1
$\begingroup$

I used a questionnaire that usually has a high Cronbach's alpha and that has a scale of 7 answers. I did a before and after (an intervention) evaluation and got a very very low alpha for the before and a negative alpha for the after. Since the 20 people who answered the questions do not represent a normal population (they have problems), the mean score for the "before" evaluation was 1.9 and had a very low sd (i.e., all answers for the 6 questions ranged between 1-3 for all participants). The mean score for the "after" evaluation was 5.9 with a low sd once again (ranging between 4-7).

I am sure there were no mistakes in the coding. Could it be that the standard deviation influenced the results? If not, what other suggestions do you have, and what am I supposed to do with this?

I should point out that the intervention was directed specifically at changing the behaviour we were measuring, so the results other than the alpha are good for us.

Any thoughts would be greatly appreciated!!

Improve this question
$\endgroup$
2
  • $\begingroup$ If you look at a formula of alpha you can see that it is based on the ratio of the average inter-item covariance to the average item variance. The sign is therefore determined by the average item covariance. If all items individual scales in a construct are co-oriented there should be no negative covariances. $\endgroup$
    – ttnphns
    Commented Oct 21, 2014 at 9:55
  • $\begingroup$ Thank you for your comment. What I'm wondering is, with these kind of scales, where you have 7 options, the "valence" of answers 1-3 or 5-7 are similar (within each group). If that's the case then the fact that in both evaluations all participants answered in the same "valence" area, then the alpha doesn't really tell us much, does it? $\endgroup$
    – Misk
    Commented Oct 21, 2014 at 10:21

1 Answer 1

Reset to default
2
$\begingroup$

You haven't given us the correlations among the items, which would help, but it sounds like you probably have a problem with restriction of range, which would lower correlations and hence alpha.

If this is the case, there is little you can do about it on this sample. But it might not be that bad as the mean rose so much.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thank you for answering. Yeah, correlations did not come out significant for obvious reasons....So do you think I should just report it as is and give my explanation for it? $\endgroup$
    – Misk
    Commented Oct 21, 2014 at 10:23
  • $\begingroup$ Yes, that's what I think you should do. $\endgroup$
    – Peter Flom
    Commented Oct 21, 2014 at 11:45
  • $\begingroup$ This is a great answer. Here's a paper that shows how easy it is to get negative reliability estimates under range restriction: journals.sagepub.com/doi/10.1177/0013164411430225 $\endgroup$
    – dfife
    Commented May 20, 2022 at 18:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.