A procedure to estimate power empirically for goodness-of-fit indices /
The purpose of the present research was to empirically investigate the power of some of the most commonly used fit indices (e.g., GFI, AGFI, CFI, NFI, and Chi-square) in detecting a small effect in a structural equation model. Two models (i.e., a reduced model with one path removed and a complete ba...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
2000.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=728409721&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | The purpose of the present research was to empirically investigate the power of some of the most commonly used fit indices (e.g., GFI, AGFI, CFI, NFI, and Chi-square) in detecting a small effect in a structural equation model. Two models (i.e., a reduced model with one path removed and a complete baseline model) and six (20, 50, 100, 200, 500, and 1000) sample sizes were used to study (1) the effect on fit indices under the two model conditions as sample size varies and (2) the power of several goodness-of-fit indices. Results of the study indicated that all five fit indices are affected by sample size. However, the degree of dependency varied among the fit indices. Of the fit indices studied, the Comparative Fit Index (CFI) exhibited very stable patterns and was less influenced by sample size or generated data types than the other fit indices. In addition, the Normed Fit Index (NFI) had very similar values to CFI's. When the power of the fit indices was analyzed, it was concluded that, as sample size was varied, the power of the fit indices also varied. Results of the study indicated that GFI was the most powerful fit index. AGFI exhibited anomalous characteristics that made it potentially unusable for comparing models. Future research should extend this research to other sample sizes, additional fit indices, and especially the root mean square and the root mean square error of approximation. Also, AGFI's performance should be investigated further. |
|---|---|
| Item Description: | Vita. "Major Subject: Educational Psychology". |
| Physical Description: | xi, 92 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 77-82). |