Cox Proportional Hazards Regression for Interval-Censored Data with an Application to College Entrance and Parental Job Loss

Abstract

This study involved conducting a survival analysis by fitting a Cox proportional hazards model to Korea Labor Panel data to analyze the impact of parental job loss on children’s delayed admission to colleges and universities in South Korea, using 376 subjects whose parental education levels were college-level or higher. Since Korea Labor Panel data are interval- and right-censored, we compared three imputation methods: simple omission, imputation as the average of the left and right values of the interval, and multiple imputation. Their integrated areas under the ROC curve (AUC) and mean square errors (MSE) were compared to assess their predictive and estimation performances. It was found that, within the simulation, the multiple imputation method exhibited a lower MSE than the other two methods. However, no difference was observed in the iAUC values. In the group where each householder had at least a college degree, parental job loss was significantly related to the delayed college or university admission of the first-born child regardless of the use of the interval censoring imputation method. In particular, when the first-born children experienced their parents’ unemployment at the age of 18, the probability of college admission was reduced nearly by 53% compared to cases where they did not. This analysis implies that the government should develop a policy in the education system offering psychosocial support for adolescents who cannot expect parental help.