Optimized Approach To TEM Band 4 Test Equating: A Tentative Research

Large-scale language tests are constantly confronted with difficulties in guaranteeing the comparability or interchangeability of scores on different test forms. Test equating, the statistical process used to adjust scores on different test forms so that the scores derived from the two forms will be directly equivalent after conversion, is thus deemed necessary to overcome measurement limitations, to make different test forms interchangeable, to ensure test consistency and decision-making objectivity, as well as to be fair to high-stake examinees.This thesis starts with an introduction to prevalent test equating theories, typical equating designs and representative equating practices applied in large-scale language tests in China and other countries. In particular, special emphasis is laid upon the application of CTT (Classical Testing Theory) and IRT (Item Response Theory) in the major equating approaches, including mean equating, linear equating, the equipercentile method and the IRT equating method.With an eye on the usefulness, feasibility and applicability of test equating approaches in TEM Band 4 (Test for English Majors Band 4), the thesis sets out to conduct a tentative experiment with the recommendable test equating design for TEM Band 4 through an empirical research. Two groups of students in a relatively small sample population took part in two separate experimental tests with common items, and the scoring results were analyzed, computed and discussed in the different statistical models constructed on the basis of the mean equating approach, the Tucker and Levine Observed Score Methods in Non-equivalent Groups, the equipercentile equating method and the IRT Rasch Single-parameter Equating Approach. The equating results thus obtained are illuminating and shed light on the appropriate equating design that caters to the realities of the TEM Band 4 test.On the basis of the empirical study, the thesis concludes that, to improve the validity, interchangeability, objectivity and consistency of the TEM Band 4 test, efforts to make the TEM Band 4 test forms interchangeable are worthwhile and long due. Although the optimum approach still merits further empirical studies, TEM Band 4 Test can be equated well by borrowing from existing equating practices widely accepted by the test measurement community. The paper also recommends that, in designing the equating method for TEM Band 4 Test, common-item non-equivalent groups design should be contemplated due to its reliability, since experiment data reveal the CTT-based Tucker Observed Score Linear Equating Method and Equipercentile Method are both as effective as, if not better than, the Rasch single-parameter equating method in a number of aspects. The paper also contends that CTT alone cannot constitute the sole basis for judging a particular item as a suitable TEM Band 4 formal test items, and both CTT and IRT equating approaches should be contemplated in the equating design for the TEM Band 4 Test. It is therefore quite essential for TEM Band 4 Test to combine the two approaches effectively, minimizing the shortcomings of both approaches while maximizing their respective strong points.