| Summary: | 碩士 === 國立臺中教育大學 === 教育測驗統計研究所 === 95 === MUML (an estimation of ignoring unbalanced facts; so called pseudobalanced procedure) was proposed by Muthén in 1990. It aimed to estimate the two-level covariance structure model for unbalanced data. Hox and Maas (2000), Hox (2001), Lawrence (2000) and Yuan (2005) all studied MUML estimation. So far, there was no study which indicates explicitly on how is sample size and either balanced or unbalanced datasets influence the accuracy of MUML estimation.
This study used two-level structural equation model (SEM) of Muthén & Muthén (1998) as the base model of the study, which includes between part and within part. Datasets are divided into balance and unbalance sets, and Mplus3.1 (Muthén & Muthén, 1998-2004) as the analyze tool. We explored on how is sample size, number of groups and unbalance level influence the accuracy of MUML estimation, and proposed a unbalanced index “CUI”. According to this research, the major findings included:
1. Using MUML estimation to analyze balanced datasets, the accuracy will de- crease as the sample size decreases, but the numbers of groups have no obvious influence on the accuracy.
2. In order to obtain higher accuracy on MUML estimation with balanced datasets, the sample size should be greater than 400.
3. Using MUML estimation to analyze unbalanced datasets, the accuracy will decrease as the sample size decreases and the accuracy will decrease as the unbalance level increases. Moreover, as the sample size becomes smaller, it decreases the accuracy of MUML estimation rapidly.
4. In order to obtain higher accuracy on MUML estimation, for unbalanced datasets, the CUI should be less than 6.
Keyword: multilevel SEM, unbalanced data, MUML estimation, CUI
|
|---|
