A Construct Validation of Adaptive Achievement Testing.

Abstract : The construct validities of conventional paper-and-pencil and adaptive achievement tests were compared using data from two independent groups of 269 and 230 college students. Two adaptive achievement tests were computer administered to each group using the stradaptive testing strategy; each group also completed two conventional classroom paper-and-pencil achievement tests. All achievement tests were drawn from the same pool of achievement test items on which item characteristic curve (ICC) parameters had been determined. Students were also administered two stradaptive vocabulary tests. All tests were scored by maximum likelihood estimation using the three-parameter logistic model. A nomological net was specified describing the relationships of the achievement tests to the achievement constructs and their relationships with the vocabulary construct and the vocabulary tests. The parameters of the net were estimated by fitting the observed intercorrelations among the test scores to the nomological net, using the methodology of linear structural equations. Maximum likelihood estimates of the parameters of the nomological net indicated essentially equal validities for the classroom and adaptive tests in four comparisons. The results of this study indicate that computer-administered adaptive tests can provide more valid measurement of achievement than conventional paper-and-pencil tests.