Exploratory Studies of Bias in Achievement Tests.

This paper considers the question of bias in group administered academic achievement tests, bias which is inherent in the instruments themselves. A body of data on the test of performance of three disadvantaged minority groups--northern, urban black; southern, rural black; and, southwestern, Mexican-Americans--as tryout samples in contrast to white, advantaged groups in the same regions, was analyzed using five different general methods for examining tests for bias. In an item tryout, a set of items is administered to a sample of the relevant population and the results are then examined item by item in an effort to pick the more effective items. The first method is an item selection routine using the point biserial correlation for each item as the criterion. The second method, group by score interactions, involves dividing the tryout group into, say, fourths, based on quartiles, and examining the proportion of the cases making each possible response in each of these levels. The third method involves plotting item difficulties so as to locate aberrant items. The fourth method involves estimating and plotting item characteristic curves separately for each group and comparing the plots. The fifth method comprises various intergroup factor analytic approaches. (Author/JM)