Predicting Work and Family Trajectories

Raffaella Piccarreta, Università Bocconi

Techniques based on sequence analysis, and in particular Optimal Matching Analysis (OMA), have been used to build clusters of life courses. Proposals on the prediction of life course sequences have been made by McVicar and Anyandike-Danes (2001), who use multinomial logit models to study the determinants of cluster membership, adopting thus a two-step approach. The main problem with this approach is that cluster found not considering the prediction purpose may be hardly predictable. We propose a new algorithm, modifying the first step of this procedure. Clusters are still obtained considering OMA, but the predictive problem is taken into account also in the first step, when clusters are formed. The aim is to define clusters that are better predictable given a set of covariates. We apply this algorithm to British Household Panel Survey data on the employment and family trajectories of women, and we show the advantage of the proposed algorithm.