Application Of Item Response Theory To Item Banking Of College English Placement Test

Item response theory (IRT) is a collection of statistical techniques that is being widely applied in the field of psychological and educational measurement. And item banks are also widely used in high-stake, large-scale assessment programs by an increasing number of test developers. An item bank will be considerably potential if it is constructed based on item response theory because IRT has valuable characteristics. This thesis aims at applying IRT to develop an item bank calibrated with item parameter estimates using data from Computer-Based College English Placement Test of Hunan University. Item discrimination and difficulty parameters were estimated and transformed on a common scale using IRT, and together with other significant information, an item bank were developed. Examinee scores could be interchangeable when administered with different sets of items from such a bank.In order to establish a standard scale against which as many as possible items could be calibrated, two test designs—the fixed forms examination and its variation, the linear-on-the-fly tests design—were used in the Computer-Based College English Placement Test. In the fixed forms examination, two fixed alternative forms with common items were administered to two nonequivalent groups of 2,461 examinees, respectively. In contrast, in the linear-on-the-fly tests design, each of the rest 2,830 examinees received a unique set of test items, each of which was strictly matched to the same content and statistical specifications.The main results were as follows: All item parameters of the two alternate test forms were estimated and transformed to a common scale using the two-parameter logistic model and concurrent calibration, thus forming a calibrated"item bank"and providing a standard scale for the rest items in the linear-on-the-fly tests design. The common-item equating to a calibrated item pool design, the EAP (Bayesian estimated a posteriori) estimator and fixed parameter calibration were used to deal with the item parameter estimates in the linear-on-the-fly tests design, which proved to be effective and accurate in calibrating item parameters in small samples, leading to reasonable and plausible estimates results. An item bank of 747 items of listening and reading comprehension has been developed and evolved successfully based on item response theory. Each item in the bank is accompanied with detailed information, such as item parameter, item characteristic curve, item information function, content, function, and specific knowledge or skill it is to measure. The bank size is reasonably big, and the items have considerably high discriminations and intermediate difficulties.The research on the application of IRT to the development of item pools, especially pools for computerized adaptive testing is still at its preliminary stage in China. Only a very limited number of tests have constructed their IRT-based item pools, such as HSK (Chinese Proficiency Test) that used the one-parameter logistic model and Economist Qualification Test that used the graded response model. This study used the two-parameter logistic model to have developed an item bank of 747 items using data from one-off examination, which has the potential to greatly improve the assessment of new comers in colleges and universities, facilitating placement testing and teaching, and maintaining consistency, stability, comparability, and justice over test forms and over years and also provides bases for computerized adaptive testing.