ABSTRACT
It is important to note that the test of assumption of normality is appropriate only if we assume that randomly determined reproducibilities for an infinite number of sets of data are normally distributed around chance reproducibility. To test this assumption, over sixty thousand data matrices were generated by computer, each set having specified numbers of items and cases and specified marginal frequencies. If an initial statistical test fails, response categories may be reclassified and items eliminated and the resulting data matrix may then pass a statistical test. This means that the statistical test, and any other test which accepts marginal frequencies and the dimensions of the data matrix as starting values of determination of the statistical significance of scalogram, is a limited test. The three phases of scalogram analysis—measurement of deviation from pattern, elimination of chance as an explanation of pattern, and scoring— though rarely treated separately, are viewed in this note as distinct operations.
