This paper examines the first large-scale panel survey of students on social-emotional learning (SEL) and asks two questions: (1) To what extent is there measureable variation across schools in the level and growth of SEL measures and can we parse out the effects of schools on SEL development? (2) Does the level of SEL development and SEL growth in schools predict student achievement gains in Math and English Language Arts?

To answer these questions, we examine a unique panel data set of over 500,000 matched responses over two years across six large districts in California on a set of survey questions related to four social-emotional constructs: growth mindset, self efficacy, self management, and social awareness. We develop methodology similar to that of conventional value-added models on standardized assessments and assess variation across schools for each construct . As a byproduct of this analysis we assess the measurement characteristics of the SEL instrument and discuss the impact and resolution of missing data issues. We then compare the outcomes of this analysis to growth results on standardized assessments in Math and English for the same students.

In assessing the first research question, we find evidence suggest that there are measurable differences between schools in students' gains on SEL constructs and that these differences are practically significant. After controlling for prior SEL levels, we find meaningful differences in SEL measures across schools such that a school that is one standard deviation above the mean in SEL gains is between 0.06 and 0.25 standard deviations above the mean SEL outcome for comparable schools. The magnitude of these results is similar to those found in the assessment of school differences in their value-added to student achievement.

In assessing the second research question, we find that there is a weak, though measurable, correlation between a school effect on SEL construct measures and a school effect on achievement measures. The noise corrected correlation is especially weak in 4th grade (on the order of 0.06-0.08). Across other grades, it is between 0.15 and 0.36. These results hold for all four social-emotional constructs.

The data providing student level SEL measures for the population of relevant-age students across multiple districts is unprecedented. These results provide evidence that there is a difference between schools on their effect on SEL and that this difference may affect academic outcomes, though we have not established this causal link. This paper concludes with a discussion on the impact this research has on our knowledge of the policy use of SEL measures as well as on the numerous questions that remain to be answered as the field constructs a knowledge base of the measurement qualities of SEL instruments.