A Simulation Study of The MSPLIT Estimator's Sample Behavior
| Autor: | 張嘉倪 |
|---|---|
| Rok vydání: | 2010 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 98 Reliability concerns about consistency and stability of test scores. A certain individual should have same or similar testing scores when he/she took the same or parallel test several times under standard testing conditions. Nevertheless, due to the difficulty to have the same group of examinees to participate two parallel form exams, Flanagan (1937) and Guttman (1945) suggested adopting the split-half reliability to estimate test reliability. Although the program developed by Feng (2008) can calculate all split-half reliabilities of a test and identify the maximal split-half reliability of the test, it consumes a very long time to complete the job for tests with item number larger than thirty. Callender and Osburn (1977) proposed the MSPLIT method to estimate the maximal split-half reliability of a test. In theory, this method is trying to find the particular split-half which is most likely to maximize the split-half reliability of a test. The research explored the sample behavior of the MSPLIT estimator by controlling the sample sizes of examinees with tests of various items selected from an achievement test. The results showed that the MSPLIT approach estimated the maximal split-half reliability of tests very well and the MSPLIT estimators behaved regularly as sample sizes changed for different tests studied. The MSPLIT estimator overestimated the maximal split-half reliability of a test when sample sizes are small, however, it slightly underestimated the maximal split-half reliability of a test when sample size exceeds a fifth of total number of examinees. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|