Research On Construct Validity Of NMET Cloze Test Using "Retrospective Report" Approach

The research of validity has long been a hot topic in the language test conferences and will continue. Among the types of validity, the construct validity draws more attention, for it is the foundation of other validities. Linguists from home and abroad did many researches on validity of cloze test from different aspects.Cloze test is a traditional test type, it is considered to assess the integrating abilities of test takers. As one of the most important test types in National Matriculation English Test (NMET, for short), the construct validity of cloze test has long been a hot topic.The designers of the NMET test believe that when the test-takers are conducting the cloze procedure, they should figure out the answers from the context information and employ the skills that are supposed to use. Therefore, how do the test takers figure out the answers and whether the actual performances are in line with the expected operations calls for attention. It is quite necessary to study the "mental process" of test takers so as to reveal the actual answering performances. However, few studies have been made on the construct validity of cloze test in NMET using "retrospective verbal report" approach, which is what this study focused on.Based on the previous research findings, three questions are raised in this study:1) Are the actual answering performances of test takers in accordance with the expected answering operations? If there are discrepancies, then to what extent are they in line with each other?2) Which levels of textual information are most frequently employed by the test takers? Do the test-wise strategies often lead to correct answers?3) Do the high group subjects employed the expected skills more effectively and frequently than the low group subjects? Are the differences significant?In this thesis, it is the "Introspection" approach that was employed, hoping to validate the construct validity of NMET Cloze Test of 2007 (Paper I) by evaluating the consistency of actual answering performances of test takers and the expected operations. 30 subjects from ShanXi DaTong Nol Middle School participated in the experiment. After the recording and data collecting process, Excel and SPSS 14.0 were employed to analyze the data, after conducting Reliability check, the following findings are concluded.1) The experiment performance of the subjects are basically in accordance with the expected performance operations, thus it could be concluded that NMET cloze test has high "construct validity" in context information comprehension and manipulation as well as vocabulary recognition and discrimination. And the two facets take approximately equal importance in this test.2) The subjects were employing sentence level and textual level information most frequently. However, there is discrepancy between the two levels information employment and the expected operations. Based on the research, it could be concluded that the test-wise strategies could not substitute the expected reading comprehension skills, and the chance of getting the correct answer by randomly guess is very small.3) Independent sample T Test showed that there are significant differences between the performance of high group and low group. The test takers in high group outperformed the low group subjects in all information level categories.The research put forward suggestions regarding the cloze test constructing, language teaching and learning, hoping to be of reference to teachers and test designers. Though this tentative study has provided satisfactory evidences on the construct validity of NMET cloze test, it is in no way conclusive. The researcher of the present study believed that further studies are needed to validate the construct validity of NMET cloze test using other methods in "Introspection" category, based on larger subjects and more regions.