I did a factor analysis in SPSS on 5 variables (ML method, since this is the method that AMOS uses) and saved the factor scores using the regression method. This factor Y, correlated with variable X .77.
Then, I do the same thing in AMOS, i.e. I create a factor (Y_a) with 5 indicators: all factor loadings are nicely the same as I got in the factor analysis in SPSS. So far so good.
Then I correlate the factor Y_a in AMOS by drawing a covariance path to the variable X (since this is an observed variable I create a latent variable with X as single indicator without error). However, now the correlation between the Y_a and X is .94.
How is this possible??
Hi,
SPSS product is AMOS. And AMOS is the special software for factor analysis. Thats why you would have been get different correlation values. It was as follows:
In SPSS, the factor correlated with the variable x was 0.77. But, the same thing you have done in AMOS, the correlation was 0.94, This was from the software AMOS. So, i concluded that the AMOS software correlation value is best one.
You follows the AMOS software for you analysis.
Regards
I suggest you post your question to the SPSS mailing list (see link below).
http://spssx-discussion.1045642.n5.nabble.com/
AMOS is more accurate as AMOS analyse or looking to data based on items correlation under each construct.
AMOS is the latest statistical analysis, so this software is much better than SPSS. However, if you do not have either mediator or moderator variable, then you can just analyse using simple regression through SPSS software. Regarding the results, it should be the same as long as you did it correctly.
You should understand that Amos is develop to cater for limitations in SPSS especially when dealing with latent constructs. What you obtained is exactly the reason why you should employ Amos.
You can do a quick check on the AMOS results by calculating the "reproduced correlations." Just multiply each of the loadings in the solution by .94 and see how close the results are to the observed results.
Thank you all for your input. David, what exactly do you mean by your suggestion? Multiply the loadings in the solution from AMOS by .94 or from SPSS?
Could other influencing factors be:
- the variance of the latent construct? In SPSS, the variance of the created variable Y is not 1 (mean is zero).
- the fact that saving factor scores via the regression method, a lineair combination of the underlying indicators is used to compute the scores (via the factor score coefficient matrix). When modeling the correlation in AMOS, this of course does not happen.
Or am I completely wrong about this?
All the estimated correlations between the single X variable and the variables associated with factor Y can be reproduced as a product of the two standardized loadings on the factors multiplied by the correlation between the two factors, which is .94. Since one of the loadings is fixed at 1.0, that gives you .94 times each of the other loads to reproduce the correlations.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS