| Summary: | 碩士 === 臺中師範學院 === 教育測驗統計研究所 === 89 === The development of structural equation models have been applied to many fields, especially in psychology and education. However structural equation models may generate improper solution in terms of estimated methods, sample size and data type. This study used simulation testing method to compare (1) the respectively accuracy of estimation of ML, GLS, WLS & ULS, (2) the effect on sample size and (3) the data type. Following are the main results of this study：
1.The estimated averages are similar for ML & ULS when the data type is continuous. The estimated averages are not similar for ML & ULS, when the data type is discrete.
2.In the continuous data, the MSE of four estimated methods tend to be zero when the sample size increases. In the discrete data, the MSE of ML & ULS drop off very slowly when the number of sample size increases.
3. In the continuous data, ML, GLS & WLS may provide accurate results when the sample size are respectively above 100,200,250. Nevertheless. When the sample size is small（n＜100）, they may not provide accurate results and they will have bias.
4.When the data type is discrete, the continuous estimated methods may not provide correct estimator. Moreover, the estimator will not be close to the true value , when the sample size increases.
5.In continuous data , the chi-square does increase when the sample size increase and the accuracy rate does not decrease when the sample size increases. In the small sample size（＜100）, The chi-square stays small , and the accuracy rate is between 0.925∼0.98.
|
|---|
