Data mining of university philanthropic giving: Cluster-discriminant analysis and Pareto effects

Abstract

A large sample of 33 000 university alumni records were cluster-analyzed to generate six groups relatively unique in their respective attribute values. The attributes used to cluster the former students included average gift to the university's foundation and to the alumni association for the same institution. Cluster detection is useful in this application because there are very likely natural groupings that may represent segments of alumni that have much in common, and for which customized marketing approaches are justified. The clustering generated six groups of sizes ranging from eight to more than 25 000 records. Once the groups were established, discriminant analysis evaluated the difference(s) between the a priori determined and mutually exclusive giving groups. The first four groups provided the foundation with a disproportionate share (88 percent) of its major gifts, while the last two groups afforded the alumni association with the largest part (88.3 percent) of its annual funds. The discriminant model predicted with 100 percent accuracy the membership of the extreme groups (Group 1 and Group 6), with an overall predictive accuracy of 98 percent. The same model calculated membership in Group 2 and Group 5 with accuracy of 94.9 percent and 91.3 percent, respectively. Other numbers of groups were attempted, including five groups and seven groups, but no other number of clusters gave the clear distinction between groups, as did six clusters. The analysis further investigated two particular groups for their giving to the annual fund of the alumni association. These two groups comprised 98 percent of the sample, and could be characterized as one group that contributed substantially and another that did not. The findings provide a basis for marketing communication efforts aimed at specific niches within the groups based on giving patterns and demographics. As per the Pareto Principle, one cluster accounts for 63.7 percent (almost two-thirds) of contributions to the alumni's annual fund from only 20.1 percent of the sample alumni, giving an average of US111 annually and making more than eight gifts since graduation. The gift amounts from members of this cluster to the annual fund are in relatively small parcels, with 95 percent of the contributions averaging less than 250. The largest cluster with 78.3 percent of the entire sample can be characterized as non-contributors to both the annual fund and the foundation. © 2009 Palgrave Macmillan.