Can I perform Principal component analysis on SPSS and then move to Smartpls to test hypotheses? (Higher and lower order constructs are all reflective).
Yes, it is possible to perform Principal Component Analysis (PCA) in SPSS and then use the results in SmartPLS for hypothesis testing. Here's a general overview of the process:
It's important to note that while PCA is a dimensionality reduction technique, it is not a direct test of hypotheses. SmartPLS, on the other hand, allows for testing hypotheses using SEM techniques. By combining the results of PCA with SmartPLS, you can leverage the dimensionality reduction insights from PCA and then conduct hypothesis testing using the SEM framework in SmartPLS.
Remember to consult the user guides and tutorials of both SPSS and SmartPLS for detailed instructions on conducting PCA and performing hypothesis testing in each software package.
Hello Tarek,
Please note that PCA (principal component analysis) presumes that: (a) all observed variance is common variance, which implies: (b) no error or variable-specific variance exists. These are very different perspectives on reality than those of common factor analysis.
As a result, unless your number of variables being analyzed is large, PCA is likely to yield loading estimates that are biased upward from what common factor analysis would yield. Why is this important? Because two people analyzing the same data set, one via PCA and one via common factor analysis, could well come up with different interpretations of factor structure and identification of salient vs. non-salient variable-factor affiliations.
Good luck with your work.
Thanks Dr. Morse. The PCA is made to investigate the factor structure for each construct separately. Does this make a difference? If not, can I do EFA using principal axis factoring with promax rotation on SPSS as a substitute for PCA? Thanks alot
Hello Tarek,
Almost any set of measures involving human responses would be better treated via common factor analysis than PCA.
When you're starting off with possible indicators, it is ok to evaluate proposed factors individually. However, unless you simultaneously evaluated the multi-factor model, you'll have no evidence for convergent validity (across factors which ought to be correlated, by your framework) or discriminant validity (across factors which ought to have little or no correlation, according to your framework).
So, even in EFA, evaluating multiple factors (and using some sort of oblique rotation, such as promax), is a better way to go. And yes, principal axis factoring is fine. Note that, were you using covariance-based method for your confirmatory work, maximum likelihood extraction would make more sense.
Good luck with your work.
Hi, please check the recent research applying PCA for early plant diseases detection:
Article Abaxial leaf surface-mounted multimodal wearable sensor for ...
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Probability
- Probability Theory
- Reliability