Combining cross-sectional market research surveys : an application of statistical matching

Abstract

In an ever complex market environment that features constant shifts in consumer preferences, marketers must understand the nature of consumer change. Information about previous shifts in consumer attitudes, lifestyles, values, and behaviour allows with varying degrees of certainty, the prediction of present circumstances from past information. However, it is purported that existing empirical research into the dynamic nature of consumer behaviour is hampered by a Iack of longitudinal panel data. Furthermore, it is well recognised that even when researchers are able to access panel data, it may be biased as a result of panel attrition, panel conditioning and non-response bias. The purpose of this study is to develop a methodology capable of modelling grosschange patterns using repeated cross-sectional surveys to replicate true panel data scenarios. Individuals at one time period are matched with similar individuals at the following time period using data fusion (statistical matching) techniques. This process allows the creation of pseudo panel data and the attitudes and behaviour of all pseudo individuals in a data set can be tracked over multiple time periods. The implications of matching data collected from two different time periods are explored and discussed. The method is developed using attitudinal data from The Nielsen Company's Panel Views Survey. The use of panel data allows for validation of the matching exercise. A simulation study is then conducted to test the method under different conditions. Lastly, the method is applied to data from the TNS Lifestyles and Opinions Survey, which is a series of repeated cross-sectional studies. This exercise is performed to demonstrate the applicability of this technique to marketing related problems This research contributes to methodology in consumer research by providing an alternative to panel data. It also extends the application of data fusion techniques beyond their usual scope of creating 'single source' data from two sources collected at a similar time point.