Joint modeling of the longitudinal student mark and the competing events of degree completion and academic dropout

Within educational data mining, it is common to model students’ academic performance using a linear regression model or the time to degree completion or dropout using the cause-specific hazard model. Yet to our knowledge, no studies have simultaneously modeled the longitudinal performance and the hazard models. We propose a joint modeling approach in which we estimate the effect of a student’s semester longitudinal weighted mark obtained after attempting t credit points, on the hazard of degree completion and academic dropout. Evidence suggests that the joint modeling approach is substantially more efficient compared to the separate modeling of the longitudinal and the time-to-event outcomes. We observe similarities in the parameter estimates of the longitudinal submodels, but smaller standard errors of the estimates in the joint model. However, the parameter estimates of the competing risk models from both analysis methods are different. A unit increase in the average log weighted mark results in a 18.5 fold increase in the hazard associated with degree completion, but reduces the risk of dropout by 49 per cent. Being in a university-type residence, not having financial aid, and having a higher number of high school matriculation points all increase the hazard rate of degree completion.