6

I have gathered data from 130 questionnaires and I have 30 variables.

Cronbach's alpha is badly negative. I have checked everything I could think of, but the result did not change.

SPSS file is attached

http://www.wikiupload.com/9M9YHEV6I4Q1HPR

Community
  • 1
shiva
  • 61
  • 1
  • 1
  • 3
  • Could you be more specific about what you checked? For example, did you check if some variables should be recoded (e.g. items with a negative formulation)? – Gala Dec 07 '12 at 07:28
  • Yes exactly, I have checked If there is any negative formulation or any mistake in typing data but I did not find any thing wrong – shiva Dec 07 '12 at 08:02
  • http://www.ats.ucla.edu/stat/spss/library/negalpha.htm – ttnphns Apr 09 '15 at 07:54
  • That may means your questionaire has a lot of open ended questions. please reconstruct it to a scale/ likert scale – Oloye Abiola May 09 '15 at 07:21

5 Answers5

8

You have only weak to very weak correlations (and sometimes negative) between your variables. Your alpha value is negative surely because the mean of all the inter-item correlations is negative. Maybe you can use a factor analysis to check the factorial structure and correlations between the extracted factors? But given the data you provide, I think it will no be very helpful, except maybe if you have a theory to guide your interpretation of the results. Do you have a theory or prior results predicting that your variables should correlates positively (i.e. allowing the use of Cronbach's alpha)? If so, then your results are pretty strange...

Alric
  • 91
  • 3
4

As @alric said, all your correlations are weak. I'd conclude that these questions are not a scale, should not be added together or combined in some other way, and are each really separate entities.

Peter Flom
  • 94,055
  • 35
  • 143
  • 276
3

This almost always means that you have some variables which should be reverse scored, and you have not reversed them.

The R package psych contains a function alpha() which checks for reversal errors and fixes them.

Jeremy Miles
  • 13,917
  • 6
  • 30
  • 64
-1

My personal observation has been that when someone calculatures Alpha for a mixture of scales like Dichotomous, policy-chotomous, likert etc, then probability of alpha being negative or low is higher. So the conclusion, from my observation, may be personal or biased, is that the use consistent scales be used when calculating Cronbach's Alpha.

The Econometrician
  • 1
  • 3
-3

Brother eliminate correlation between the items by manually editing the results i.e. Q1 4 Q2 5 Q3 3 Q4 4 Make these like Q1 4 Q2 4 Q3 5 Q4 4

Usama
  • 1