GROUP AND ITEM INVARIANCE OF ITEM DIFFICULTY PARAMETER BASED ON ITEM RESPONSE AND CLASSICAL TEST THEORIES
The main concern of any assessment procedure is to adopt a measurement approach that will yield valid and reliable test items and test scores on which decisions about the examinees are based. Classical Test Theory (CTT) and Item Response Theory (IRT) are two major measurement frameworks employed in psychometrics. CTT, used over the years, has been theoretically criticized for its inability to solve measurement problems such as test equating, differential item functioning, item banking, and invariance among other. The emergence of IRT as a preferred framework sparked off a debate in the psychometric community on the superiority of IRT over CTT, particularly in the provision of ability estimates and item parameters that are independent of test and sample respectively. This study, therefore, compare the CTT and IRT in terms of item/person parameters. Survey research design was adopted. The sample comprised 1150 senior secondary three students drawn from 50 schools in Abia State using multistage sampling technique. Two parallel Chemistry Achievement Tests (CAT A & B) developed by the researchers were used for data collection. Each instrument consisted of 60 items of a 4-option multiple choice test. Kuder-Richardson formula 20 reliability estimates of CAT A and CAT B yielded coefficients of 0.88 and 0.90 respectively. Four hypotheses guided the study. The item difficulty and person parameters from IRT and CTT were tested for invariance using independent and dependent t-tests at 0.05 alpha levels. Findings indicated a significant difference between item/person parameters of CTT and IRT. Furthermore, the item difficulty parameter and ability estimates of IRT were invariant as against those of CTT. Based on the findings, it was therefore, concluded that, given that the data fits the IRT model used, IRT is empirically superior to CTT. Logical recommendations were highlighted which include that IRT should be used in solving measurement problems.
Ijeoma Joy Chikezie (Ph.D) & Eme U. Joseph (2016)
GROUP AND ITEM INVARIANCE OF ITEM DIFFICULTY PARAMETER BASED ON ITEM RESPONSE AND CLASSICAL TEST THEORIES
The main concern of any assessment procedure is to adopt a measurement approach that will yield valid and reliable test items and test scores on which decisions about the examinees are based. Classical Test Theory (CTT) and Item Response Theory (IRT) are two major measurement frameworks employed in psychometrics. CTT, used over the years, has been theoretically criticized for its inability to solve measurement problems such as test equating, differential item functioning, item banking, and invariance among other. The emergence of IRT as a preferred framework sparked off a debate in the psychometric community on the superiority of IRT over CTT, particularly in the provision of ability estimates and item parameters that are independent of test and sample respectively. This study, therefore, compare the CTT and IRT in terms of item/person parameters. Survey research design was adopted. The sample comprised 1150 senior secondary three students drawn from 50 schools in Abia State using multistage sampling technique. Two parallel Chemistry Achievement Tests (CAT A & B) developed by the researchers were used for data collection. Each instrument consisted of 60 items of a 4-option multiple choice test. Kuder-Richardson formula 20 reliability estimates of CAT A and CAT B yielded coefficients of 0.88 and 0.90 respectively. Four hypotheses guided the study. The item difficulty and person parameters from IRT and CTT were tested for invariance using independent and dependent t-tests at 0.05 alpha levels. Findings indicated a significant difference between item/person parameters of CTT and IRT. Furthermore, the item difficulty parameter and ability estimates of IRT were invariant as against those of CTT. Based on the findings, it was therefore, concluded that, given that the data fits the IRT model used, IRT is empirically superior to CTT. Logical recommendations were highlighted which include that IRT should be used in solving measurement problems.