Update
I have a scale (12 items)
I go to Analysis -> Scale -> Reliability analysis and get my Cronbach alpha (0,5)
BUT 2 of my items are «inverse». If I recode this two items as it was not inverse I get alpha=0.8
Am I right? I should recode this items before counting Cronbach alpha?
written earlier
I conducted a study (correlation plan).
I used (including) 2 psychological tests, which were adapted by another author according to all the rules.
And I run into problems:
Situation1 (solved)
My first test (14 items) has 2 subscales. In Ukrainian adaptation, the Cronbach alpha for the scales is 0.73 and 0.68. But I did my own research and counted Cronbach's alpha. 0.65 and 0.65 came out.
Question1: Should I count correlations with this test or, maybe, exclude this test from analysis?
Situation 2 (see update)
My second test is Zimbardo’s Time Perspective Inventory (56 items). In Ukrainian adaptation, four of five scales have Cronbach Alpha above 0.7. One scale is 0.65.
But in my research everything is ok only with 3 scales, they are higher than 0.7.
Two scales have a very low Cronbach Alpha: 0.55 and 0.49.
Question2: should I exclude this two low scales and count correlations with only that 3 scales which Cronbach Alpha more than 0,7?
PS: N=336 in my study
Cronbach alpha doesnt mean a scale has sufficient psychometric properties or not. It gives only internal consistency. If you want an answer about psychometric properties or quality, at least one validity evidence require. Factor analysis, inter-item correlations or maybe item response theory applications.
Only Cronbach alpha values means nearly nothing.
Before becoming too desperate, I would look carefully at the interitem correlations as well as corrected item-total correlations, to see whether there are any items that, in your context, might be particularly weak.
Perhaps any weak items could be disregarded, with an increase in your coefficient alpha.
That's a start, anyway.
Hakan Koğar
Thank you for a quick response. Maybe you got me wrong, I don’t want to create a questionnaire. I just use them in my study. And as I see in many articles authors report “Estimates of internal consistency reliability for the current study ranged from a=.71 …. to a=.85” I want to write only one sentence like this.
Robert Trevethan
I looked carefully and now I think I did a mistake counting Cronbach alpha. because I did not pay attention to the reverse questions. If you want to help, please see updates in the beginning of the question.
Hello Olkesandra: Ah that's an old oversight - failing to attend to items that need to be reverse coded. You must be feeling relieved now that you have a better alpha.
However, please be aware that coefficient alpha (that's apparently a more appropriate name than is Cronbach's alpha) does, indeed not guarantee you a lot about the validity of a scale, as Hakan K. pointed out above.
Many researchers cite alphas because everyone else does so, and probably because they're easy to obtain. But whether they're meaningful or not is another matter. By chance, some colleagues and I dealt with that in an article that I put onto ResearchGate earlier today. If you're interested, I'd be happy to give you the reference - or you could check it out among the publications I've uploaded. It's a long article about measurement of teacher sense of efficacy - and in the middle of that article we deal with some ins and outs of alpha.
I'm glad you have a good number of participants.
No matter create or use Oleksandra Shatilova any measurement tool must be valid. And cronbach alpha's say nothing direct things about validity. When you handle with reverse questions, Cronbach's alpha must be rise. But it is not sufficient.
So i agree with Robert Trevethan 's advices
Hakan Koğar Thank you very much! I will check the other options you wrote.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS