Data mining analysis of the effect of educational, demographic, and economic factors on time from doctoral program entry to degree completion in education.

摘要

The duration of doctoral studies has been linked to low persistence rates and can therefore be viewed as an indirect measure of risk for non-completion. In 2004, the median time between masters and doctorate in Education was 12.7 years, which was 4 years longer than the median for all fields. The purpose of this study was to identify demographic, educational, and economic factors associated with atypically long time (i.e., the highest 33%) between doctoral program admission and degree completion. The population included all doctoral recipients in Education from Florida public universities between 1998 and 2004 (n=773). Data mining was used to generate six models which were compared on the basis of variable importance and predictive accuracy. Tree and classification models were compared to models developed through the traditional statistical methods of discriminant analysis and logistic regression. Assessment of model predictive accuracy was based on four criteria as follows: cumulative misclassification rate, weighted misclassification rate, misclassification of the dependent variable, and model consistency. The models' predictive accuracy differed but there was general consensus on variable importance with educational and institutional variables superseding all demographic variables. The highest predictive accuracy was observed in the three tree models which validated the analytic merit of data mining in this study.