GROUP SIGNIFICANCE TESTS
THE PROBLEM we seek to solve in this annex is the problem of group significance, the so-called dredging, combing, or mudsticking problem. "They throw the mud on the wall and see if it sticks," is the way one critic has described this cross-cultural survey method. In this study, the mud that stuck consists of our Communication Cluster. We did not posit these associations in advance. They were a comparatively small number of significant associations gleaned from a much larger number of correlations which did not attain significance. But as common sense tells us, and as Banks and Textor ( 1963) and Textor ( 1967) have shown, even if the data are mere nonsense collections of random numbers, if thousands of correlations were run with a computer, dozens of them are bound to turn up statistically significant.
Our solution to this problem is an elaboration of the Whiskers solution used by Banks and Textor ( 1963) and Textor ( 1967:54). We ourselves ran nonsense correlations and compared these findings as a group with the findings of our real data runs as a group.
The first step in our solution was the preparation of a special computer program to generate Whiskers variables. Banks and Textor ( 1963) and Textor ( 1967) had several sets of these variables. In Textor ( 1967:54) the Whiskers variables differed in the dichotomous cut. Of the 400 societies in his sample: 50% had purple whiskers, 50% did not; 40% had blue whiskers, 60% did not; 30% had green whiskers, 70% did not; 20% had pink whiskers, 80% did not; 10% had yellow whiskers, 90% did not; 5% had white whiskers, 95% did not.
The point of varying the dichotomous cuts lies in their implications on the marginals of fourfold contingency tables and thus on the number of possible arangements of the four cells in the table. Consider the examples in table D.1. Suppose we have a sample of 100 houses. 50 of these have red roofs; the other 50 have black roofs. 40 have red doors; the other 60