Bayesian Proportional Hazard Analysis of the Timing of High School Dropout Decisions
Mingliang Li
Abstract
Mingliang Li
Abstract
In this paper, I study the timing of high school dropout decisions using data from High School and Beyond. I propose a Bayesian proportional hazard analysis framework that takes into account the specification of piecewise constant baseline hazard, the time-varying covariate of dropout eligibility, and individual, school, and state level random effects in the dropout hazard. I find that students who have reached their state compulsory school attendance ages are more likely to drop out of high school than those who have not reached compulsory school attendance ages. Regarding the school quality effects, a student is more likely to drop out of high school if the school she attends is associated with a higher pupil–teacher ratio or lower district expenditure per pupil. An interesting finding of the paper that comes along with the empirical results is that failure to account for the time-varying heterogeneity in the hazard, in this application, results in upward biases in the duration dependence estimates. Moreover, these upward biases are comparable in magnitude to the well-known downward biases in the duration dependence estimates when the modeling of the time-invariant heterogeneity in the hazard is absent.
OpenAlex reports 13 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
In this paper, I study the timing of high school dropout decisions using data from High School and Beyond. I propose a Bayesian proportional hazard analysis framework that takes into account the specification of piecewise constant baseline hazard, the time-varying covariate of dropout eligibility, and individual, school, and state level random effects in the dropout hazard. I find that students who have reached their state compulsory school attendance ages are more likely to drop out of high school than those who have not reached compulsory school attendance ages. Regarding the school quality effects, a student is more likely to drop out of high school if the school she attends is associated with a higher pupil–teacher ratio or lower district expenditure per pupil. An interesting finding of the paper that comes along with the empirical results is that failure to account for the time-varying heterogeneity in the hazard, in this application, results in upward biases in the duration dependence estimates. Moreover, these upward biases are comparable in magnitude to the well-known downward biases in the duration dependence estimates when the modeling of the time-invariant heterogeneity in the hazard is absent.
Key concepts: Covariate, Dropout (neural networks), Hazard ratio, Bayesian probability, School dropout, Hazard, Econometrics, Duration (music)