Exploring repeated measures data sets for key features using Principal Components Analysis

Repeated measures data sets are a very common data structure in marketing, ranging from weekly household purchase data obtained via supermarket scanners, household penetration rates for products over time, panel survey questionnaire responses. etc. In many circumstances, it is desirable to glean key features of the data without having to run a formal model and/or perform a large number of exploratory analyses. Such features of interest may include defining market segments, identifying change points (a sudden shift in aggregate behavior), and. in general, identifying "typical" behavior patterns. In this research, we apply an exploratory approach for analyzing repeated measures data sets, initially considered by Tucker [Psychometrika 23 (1958) 19] and fatly explicated in Jones and Rice [Am. Stat. 46 (1992)], that utilizes Principal Components Analysis (as its core), regression, and simple bivariate plots. It is clearly our intention, via the algorithm provided, that a user can read this manuscript and then sit down at his or her computer and implement the approach immediately. We first demonstrate our approach on simulated data, therefore providing a "blueprint" for users of this approach, i.e. we simulate different types of structural variation and provide the corresponding principal component rotation curves, and how to glean features from them. Our approach is then applied to a data set on cumulative penetration rates for a set of durable products. (C) 2002 Elsevier Science B,V. All rights reserved.