The Pearson formula, and a further note on the Kuhlman-Anderson tests
Abstract
One fault of the Pearsonian is that "two measures on a given case will affect in proportion as the measures are above and below averages, rather than in proportion to the amount of agreement between them." Or, whenever a case measures below average in one series and above in the other (so that is negative), is decreased, "although actual agreement . . . may be close." Detroit Alpha group and K-A tests were given to 136 pupils in grade V. The s were, for O, 0.794; for A, 0.682; for B, 0.912; for C, 0.803; and for D, 0.866. Measuring the agreement between groups by the actual average difference in M. A.'s as given by the two tests, A gives better agreement than O, although the s for the two groups indicate the reverse. Only in D does change to agree with the real difference. One may also determine a more refined difference between groups by means of a formula given and obtain the per cent. of disagreement. Again the condition above showing the greatest disagreement (B) exhibits the highest In studying these results in the same way it is shown that no comparison can be made from as to the relative amounts of correlation between the tests or as to changes in this correlation from grade to grade. For all tests except K-A, the agreement between the two halves becomes strikingly better for the higher grades. No tests can give uniform results at various levels. Tests of apparently wide range will show high reliability and validity as measured by and the higher the the poorer the tests in the traits under consideration.