Single-Level Weighted Analyses of NAEP Teachers

• In Event: Feasibility of Teacher-Centric Analyses Using NAEP Data

Abstract

This section examines whether there is any advantage to conducting a teacher-weighted single-level analysis of teacher outcomes. In the second, third and fourth presentations of this symposium we demonstrated that NAEP analysis using representative teacher sampling weights is possible and valid. But of greater interest to us as researchers in general, and to NCES in particular, is the question of whether such analyses are useful. Is there any additional benefit to analyzing teacher-weighted NAEP data that cannot already be achieved using student-weighted data? This question has consistently been raised by the NAEP sampling director (2013, 2014, 2015) in technical memoranda on the topic. Unless weighted teacher analyses offer a substantial advantage over existing analyses, NCES must question whether it is worthwhile to invest resources in developing them. This presentation, and the one that follows it, address that question.
In the first part of this presentation we examine the weighted distributions of fourth grade public-school mathematics teachers by a set of ten teacher and school characteristics (Tables 6 and 7). We then compare these distributions to a parallel analysis of student-level data, using student sampling weights. Our comparison shows that teacher-weighted crosstabs provide useful information that cannot be obtained using student-weighted crosstabs, especially with regard to class size and school enrollment.
In the second part of this presentation we examine a logistic regression model predicting the odds of a fourth grade teacher having a Master’s degree (Table 9). Results for school predictors indicate that odds were decreased by 31% for teachers in towns and by 39% for teachers in rural locales, relative to teachers in suburban locales. Results also suggest lower odds for teachers having a Master’s degree in schools with higher proportions of non-White and NSLP-eligible students. Results for teacher predictors suggest that odds were also significantly lower for teachers with fewer years of experience and for Hispanic teachers, but higher for certified teachers and Black teachers, holding all else constant.
The teacher-centric analyses presented in this presentation are novel in that they reflect use of NAEP data in ways that are currently not possible. These analyses represent an innovative use of NAEP data with regard to teacher outcomes. In light of these findings, and given our assertion that fourth grade assessments contain a representative sample of teachers, NAEP could be added to the list of NCES data sources (such as SASS and ECLS-K:98) available for analysis of teacher outcomes.

Author

• Jonathan Phelan, American Institutes for Research