Methodology to combine theoretical knowledge with a data-driven probabilistic graphical model

Abstract

ABSTRACT This study presents an analytic inference methodology using probabilistic modeling that provides faster decision-making and a better understanding of complex relations. Two educational psychology models (i.e., the MUSIC Model of Motivation and the domain identification model) were coupled with a data-driven Probabilistic Graphical Model to provide a top-down and bottom-up combination for reasoning. Using survey data from middle school students, Bayesian Network models captured the probabilistic interactions between students’ perceptions of their science class, their identification with science, and their science career goals. Complex/non-linear relationships among these variables revealed that students’ perceptions of their science class (i.e., eMpowerment, Usefulness, Success, Interest, and Caring) were significant predictors of their science-related career goals. These findings provide validity evidence for using the MUSIC and domain identification models and provide educators and school administrators with a web-based simulator to estimate the effect of students’ science class perceptions on their science identification and career goals.