need help with transformation

Hey all

I have questions about transforming variables. I'm sorry if this comes across as stupid.

I have a set of variables (45 variables with 5-point likert scale). After screening these data set, 39 of them were transformed and 5 weren't.

My question is, to do factor analysis of these variables, can I submit all of them at once (both transformed and original), or do I have to do anything else?

thanks in advance



TS Contributor
Why were they transformed? What were they transformed to?

I would submit the originals.....I'm not a big fan of transforming variables.
thanks for the reply JohnM. They were transformed to correct normality (they have moderate to strong negative skew. mostly beyond 3.5 standard score. Some were transformed using LG10 and some using SQRT, depending on its severity.

I also did the analysis with the original - this is a second round analysis to make sure the solution is similar, despite having nomality issue with my variables.

Some suggest that multivariate assumption is not in force (I remember I read this somewhere :D ) when it comes to exploratory FA, but since i will be using the same data set for confirmatory FA. I thought i should address the assumptions before hand.

comes back to the - if a data set contained a mix of transformed and original, is it appropriate/correct to do the factor analysis using all variables at once or do I have to address anything else prior to the analysis.




TS Contributor
I would do the factor analysis with all variables either transformed or in their original state.

- although I have only a little experience with it (in grad school), I would imagine it would make interpretation very very difficult if they were a mix of original and transformed variables.

If you want to analyze the mixed variables, then I would take the mixed variables and transform them all to z scores (standard scores) to at least get them on the same scale.

In any event, taking multiple approaches to analysis is always a sound idea, and if you get consistent results across the different methods, then that just gives you a little more confidence in your results...