Modeling Teacher Attrition

Abstract

This paper describes the literature on teacher attrition as either focusing on the working conditions faced by beginning teachers or highlighting variations in teachers' characteristics as causes for early teacher attrition. This study uses responses to the School and Staffing Survey (SASS) along with the Teacher Follow-Up Survey (TFS) to compare these contrasting views of early teacher attrition. Two logistic regression models were constructed and their relative efficacy in explaining teacher attrition were compared using three statistical techniques; model fit characteristics (e.g. pseudo-R2, Akaike Information Criteria, Bayesian Information Criteria); a comparison of their classification effectiveness, and results from Davidson and MacKinnon's J test (1981). A final model was also constructed using the predictive elements of each of the previous models. Results suggest that the working conditions model better fits the observed data than the teacher characteristics model. The final model highlights the importance of teacher commitment and engagement in the profession in teachers' career decisions.