A PRIMER ON SEQUENCE METHODS

This paper considers the technical problem of analyzing sequences of social events. Examples of such sequences from organizational behavior include organizational life cycles, patterns of innovation development, and career tracks of individuals. The methods considered here enable the analyst to find characteristic patterns in such sequences. Forces shaping those patterns can then be found by more conventional methods. After a brief definitional section, the paper begins by discussing three types of sequence questions: (1) questions about whether a typical sequence or sequences exist, (2) questions about why such patterns might exist, and (3) questions about the consequences of such patterns. The theoretical foundations of the first type of question, which is in fact the most important, are then considered. Having established the legitimacy of the approach here taken, the paper then introduces two exemplary datasets with which to focus discussion. These raise the issue of conceptualization and measurement of sequence data. Illustrative cases are presented to show the importance of extreme care in conceiving a sequence to measure and then choosing indicators for it. The paper then turns to methods proper, considering them in several categories. It first briefly mentions methods not employing "distance measures" between events: permutational techniques, stochastic (e.g., Markov) models, and durational methods. Most of these do not directly address sequence questions but can be used to do so if necessary. Turning to the methods based on event distance, the paper first considers the problem of measuring distance between events (1) in terms of elapsed time, (2) in terms of categories of events, and (3) in terms of observed successions. It then considers methods for unique event sequences (sequences in which no events repeat), proposing the use of multidimensional scaling and illustrating it with an analysis of data on medical organizations. For the separate case of repeating event sequences, the paper discusses optimal matching methods, which count the number of individual transformations required to change one sequence into another. These methods are illustrated by an analysis of data on musicians' careers. The paper then briefly considers the problem of findings subsequences common to several longer sequences (or repeated in one longer sequence). It closes with a discussion of assumptions made and caveats required when these types of methods are used.