Statistical Refutation of Comparative Functional-Causal Models

Abstract

The variables correlated by Murdock in his Social Structure (1949) were selected to demonstrate positive relationships anticipated in advance. Coult and Habenstein's (1965) printout of Murdock's (1957) "World Ethnographic Sample," in contrast, computes all the bivariate relationships among the 210 trait categories. The thousands of phi coefficients average close to zero, and the standard deviation about this average is only slightly greater than that which would result from sampling error if all the true values were zero. Sawyer and LeVine (1966) and Driver and Schuessler (1967) show that by collapsing the 210 categories into 30 variables and using a multivariate technique (i.e., factor analysis), correlations of substantial magnitude may be obtained from the entire sample without the kind of selection in favor of a preconceived theory that Murdock made. Some simple models of comparative functional theory and the resulting correlations show how far this over-simplified theory is from the actual statistical results obtained from Murdock's (1957) "World Ethnographic Sample."