A Statistical Process to Incorporate the Use of Demographics to Help Select the “Best” Number of Market Segments

The traditional method to identify/select the proper number of segments (clusters) has been to (i) cluster analyze the core data (e.g. AIOs, satisfaction statements), (ii) use one (or more) of several suggested criteria to select the most homogenous cluster solution, and finally (iii) profile the selected number of clusters by cross tabbing them with demographic data. This final step enables the researcher to design specific/directed media messages for each segment from the “best” number of clusters. Our approach is somewhat in the reverse of this process (at least in determining the “best” number of segments). We first do the clustering with the core data (as in (i) above) and then use a matrix matching criterion, the Adjusted Rand Index, to select the most appropriate number of segments by comparing how each cluster solution set is defined by the demographic variables. The cluster solution set that has the best match, i.e. highest average Adjusted Rand Index, with the demographics is selected. By selecting the “best” number of clusters in a reverse matching of the demographics, we yield a more unique profile for each segment. This process, does not use a traditional clustering criterion to select the number of clusters, but instead uses the Adjusted Rand Index to maximize the uniqueness of the profiles. To test the efficacy of this process, we designed a simulation study using core data with varying amounts of cluster and demographic structure. We then compared the results of number of clusters selected using the proposed process to that of the original cluster structures. Results indicate that the average Adjusted Rand Index value reached its highest levels when the “correct” numbers of clusters are selected. Implications for research are discussed.