An Integrated Heuristic for Validation in Sequence Analysis

Sequence Analysis is a collection of tools to describe life courses represented as sequences that are increasingly applied in different fields, particularly in demography, sociology, and political sciences. Identifying typologies through cluster analysis, thus disregarding individual sequences’ peculiarities, is the aim of most applications. However, a substantive interpretation of such typology can be questionable when clusters include sequences deviating from the others. We propose an integrated approach to identify such sequences, distinguishing between sequences presenting structural peculiarities and randomly deviating sequences. We monitor the quality of partitions with respect to the amount and type of deviation in each cluster relying on novel graphical tools allowing to properly visualize and closely inspect the structure of deviating sequences. We demonstrate that the identification of deviating sequences provides relevant insights also when clusters are used as dependent or independent variables in an explanatory framework, for example in combination with multinomial logistic regression analysis.