Identification of Segmentation Criteria for the Improvement of Population Forecasts

It is often stated that the single most important influence on the quality of forecasts of behaviour is the set of predetermined exogenous variables (typically sociodemographic) that categorise the population. These variables are used to carry the sampled population into the future as a representation of the composition of future populations. Such variables or segmentation criteria are typically dictated by the limited set of descriptors projected by specialised forecasting agencies. Although the wisdom of their efforts is respected, from time to time it seems worthwhile to reappraise the set of exogenous variables which are subject to intense forecasting to see if there may be a case for considering additional variables (and even ‘resting’ the current set). It is argued that a desired segmentation set should be linked to preference stability, and that a suitable procedure for establishing such a link is via preference data derived from a controlled experimental design. From the empirical study it is illustrated how preference data can be combined with socioeconomic data to seek out the role of the ‘current’ set and the ‘new’ set of socioeconomic variables in a particular forecasting context. Since the selection of variables must be application specific, the main emphasis of this paper is on the methodology. The reported empirical findings are of illustrative use only.