Dear researchers and Professors
I am going to run exploratory factor analysis for a questionnaire with 16 items. Prior to running an EFA, it was calculated the internal consistency with Cronbach's alpha. Alpha was estimated .41!
I have two questions:
1- Can I run that EFA?
2- If your answer is “No” (because of low alpha and inadequate internal consistency), what is your comment?
Sincerely yours
Hamid
You can certainly run the EFA. One possible reason for a low alpha is that your measures do not all fall on the same scale, which would show up in the EFA as two or more distinct factors.
Before doing that, however, I would check for two other common problems (which can both affect your EFA).
1. You have negatively correlated items that need to be reversed.
2. You have problems with missing data so that some of your correlations are very low.
You can certainly run the EFA. One possible reason for a low alpha is that your measures do not all fall on the same scale, which would show up in the EFA as two or more distinct factors.
Before doing that, however, I would check for two other common problems (which can both affect your EFA).
1. You have negatively correlated items that need to be reversed.
2. You have problems with missing data so that some of your correlations are very low.
I agree with Prof. Morgan. However, it is also possible that your sub-scale has few items. Is this the case?
You can always run EFA . Reasons for getting low alpha is already explained by David. Their are some pre conditions for EFA :
1. Multi-collinearity exists in data
2.KMO should be greater than 0.6.
3. Bartlett’s Test of Sphericity must be less than 0.05.
For running the EFA, I think a pre condition is that the Cronbach's alpha of all sub-scale are high.
Dear Hamid,
You have been given some excellent advice above, starting with David's. I would offer, however, that it's not really a precondition of EFA that you have high Cronbach's alphas (this is really what justifiably concerned you in your original question), but that a low alpha suggests that EFA is probably worth conducting (your hunch) and that a desirable outcome of the EFA is that you can group your items so that each group has an appropriate Cronbach's alpha. What is appropriate is not always what is highest when it comes to these alphas.
Nicola raises an important point, namely that a small number of items can produce a relatively low Cronbach's alpha. This can occur even when those items are quite closely related to each other simply because of the mathematics behind the Cronbach's alpha. Conversely, it is possible to get quite a high alpha with a large number of items without all of those items being strongly related to each other. Your alpha of .41 with 16 items is certainly troubling (something like .80 or would have been nicer), but an alpha of .65 would probably be OK for 4 or 5 items.
A really high alpha with a small number of items might be troubling because it would indicate that all of those items are repeatedly tapping into pretty much exactly the same thing and that the underlying construct is not being represented with sufficient complexity.
One final comment: Do make sure that, as well as things like appropriate results for your KMO index and Bartlett's test of sphericity, you have a sufficient number of participants to run factor analysis. With 16 items, ideally you should have at least 160 participants.
All the best with your research. I hope you obtain some helpful outcomes.
Robert.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire