A STUDY OF VALIDITY OF ESTIMATION IN MULTILEVEL STRUCTURAL EQUATION MODEL

Abstract

The purpose of this research was to study a validity of estimation methods in Multilevel Structural Equation Modeling under conditions of sample size variation at each analysis level. This research used a Secondary analysis of a data from the project of learning opportunities in mathematics and science of Mathayomsuaksa 3 students of the Institute for the Promotion of Teaching Science and Technology. The analysis used 30 samples from 9 conditions comprising level 2 sample size of 50, 30 and 15 schools in combination of the level 1 sample size of 10 and 5 persons a school, and the actual students of the selected school. Each conditional data was analyzed by 2 methods of estimation, Full Information Maximum Likelihood (FIML) and Robust Maximum Likelihood (RML), and checked validity by considering estimates of factor loading, path coefficient and cross level coefficients by comparing with the estimates and the parameters using t-test, confidence interval and range. The results were as follows : 1. The validity of the parameter estimation between the FIML and RML when level 2 samples consisted of 50 and 30 schools and level 1 samples consisted of 5 students, 10 students and actual student of the school, both of the two parameter estimation methods gave the same estimations result of Factor Loading, Path Coefficient and Cross Level Coefficients and both estimations were closed to the parameters. So the validity of the 2 parameter estimation is not statistically difference. In the condition of the 15 schools of level 2 samples and actual students the school of level 1 the RML parameter estimation was complete but the FIML parameter estimation was incomplete. 2. The estimate when the level 2 sample size was 50 and 30 schools, and the level 1 sample size was 5 students, 10 students, and the actual students of the school, both FIML and RML parameter estimation gave an valid estimation result of Factor Loading, Path Coefficient and Cross Level Coefficients. However for the level 2 samples of at least 15 school, the RML parameter estimation gave a more valid estimation than the FIML. 3. The estimate when the level 1 sample sizes was at least 5 students a school, the FIML and RML parameter estimation gave an valid estimation results of Factor Loading, Path Coefficient and Cross Level Coefficients.