RACIAL TEST BIAS: AN ARTIFACT OF TEST DEVELOPMENT METHODOLOGY?

The issue of racial test bias in intelligence tests, primarily raised when Blacks as a group consistently score approximately one standard deviation less than Whites as a group, has been examined from many perspectives. The suggestion has been made that Blacks obtain lower mean scores simply as a result of their small representation in the test development population. This study examines the possible differential effects on the test performance and the predictive validity of intelligence tests resulting from using equal numbers of Blacks and of Whites in the test development sample. From the standardization sample of an existing intelligence test (K-ABC), two new intelligence test forms--a Black form and a White form--were constructed using equal numbers of Blacks and of Whites as the test development samples. Three one-way MANOVA\u27s were used to examine the differential test performance across racial group of the newly developed scales. Results indicate that Blacks as a group did not perform significantly better on the test form developed solely on their own item statistics. A Potthoff procedure was used to determine if the newly developed test forms equally predicted achievement across racial group. The only bias in predictive validity occurred on the intercepts of the regression lines and in all cases the Whites scored higher on the criterion measure. These findings seem to indicate that, at least from a statistical perspective, using equal numbers of Blacks and of Whites in the test development sample results in the same findings of little or no differential construct and predictive validity of intelligence tests across racial groups as reported in studies using tests developed using population proportionate sampling procedures. Mean differences between racial groups were reduced which may have social, ethical and political implications for our current test development procedures