Hi everyone, I apologise in advance for the rather probably a daft question.
I am currently performing Structural Equation Model (SEM) with AMOS, and also use SPSS for the descriptive and "simpler" analyses to the dataset. One analysis that I am performing on SPSS is the Pearson correlation to have the correlations between variables. And, as you may have known, that we can also obtain correlations of variables of a model in AMOS.
I noticed that the correlation coefficients between variables are different between SPSS output (Pearson) and AMOS output, and the difference is rather significant (some have 0.2 difference in correlation coefficients). I also found this in some papers. So my questions are, why are AMOS and SPSS showing different correlation coefficients within the same variables? Is that normal (if so, why?)? Or are they (SPSS-Pearson correlation and correlation obtained in AMOS) different "thing" of correlation?
I would appreciate and be grateful of any responses to this, thank you so much.
Regards,
Rizky
Have you compared the sample sizes reported in SPSS vs. AMOS for these correlation analyses? These differences could be due to differences in how missing values are handled by the programs (e.g., listwise or pairwise deletion versus full information maximum likelihood estimation). The different approaches to handling missing scores may result in different sample sizes and differences in the estimated correlations.
Thank you for your reply Dr Christian Geiser and sorry for the late reply. As far as I know, both SPSS and AMOS reported the same sample size in what I have been analysing, as pictures attached. Do you have any further advise on these differences in the reported correlations between SPSS and AMOS? Thank you.
Since you did not show the model, it is difficult to make an educated guess, but these are my 2 ideas:
1) Are you sure that you calculated plane correlations or may there be regression coefficients? Are there other variables associated with your targetvariables?
2) Do you compare only manifest variables or do you compare correlations between manifest and latent variables?
It would be best if you could share the complete AMOS model.
Were the correlations in AMOS estimated under a specific model or simply as sample correlations without a specific model structure? Correlations that are estimated based on a specific overidentified model may differ from observed (sample) correlations due to model-implied restrictions.
As Rainer Duesing wrote, without more information on the model/type of analysis that you ran in AMOS, it is difficult to make an informed guess.
- Badges
- Science topic