1. Demonstrating Local Item Dependence for Recognition and Supply Format Tests.
- Author
-
Bastick, Tony
- Abstract
This study tested the hypothesis that the common approach to test construction in which recognition questions (RQs), such as multiple-choice items, are followed by constructed response questions (CRQs) encourages students to use the informationally rich RQs to gain marks on the CRQs, thus introducing Local Item Dependence (LID) and inflating the CRQ test scores. This was tested with 188 children aged 10 to 16 years in 5 schools using class tests in 4 topic areas. The children in each class were randomly assigned to take the test in the traditional RQ-CRQ order, or in the experimental CRQ-RQ order. Using two independent t-tests, the groups were then compared on their RQ scores and on their CRQ scores. The results indicate that a statistically significant advantage was gained on the CRQs when the traditional order of test construction was used. Differences in mean RQ scores were used to check if factors other than LID, which could be associated with the nontraditional order, might have influenced CRQ results. These checks showed no statistically significant differences between the two groups. It is concluded that the traditional order can produce LID and result in inflated test scores for the constructed response part of the test. (Contains 1 figure, 1 table, and 24 references.) (SLD)
- Published
- 2002