Longitudinal categorical data arise in many diverse areas of the social sciences and methods for its analysis have taken two broad directions. Heuristically, one can attempt to model the state space (i.e., the categories) or the sequence space (i.e., the subjects), typically with event history models or optimal matching respectively. This study proposes a more general framework for inference from such data which acknowledges not only the analytic approach (split into stochastic models and algorithmic differencing) but also hypothesis, sequences, categorisation and representation. The individual sequence can be thought of as a map from time to the state space. The hypothesis relates to how these maps are similar and how they deviate from this structure. The analytical frameworks define what is assumed, what is uncertain, and how this is modelled. The categories of the state variable define what is considered pivotal as an event. Representations create explorative tools and describe structure, as well as communicating high dimensional inferences. It is the interaction between these ideas which is fundamental to making inferences, as well as their relationship to time, which is essential to the social science treatment of sequences. Thus, the analysis should not prefer one approach to analysis over another but appreciate the origin of the data and the theory under examination. Copyright Springer Science+Business Media B.V. 2013
