| Summary: | 碩士 === 臺中師範學院 === 教育測驗統計研究所 === 89 === This study compares robust the WLS estimators with conventional WLS estimator by using a dichotomous CFA model known as the TIMSS CFA model. The robust WLS estimators contain both the mean- and variance-adjusted WLS (WLSMV) estimator as well as the mean-adjusted WLS (WLSM) estimator. Monte Carlo studies show that WLSMV and WLSM perform better than WLS under all three kinds of conditions. In addition, performance of WLSMV and WLSM is very good when different sample sizes and thresholds are considered. But under the condition of changing correlation matrix, both failed to perform satisfactorily. Futhermore, WLSM performs as well as WLSMV, except for one criterion. By judging from the observed chi-square p-values to expected ones, WLSMV performs somewhat better than WLSM.
Besides Monte Carlo studies carried out, this study also provides an example of TIMSS to test this CFA model. Results of CFA analysis show that this model is acceptable for both case Taiwan and case USA. In addition, the results of multiple group CFA analysis show that the parameters of thresholds and factor loadings are invariable over groups. Finally, other findings of this study show that the students of group Taiwan perform better than those of group USA.
|
|---|
