Hi,
while calculating Cronbach's Alphas for some of my constructs in a questionnaire I am obtaining several Alphas larger than 1 (e.g. 1.1, 1.3, etc...). How am I supposed to discuss such values? Is this good because larger than 0.7 or is this bad because Alpha is generally considered in a range 0-1?
Thanks
I do not see how anything you describe could lead your alpha to be greater than 1, but it can be wise to exclude cases with missing data.
As for using variances and covariances as opposed to correlations, they will lead to the same standardized value, which is what you want to report. I gave you the formula using the average correlation because that is the easiest one to compute.
Are you doing the calculations by hand? If so, you must be making an error.
Here is a link to a simple formula for calculating alpha using the average correlation among the variables:
http://www.analytictech.com/ba762/handouts/alpha.htm
If you do a little more algebra on that formula, you will see that numerator has to be larger than the denominator, so alpha must always be less than one.
Thanks Jesus, I had found those links before, I was struggling to find some more academically grounded perspective.
Thanks David, I am not doing the calculations by hand, the formulas are apparently correct (although I am using variances as inputs). I realized that the problem is that a some respondents did not answer to one or two items that concur to the assessment of the variable, and this is causing the distortion of Alpha. I am solving the problem by excluding them from the calculations.
I do not see how anything you describe could lead your alpha to be greater than 1, but it can be wise to exclude cases with missing data.
As for using variances and covariances as opposed to correlations, they will lead to the same standardized value, which is what you want to report. I gave you the formula using the average correlation because that is the easiest one to compute.
It is difficult to say something of interest after the comments of Prof Morgan.
In my little experience with Cronbach´s alpha, I had an unexpected value of -2,67 in a quite homogeneous group of answers. After trying some procedures, lastly I changed the codification of one of the variables (1=no/2=yes, instead of 1=yes/2=no), and alpha turned into 0,73.
HTH
Regards,
Emilio
Could you please tell me. Cronbach alpha is meant for testing the reliability of data or adequacy of the sample?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire