Test Reliability by Ability Level of Examinees / Kathy Green and Gilbert Sax.

Achievement test reliability as a function of ability was determined for multiple sections of a large university French class (n=193). A 5-option multiple-choice examination was constructed, least attractive distractors were eliminated based on the instructor's judgment, and the resulting three forms of the examination (i.e. 3-, 4-, or 5-choice question form) were randomly assigned to quiz sections with similar mean cumulative grade point averages. Students were later grouped into high (3.6-4.0), average (3.1-3.5), and low (0-3.0) ability levels based on their final course grades in French where B=3.0 and A=4.0. A Kuder-Richardson 20 reliability coefficient was computed for each test form for each ability group and adjusted by the Spearman-Brown formula. Differences among reliabilities for the three forms were: (1) significant at alpha=.05 for the low ability group; (2) not significant for the high ability group; and (3) significant at alpha=.10 for the average ability group. The ability groups were combined and differences among reliabilities for the three forms were significant at alpha=.05. The optimal number of alternatives for all ability groups combined was four. (Author/RL)

Microfiche. [Washington D.C.]: ERIC Clearinghouse microfiches : positive.