CLUSTER ANALYSIS / Maxwell, Pryor, and Smith 22 2002 World Cultures 13(1): 22-38 CLUSTER ANALYSIS IN CROSS-CULTURAL RESEARCH Bruce A. Maxwell Department of Engineering, Swarthmore College, Swarthmore, PA 19081; Maxwell@swarthmore.edu Frederic L. Pryor* Department of Economics, Swarthmore College, Swarthmore, PA 19081; Fpryor1@swarthmore.edu Casey Smith Swarthmore College, Swarthmore, PA 19081 * Corresponding Author This essay has three purposes. The first is to present a relatively non-technical description of cluster analysis. The second is to describe a computer program available on the World Wide Web, which allows such a statistical technique to be carried out in a very simple way. The third is to show how this approach can be used with cross-cultural data to extract similarities and differences between societies in a systematic fashion. Although the example used focuses on the economic systems of foragers, the methodology is also applicable to a wide variety of other cross-cultural research problems. ? 1. INTRODUCTION Although considerable cross-cultural data are available - for instance, the 1700 series for the Standard Cross-Cultural Sample published by World Cultures - such information has been underutilized. Part of the problem is that so much information is available that it is difficult to discern patterns in a sufficiently objective manner to allow others to be able to replicate the results. One traditional way to reduce the dimensionality of the data is to use some variant of principle component analysis, a technique that permits us to determine which traits are related. Nevertheless, if we wish to determine which societies are the most similar or different using the results of the derived principle components, difficulties begin to arise because, according to one factor, two societies may be very different while, according to another factor, they may be quite similar. Other analytic problems arise because the results may depend upon whether we employ a standard principle component analysis, where, in effect, each factor is removed before the next factor is derived (thus deriving orthogonal factors) or some type of varimax technique in which the factors may be related. Cluster analysis approaches the problem of determining similarities and differences among societies more directly, namely by determining the multidimensional distances between various societies and then picking out those groups of societies within which the distances are relatively small. This statistical technique has been used in a wide variety of data analysis and pattern recognition applications, and a number of clustering techniques exist, the most common ones being the K- means and hierarchical clustering algorithms (MacQueen 1967; Johnson 1967). The k-means