Hello everyone,
I am in the process of performing a EFA with maximum likelihood, but it seems like the result is unidimensional. From what I understand, no rotation should be done on data that shows unidimension, but how do I then proceed? I cant seem to find much or any information on EFA and unidimension. Anyone have any advice for how I can proceed or any articles at hand?
Regards
Anneli
varimax rotation is recommended for one-dimensional sometimes maximum likelihood should apply only if you have adequate multivariate normality. It should be noted that the variance explained should be greater than 50% and items greater than .50.
Hello Anneli,
A one-factor solution means that you have the simplest possible structure; variables do or don't affiliate (depending on your chosen threshold for salience of loadings) with a single latent variable. You are correct that no rotation is either possible or necessary.
What to do with this?
0. Report pertinent background information:
a. What were the variables?
b. What was the sample?
c. What were the correlations among the variables?
d. Means & SDs for each variable?
e. Did indicators such as K-M-O (measure of sampling adequacy) and Bartlett's test suggest that the data set was suitable for factoring?
f. Your chosen method of factor extraction (e.g., ML estimation? principal axis?)
g. Your decision rule for determining the number of factors to extract.
1. Report the variable-factor loadings as they were computed,
2. Indicate which, if any, variables you believe are insufficiently affiliated with the lone factor to be retained in the solution. Indicate your threshold for salience,
3. Indicate what you think the factor represents, and
4. Discuss the implications/utility of this factor (or this measure--I really wasn't clear about your research question/s).
Good luck with your work!
Thank you Cristian Ramos-Vera and David Morse for answering my quesiton. I am doing a validation study of a scale where ML EFA was used to extract the final items. Since I am doing the validation in a different setting/country with its original items instead of the final ones, I thought the same procedure would be a good first step.
The KMO and Bartlett's test did show that it was suitable for factoring and I chose ML since the study belongs to the social sciences. I did attempt to do a PCA as well, which indicated 2 factors (of items that were negatively worded vs positively worded) that I felt were not an accurate representation.
I will try to see if I can find some previous research and/or articles that I can use for my reasoning in the thesis. Thank you so much for the feedback/suggestions!
- Badges
- Science topic