Hello Gjergji Josifi My idea is to think this way and search about it.

Principal Component Regression (PCR).

PCR principal component regression is a fast, efficient and optimal regression method based on covariance and is usually used when there are many explanatory variables that are likely to be correlated with each other in the study.

PCR analysis is a method that reduces variables used for prediction to a smaller set of predictors. It then uses this reduced set to perform a regression. These reduced sets, which are made from Independent Variables, are called Components.

PCR combines the features of Principal Component Analysis (PCA) and Multiple Regression. First, it obtains a set of factors or components that explain as much covariance between independent quantities as possible. This part is what PCA does. In the next step, the multiple regression model is created between dependent quantity values and components.

https://graphpad.ir/principal-component-regression-prism/

Thank you Mr Ghoodjani i appreciate the advice

In general PCR is not recommended for working with attitudinal variables for technical reasons related to assumptions about the underlying measurement model.

Instead, I suggest you use Principal Factors for your extraction method, followed by an oblique rotation to allow for correlated factors.

Thank you David L Morgan . After that does point 2 and 3 make sense?

Yes, especially if your factor analysis shows what is known as "simple structure," where each variable loads on one and only factor.

Once you do the factor analysis (or any similar technique) to choose variables to be in certain factors, calculating Cronbach's alpha with those same data is problematic because you have already chosen those variables because they hang together and Cronbach's alpha is related to "hanging together" (and things like the number of items).

Using factor analysis with Likert scaled items is not a recommended option. The measurement level is not as required for factor analysis which requires interval level. Likert scaled variables are at most ordinal level. Use somehing more appropriate like graded response model, Mokken or another IRT model. These models can be calculated with R.

I will post an SPSS script to run several IRT models when you have installed the R extension (and all required packages) to SPSS. I will need to clean an SPSS syntax and make sure the project it was used for can not be identified.

While I agree that IRT more directly addresses the ordinal nature of Likert-type items, there are methods in the factor analysis (FA) tradition that do this too. There is, for instance, FA of polychoric correlation matrices.