New Releases by Jen Wang

COLOUR-STRUCTURE RELATIONSHIPS, ANTHRAQUINONE DERIVATIVES, 14H-NAPHTHO[2,3-a]-PHENOTHIAZINE.., NEAR-IR ABSORBING DYES.

One of the major categories addressing reliability of criterion-referenced testing is suggested by the concept of decision-consistency--the consistency of mastery-nonmastery classification across repeated measures with one test form or parallel test forms. The purpose of this study is to investigate the accuracy of four decision-consistency approaches in estimating reliability of criterion-referenced tests: (1) Swaminathan method, (2) Huynh method, (3) Subkoviak method, and (4) Marshall method. A computer simulation of test performance was used to investigate the effects of the test variability, the number of items, the number of examinees, and the cut-off scores on the accuracy of estimates derived from these four methods. For each of the six combinations of number of item (n = 5, 10, 20) and number of examinees (N = 15, 30), three matrices were constructed with elements p(,ij) (i = 1,2,...,N; j = 1,2,..., n) representing the true probability of success for the i(''th) examinee on the j(''th) item. These matrices were used in simulating the responses of three hypothetical groups (low, medium, and high true score variability) of N examinees to n items. The computer simulation for each of the 18 tests was accomplished by a Fortran program over 100 replications. Means and the mean deviations of each estimate from true agreement index over 100 replications were computed as evaluative indices. The results indicate that: (1) As number of examinees decreased from 30 to 15, the mean estimates of agreement index remain unchanged; (2) The effect of cut-off score changes on the mean estimates is a function of true score variabilities and methods; (3) The mean estimates increase with increasing test length and true score variability under most conditions studied; and (4) All things considered, the Huynh and Subkoviak methods are quite similar to each other. These two methods tend to underestimate under most of the conditions and tend to provide smaller standard error of estimates. Marshall''s method is less sensitive to cut-off score changes. Swaminathan''s estimator is closer to "true" agreement index, compared to other methods, although it tends to underestimate. The variability of estimator for Swaminathan''s method is always larger than the variability of estimator for other methods.