ERIC Number: ED247297
Record Type: RIE
Publication Date: 1983-Feb
Reference Count: 0
The Effect of Number of Ability Intervals on the Stability of Item Bias Detection.
The chi-square procedure has been suggested as a viable index of test bias because it provides the best agreement with the three parameter item characteristic curve without the large sample requirement, computer complexity, and cost. This study examines the effect of using different numbers of ability intervals on the reliability of chi-square statistics as indicators of item bias in educational tests. In 1980 the English Usage exam from the American College Testing Program was administered to 922 black and 4423 white high school juniors and seniors at 16 test sites. Four randomly equivalent samples of black and white examinees were created. Three sets of ability grouping intervals, based on total raw score on the 75 items, were established using three, four, and five intervals. The same number of examinees were represented in each interval. A chi square index for each test item and three measures of stability (correlations of chi-square values; correlation of 0, 1 bias decisions; and the kappa index) were calculated. Results confirm previous studies and suggest that the chi-square index is fairly unreliable regardless of the number of grouping intervals and cannot lead to consistent decisions about item bias. (BS)
Publication Type: Speeches/Meeting Papers; Reports - Research
Education Level: N/A
Authoring Institution: N/A
Identifiers: American College Testing Program; Chi Square; Kappa Coefficient
Note: Paper presented at the Annual Meeting of the Eastern Educational Research Association (Baltimore, MD, February 24-27, 1983).