It is recommended to make the exploratory factorial analysis with prominent rotation if its factors are correlated, also it has to see if it presents indicators in negative redaction, significant KMO bartlett, to fix the distribution of normality of the data.

Part of your problem may be that you are using PCA with orthogonal rotation, and neither of these techniques is very good match to typical questionnaire data. I recommend that you try Principal Factors Analysis with oblique rotation (i.e., correlated factors).

@David Morgan my dear it is a pleasure to read your expert input once again .

Your suggestion about doing an obliqie rotation on the output unrotated factors will resullt in non-independent rotated components (factors)

However, the questioner wants to eventually use the the final rotated factors as independent (and sepatate) variables for further use in regression analysis. He actually made that clear. In effect, he really needs to do orthogonal rotation.

From my experience, the researcher needs to know that it is the data supplied to the analysis that actually determines the number of output factors .

This is mathematical linear algebra issue.. It is also the linear algebra solution that yields the communalties of the variables and the eigenvalues which are the variances of the components.

So although the researcher my wish to deal with 8 factors, the data set contains more or less than that.

The Principal Components Analysis determines the number of factors in a data set not the researcher.

This is a common mistake made by students .

Sorry for my lengthy explanation.

There is no problem with using correlated factors in regression analysis, because regression is designed to handle correlated independent variables, so there is no requirement to do an orthogonal rotation.

Also, both Principal Factors and Principal Components only determine the number of factors due an external assumption, such as eigenvalues greater than one. Other than that, they will find as many factors as there are degrees of freedom.

Thankyou David L Morgan Harold Chike Cristian Ramos-Vera

But as i am very much beginner in this and doing this for an assignment. So unable to understand it to the core. could you please me in detail for the entire process. it will be really helpful

@Siddharth Tiwari, you have actually completed the Principal Analysis task yourself. That was fine.

The new information from our discussion is that the number of outputs is not 8 as you had wanted.

Use the number of output factor which were statistically determined by the data you yourself supplied.

Study each of them and determine its meaning using its significantly large factor loadings. Give an identification name to it.. The total number of these identifed groups is actually the total number of factors you should have based on the data you supplied. Then use their factor scores as the new and surrogate variables for your regression analysis .

Good luck.

just do one thing. This SPSS file transform into CSV file then run on SEM-PLS. you will me amassed.

I would suggest to go back at your conceptual model and see if your statements are truly reflecting your construct. It is important to think theoretical, rather than focusing on numbers.

Harold Chike Hello Harold, I believe that your original concern about correlated IVs in the regression was well taken. If that occurs something like ridge regression or Elastic Net algorithm could be used to cure that. Nice R programs are available in R for Everyone 2nd Ed. by Jared Lander. Code and examples are IMO good. I would be inclined though I think to do a regression on principle components because the resulting IVs are orthogonal. However the two regression procedures are perhaps what David L Morgan had in mind. Hastie and Tibshirani at Stanford have a new elastic net program that does everything it seems but prepare your morning coffee. Code is available from their website. Best to all, David Booth

David, you remember that the reason for preference of orthogonal projection is to maximize loading of variables on the components they best belong. The resulting components automatically become independent. There will then be no problem of any variable loading appreciably on two separate components.

Besides, Siddharth Tiwari wants to use the SPSS and not the R package

Thank you David for calling my attention again.

No, I was not referring to advanced procedures in regression analysis, but to the basic approach to calculating partial coefficients in Ordinary Least Squares regression. There certainly can be problems with multi-collinearity if there are high correlations among independent variables, but I am not aware of any assumptions that the variables in OLS should be completely uncorrelated.

David Morgan, you probably have assumed that the researcher plans to load the output components from the components analysis back into regression analysis in which they will play dependent and independent variables roles. If that is the intention, your observation is in order.

However, the researcher might have wanted to create new variables of which the components are surrogates for further investigations not connected to the other components previously obtainainrd.

That is where orthogonal projection becomes handy.

Moreover, it is common practice to create surrogate variables using

Principal Components Analysis

But my experience is that the labels attached to "surrogate variables" inevitably imply concepts that are theoretically related to each other. To my way of thinking, creating variables that do not reflect the underlying relations between the concepts they are supposed to measure is artificial at best.

would you please explain your problem to give you help, even msg. privately as you wish

David Morgan, let us avoid derailing from the issue at hand.

You are now quarrelling with issue of surrogate variables. Many disciplines make use of surrogate variables irrespective of any contrary opinion.

But get one thing clear. Remember that when components have been orthogonally rotated, they are at right angles to each other, implying little or no correlation.

However, you are entitled to your opinion. But many disciplines have successfully carried out valid research with surrogate variables.

Following