?Papers such as Rudner's have addressed theoretical and methodological concerns about the nature, accuracy, and interchangeability of person-fit statistics. Other researchers have addressed the frequency and amount of abnormal response patterns and, therefore, the practical utility of person-fit statistics.
Frary (1980) calculated four different fit statistics on a large sample of eighth graders who had taken a commercial achievement test battery. Along with the moderate to strong intercorrelations mentioned earlier, Frary found that blacks and females differed from whites and males on some tests. Overall, females showed fewer aberrant responses than males, but racial differences occurred in both directions. Among low-scoring students, the effects were consistent: White and female students made fewer unusual choices.
These findings raised the possibility of test bias. But using a knowledge assurance statistic, Frary concluded that blacks and males did better than whites and females when it came to correctly guessing items for which they had only partial knowledge.
Doss (1981) applied a residual mean-square statistic from the computer program RASCH in the PRIME system to a fifth-grade Chapter 1 setting, where children were given the Iowa Tests of Basic Skills with https://writeanypapers.com. He examined how removing the poorest fitting 10%, 20%, and 30% of students affected the accuracy of pretest predictions. While the N dropped substantially as he removed students, prediction accuracy increased with each removal.
Although the improvement in accuracy is interesting, Doss's setting may not have provided a meaningful testing situation. The test was badly matched to the students abilities: Even though some (Doss doesn't say how many), took the fourth-grade level of the battery, 25% scored at or below the chance level. After the worst fitting 30% had been removed, only 13% of the Chapter 1 students remained. Finally, the students showed losses from pretest to posttest. The study does, nonetheless, point out another use of person fit statistics--to objectively document whether the testing situation is meaningful.
Schmitt and Crocker (1984) investigated the relationship between scores on the Test Anxiety Scale for Adolescents and person-fit. They used various indices and the Metropolitan Achievement Tests in reading, mathematics, and science in seventh and eighth grades. Students in the middle ability range showed no relationship between test anxiety and person-fit indices. High-ability, low-anxiety students showed greater misfit than high-ability, high-anxiety students. At the low-ability end, the reverse was true: Low-ability, low-anxiety students showed less misfit. The authors offer some conjectures on the findings in terms of a Cognitive-Attentional Theory of Test Anxiety, but present no data that might support their notions.
