Computerized Adaptive Testing

Historical Perspectives

Adaptive testing is not new. Through the ages examiners have asked questions and, depending on the response given, have chosen different directions for further questioning for different examinees. Clinicians have long taken adaptive approaches, and so since the advent of standardized intelligence testing, many such tests have used adaptive techniques. For both the 1916 edition of the Stanford-Binet and the 1939 edition of the Wechsler-Bellevue intelligence tests, the examiner chose a starting point and, if the examinee answered correctly, asked harder questions until a string of incorrect answers was provided. If the first answer was incorrect, an easier starting point was chosen.

Early group-administered adaptive tests faced administration problems in the precomputerized administration era. Scoring each item and making continual routing decisions was too slow a process. One alternative explored by William Angoff and Edith Huddleston in the 1950s was two-stage testing. An examinee would take a half-length test of medium difficulty, that test was scored, and then the student would be routed to either an easier or a harder last half of the test. Another alternative involved use of a special marker that revealed invisible ink when used. Examinees would use their markers to indicate their responses. The marker would reveal the number of the next item they should answer. These and other approaches were too complex logistically and never became popular.

Group-administered tests could not be administered adaptively with the necessary efficiency until the increasing availability of computer-based testing systems, circa 1970. At that time research on computer-based testing proliferated. In 1974, Frederic Lord (inventor of the theoretical underpinnings of [Page 203]much of modern psychometric theory) suggested the field would benefit from researchers’ getting together to share their ideas. David Weiss, a professor at the University of Minnesota, thought this a good idea and coordinated a series of three conferences on computerized adaptive testing in 1975, 1977, and 1979, bringing together the greatest thinkers in the field. These conferences energized the research community, which focused on the theoretical underpinnings and research necessary to develop and establish the psychometric quality of computerized adaptive tests.

The first large-scale computerized adaptive tests appeared circa 1985, including the U.S. Army's Computerized Adaptive Screening Test, the College Board's Computerized Placement Tests (the forerunner of today's Accuplacer), and the Computerized Adaptive Differential Ability Tests of the Psychological Corporation (now part of Pearson Education). Since that time computerized adaptive tests have proliferated in all spheres of assessment.

Goals of Adaptive Testing

The choice of test content and administration mode should be based on the needs of a testing program. What is best for one program is not necessarily of importance for other programs. There are primarily three different needs that can be addressed well by adaptive testing: maximization of test reliability for a given testing time, minimization of individual testing time to achieve a particular reliability or decision accuracy, and the improvement of diagnostic information.

...