Combining optimal scaling and survival techniques to identify possible predictors for unemployment duration

Abstract

In this article we propose a new approach in survival analysis in a non-traditional field of application; unemployment data. A common practise is to use factor analysis to first summarize survey data, and then fit a binomial logistic regression model to estimate the regression weights, which are used to identify factors associated with work status at a prespecified time point. In this article, a combination of optimal scaling and survival analysis methods is proposed as an alternative to find possible predictors for unemployment duration. This combination of techniques is illustrated and compared to the traditional approach. Data from the Dutch Employee Insurance Agency are used to illustrate the methodology.