Analyzing Faculty Workload Data Using Multilevel Modeling

The results indicate a large grouping effect on research productivity, and the statistical as well as the substantive results of the random coefficients model differ significantly from the regression approach.

Abstract

Research on faculty productivity fails to account for the hierarchical nature of the data. Faculty members within an academic discipline more closely resemble one another than faculty in other disciplines, resulting in dependent observations and thus inaccurate statistical results. Unlike ordinary least squares, multilevel modeling takes into account this grouping effect. This article analyzes the research productivity of 1,104 tenured/tenure-track faculty from the 1993 NSOPF survey to compare traditional regression models with a random coefficients model. The results indicate a large grouping effect on research productivity, and the statistical as well as the substantive results of the random coefficients model differ significantly from the regression approach.