Date of Award
Doctor of Philosophy
Robert F. Strahan
Golden section search strategies (GSSS), dichotomous search strategies (DSS), and Z-score strategies (ZSS) are simple and robust computerized adaptive testing strategies. GSSS, DSS, and one version of ZSS are similar in that statistical hypothesis testing occurs at each successive testing point in determining the current ability estimates. After each item is administered, the examinee's obtained score is compared with the expected score at successive testing points. If the examinee's obtained score does not exceed a confidence interval of an expected score at a testing point, the examinee's current ability estimate is assumed to be equal to that of the testing point. Otherwise, a hypothesis testing will be conducted at the next testing point and the process is continued until the examinee's current ability estimate is determined and the next item is selected. The three strategies differ in the successive testing points' allocation. Each middle point of successive golden search regions is a testing point in GSSS; each middle point of successive dichotomous search regions is a testing point in DSS; each Z-score estimate evaluated at the previous testing point is the next testing point in ZSS. No hypothesis testing is involved in another version of ZSS in which the current ability estimate is the Z-score estimate evaluated at the previous ability estimate. Results of Monte Carlo studies in three hypothetical item pools and one SAT Verbal item pool showed that GSSS, DSS, and ZSS were computationally efficient, precise throughout the ability continuum, and robust against aberrant responses. Optimal measurement occurs using moderate size confidence intervals. Both versions of ZSS measured well under general conditions. GSSS, DSS, and ZSS provided more accurate and efficient ability estimates than did maximum likelihood estimate strategies (MLES) whenever guessing effect existed. GSSS, DSS, and ZSS were more efficient but not more accurate than MLES whenever guessing was not a factor. For GSSS, DSS, and ZSS, there were no differences in measurement accuracy, but occasional differences in measurement efficiency. GSSS, DSS and ZSS were more robust against guessing and inaccuracy of item parameters and took less time to execute than did MLES.
Digital Repository @ Iowa State University, http://lib.dr.iastate.edu/
Xiao, Beiling, "Strategies for computerized adaptive testing: golden section search, dichotomous search, and Z-score strategies " (1993). Retrospective Theses and Dissertations. 10203.