Analysis of Mediators of ERF's Impacts on Classroom Instructional Practice and Children's Language and Literacy Skills

As a final part of the analysis of ERF, we explored potential channels, or mediators, through which ERF generated its positive impacts on classroom and child outcomes. Unlike the impact analyses, this analysis is correlational, rather than quasi-experimental, because we could not use the regression-discontinuity design to identify the causal effects of particular mediators.

Consequently, any observed effect of mediators on child or classroom outcomes might be due to the effects of unobserved factors that happen to be correlated with these mediators, rather than to the mediators themselves.

For our analysis of the channels through which ERF generated positive impacts on classroom and child outcomes, we hypothesized that the additional hours of professional development attributable to ERF and the increased proportion of teachers receiving professional development through intensive, individualized mentoring account for at least some of ERF's impact on the classroom language and early literacy environment. The impacts on classroom environments, in turn, might account for at least some of the program's impacts on children's language and literacy skills.

To investigate this hypothesis, we first examined the extent to which hours of professional development and the use of mentoring as a mode of training were associated with the classroom outcomes affected by ERF. We then examined the associations between classroom outcomes and the child outcome on which ERF had a positive impact—print and letter knowledge. Thus, our model of print awareness includes as mediators the number of phonological awareness activities, print- and letter-knowledge learning opportunities, written-expression learning opportunities, the classroom print environment, opportunities and materials for writing, book-reading practices, child portfolios, and teacher sensitivity.

The estimated marginal effect of hours of professional development is generally small and not statistically significant on each of the 10 measures with the exceptions of classroom print environment and teacher sensitivity; we estimated positive and statistically significant effects of professional development on those two measures. Similarly, the estimated marginal effect of mentoring on each of the 10 outcomes is generally small and not statistically significant, with the exceptions of child portfolios and teacher sensitivity; the estimated marginal effects of mentoring are negative and statistically significant on those two outcomes. The mediators are jointly statistically significant only for child portfolios and teacher sensitivity.

The estimated marginal effects on print and letter knowledge are not statistically significant for any of the potential mediators except print and letter-knowledge learning opportunities, which account for 27 percent of the total implied impact on print-awareness scores. Together, all eight mediators account for 60 percent of the total implied impact on print and letter knowledge and are jointly statistically significant at the 5-percent level.