Michaels, H. R., Ferrara, S. F., Huynh, H. (1995, April ). Contextual Characteristics of Local Item Dependence in a Mathematics Performance Assessment. Paper present at the 1995 annual meeting of Education Research Association, San Francisco, CA.

Abstract: This study continues the examination of local item dependence (LID) in Maryland's large scale assessment program to identify contextual characteristics associated with low and high LID. It also begins to validate the logically derived cases and the empirically derived ones. The test in question was the mathematics content portion of the Maryland School Performance Assessment Program, administered to approximately 55,000 students at each of grades 3, 5, and 8. Hypothesized causes of LID from S. Ferrara, H. Huynh, and H. Baghi (1994) and W. Yen (1993) were used to predict high and low LID in the mathematics content items, and these predictions were compared to the empirical calculation of LID. Overall, a large number of cause codes (n=103) were identified. The contextual analysis correctly identified LID 95% of the time. High LID was overpredicted four times, since the contextual analysis did not predict medium LID. Results show the ability to predict LID, which begins to support the validity of some of the hypothesized causes of LID. Most LID was explained by using items that required explanations of processes, extensions of data or learning, or conclusions to problems. This paper, presented at the annual meeting of the American Educational Research Association, San Francisco, California, April 1995, is also available in the "Journal of Educational Measurement," volume 36, number 2, pages 119-140, Summer 1999 (TM 521 952). (Contains 7 tables and 13 references.)