I have 6 varaibles in my thesis model among which for 4 variables I have constructed my own scale. The other two varaibles I have taken from previous literature. now I am confused do I have to perform EFA and the first order factor analysis on those two items on as well?
Since you already have a specification for how variables go with their associated scales (factors), you should use Confirmatory Factor Analysis.
Hi , EFA to be generally used for the unexplored factors . Since you already have taken the items by collecting data , you need to do first perform the validity test of the instrument and then you can perform EFA/ CFA.
If you have taken the items from the literature better to go with CFA or you can go with EFA.
Note : You need to do the EFA/CFA for all six items irrespective of the source of these items
Regards,
Dr Uday Bhale
I'd begin with parallel analysis to determine how many factors you most likely have. It's possible, for example, that one or more of your intended factors are partially or fully redundant to the pre-existing measures or even that there's full or partial redundancy between them. I'd go with EFA, because empirical fact trumps theory.
Paul Max Kohn. As I am working on my thesis I have total of 6 variables. Four Variables I have developed my own scale based on their sub dimensions. the overall items in my study are 151 which majorly contribute to the varaibles for whom I have developed my own scale. I performed EFA on all 151 items out of which around 45 items got due to low factor loading after keeping the threshold above (.60). I am confused, what after I have to do exactly. As I transfer the EFA table to pattern max builder in Amos I get an error of having negative eigen values. Moreover I am confused about the first order and second order factor analysis, when I am supposed to perform them. Do I have to perform that anaysis on the data collected for EFA or the data on which I have finallly perform CFA.
When performing a factor analysis, the most consequential decision one makes is how many factors to extract. Historically, there have been numerous strategies for making that decision, including theorization.
The current consensus is that parallel analysis is the closest to definitive. It is a bootstrapping of eigenvalue calculation which enables one to compare one's actual eigenvalue against the 95th percentile value of some very large number of bootstrapped eigenvalues, typically 1,000. I think of this as a test of significance, alpha=.05, for the eigenvalue corresponding to the estimated correct number of factors.(The number of factors corresponding is typically displayed in the program output).
Higher order factor analysis may or may not apply to a dataset. I'd do parallel analysis on the intercorrelations of the factors identified at the previous order to find out.
- Badges
- Science topic
- Similar topics
- Psychology