I am doing a pilot study on software professionals with a sample size of 139.
My model consisting of two IV (X1,X2) and their consequent relationship with three DV (Y1,Y2,Y3). All these variables are taken as latent variables.
On conducting the path analysis in AMOS the relation between IV and DV are found to be significant at the level of 1% and 5% and most of the beta weights are above 0.5 value. Whereas on conducting the regression analysis in SPSS taking the composite scores of all the items of individual variables, the beta coefficients have found to be extremely low and insignificant in some cases. The model fit is moderately fine for AMOS (RMSEA =0.06) but the r squared values in SPSS are very low ranging only from 0.007 to 0.051.
Excerpt of my analysis
Results AMOS(loading of Path Anlysis) SPSS(beta value for linear regression)
X1-->Y1 0.455 (p=0.001) 0.247 (p=0.004)
X1-->Y2 0.646 (p=0.001) 0.240 (p=0.005)
X1-->Y3 -0.295 (p=0.021) -0.073(p=.394)
X2-->Y1 -0.577 (p=0.001) -.119(p=0.167)
X2-->Y2 -0.681 (p=0.001) -.278 (p=0.070)
X2-->Y3 0.694 (p=0.001) 0.312 (p
Basic linear regression uses OLS method to obtain beta values. AMOS is using maximum likelihood. That could be a reason for the difference. Could you post your AMOS model? If you AMOS models is allowing your predictors to covary then you will get different estimates from OLS where predictors are assumed to be independent.
The algorithm or estimator used in SPSS os Generalized Least Square (GLS) while in AMOS the algorithm is Maximum Likelihood Estimator (MLE). The MLE is much more efficient than GLS. So for any difference, the result from Amos is more accurate.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS