How high the Cronbach's alpha should be to be recognized as satisfactory? Majority of sources say 0.7. Does the purpose of a questionnaire make the difference: research or diagnosis? Any limitations for such categorisation?
Thank you!
When the instrument is tested in a new environment a cronbach alpha coefficient of 0.5 is acceptable
Thank you, Joseph!
Could you please provide any reference for that 0.5 value?
That 0.7 is also a rule of thumb. Some studies recommend 0.8-0.9
The Cronbach alpha could fall below 0.7 due to presence of reverse coding or multiple factors. I would think that reverse coding should be avoided as much as possible not to confuse the scores on a Likert scale. Regarding multiple factors, one may have several questions but they are testing different factors. Mixing all the questions into the Cronbach alpha computation may lead to a low score if there are multiple and distinct factors.
0.70 or higher is the recommendable criterion. Two consensual authors are: DeVellis (1991) and Nunnally & Bernstein (1994).
DeVellis, R.F. (1991). Scale development: Theory and applications. London: SAGE.
Nunnally, J.C., & Bernstein, I.H. (1994). Psychometric theory (3rd ed.). New York: McGraw-Hill.
You can consider coefficients little below 0.70 acceptable (not below 0.65) if the scale shows robust validity evidences (factor validity, criterion validity) or if you can prove the reliability through other method, which is the test re-test reliability (temporal stability of the results).
In fact, Cronbach's alpha is not a very good measure.
First, it presupposes that loadings of all items on the scale are equal (tau-equivalence, or Raschmodel) and item residuals are not correlated, conditions which are only rarely satisfied. Secondly, even if these condition are satisfied, alpha underestimates relability. If the conditions are not satisfied, it is undetermined if alpha over- or underestimates reliability, and more so if loadings are rather low (which is common in the early phases of the development of an measurement instrument).
Instead, use confirmatory factor analysis to examine if the scale is satisfactory and which items perhaps are problematic,
Literature:
Papers by Tenko Raykov
Raykov & Marcoulides (2011) Introduction to Psychometric Theory. New York :Routledge.
Hello Michal,
I hope the following information that I wrote in a journal article about 7 years ago is helpful:
Theoretically, alpha values can range from zero to one, but typically they range from around .65 to .90. From a naïve perspective, low values of alpha indicate heterogeneity among items and high values indicate homogeneity. Lower alpha values are generally regarded as less acceptable than higher values. However, whether an alpha value is satisfactory or not should be made with reference to the number of items involved. Because of the way they are derived mathematically, high alpha values are more easily obtained with a larger number of items,33 even when some of those items are not strongly related with others. The opposite also pertains: Low or moderate alpha levels often occur with small item numbers, even when those items are quite highly associated with each other. This has several important implications that are often overlooked. When large numbers of items are involved, low alpha values are indeed undesirable because they indicate too little inter-item association, but a high alpha value may spuriously imply inter-item homogeneity because it could simply be a mathematical inevitability. When small item numbers are involved, extremely low alpha levels remain undesirable, but high alpha values are problematic because they indicate excessive inter-item association. If this latter situation pertains, most or all of the items are likely to be repeatedly tapping the same thing, at least from the respondents’ perspective, and they are therefore probably not capturing the breadth or complexity of a construct.12, 31 Under those circumstances a number of items could be discarded without reducing an instrument’s effectiveness.
It has been recommended that “as with food and other good things, moderation in internal consistency is best”.31, p.40 If researchers aim for a moderate amount of association among a set of items, alphas of around .65 are probably satisfactory for four or five items, whereas alphas of around .93 are equally satisfactory for about 30 items.33 In line with this, alphas of around .75 and .80 would be appropriate for 10 and 15 items respectively.
This text is from an article published in the Journal of the American Podiatric Medical Association, under my name, titled Self-Assessment of Foot Health: Requirements, Issues, Practicalities, and Challenges. http://dx.doi.org/10.7547/0990460
That article might have other information you'd find useful.
Robt.
Cronbach's alpha should exceed 0.70. Otherwise, do exploratory factor analysis with varimax rotation. Delete the item(s) that can make other items to have higher Cronbach's alpha. Good luck.
Let me add my one cent here. Most of the rules of thumb stem from misreading or unreading the original writing of Cronbach. He never said .7 is 'good enough' or 'adequate.' What he meant was that .7 is bare minimum and most researchers would want much higher alpha. Because he mentioned .7 first, somehow it became the rule... Cortina's (1993) Journal of Applied Psychology article will inform you well about influencing factors on alpha, especially the role of the number of items Robert mentioned.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics
- Survey Design
- Questionnaire