# Where do the descriptors for Cronbach's alpha values come from (e.g., poor, excellent)?

It seems fairly common to describe Cronbach's alpha values as follows:

• α ≥ 0.9 Excellent
• 0.7 ≤ α < 0.9 Good
• 0.6 ≤ α < 0.7 Acceptable
• 0.5 ≤ α < 0.6 Poor
• α < 0.5 Unacceptable

Where do these values come from? I cannot find an original research article describing these.

Edit: I am 90% sure its merely based on convention and there is no classic research article outlining these.

• Nunnally has largely popularized those thresholds (emphasizing the distinction between individual vs. group-level decision, though); but there are so many flaws with its use in current research that it's probably not worth to worry about them :-) – chl Sep 17 '13 at 20:53

The following two papers discuss cut-off values for reliability indices:

• Lance, C.E., Butts, M.M., & Michels, L.C. (2006). The sources of four commonly reported cutoff criteria: What did they really say? Organizational Research Methods, 9 (2), 202-220.
• Henson, R.K. (2001). Understanding internal consistency reliability estimates: A conceptual primer on coefficient alpha. Measurement and Evaluation in Counseling and Development, 34 (3), 177-189.

Strictly speaking neither of them supports the specific scale you describe – the first one in particular is rather critical of the whole idea of a conventional cut-off values – but they do point to many key publications on this topic so digging up those references might bring you to the original sources.

Kline (in the 1993 edition of the Handbook cited by Gavin in his answer) traces his cut-off value to Guilford and Nunnally. IIRC, Nunnally never provided much justification for his recommendation and actually changed it from one edition to the next of his Psychometric Theory but his writings have been very influential so he might very well be most responsible of the popularity of the notion that .7 is acceptable and .9 excellent.

Incidentally, Cronbach's $\alpha$ is often misinterpreted and has been thoroughly criticized. Even the very idea of aiming for higher internal consistency has been called into question (most notably by Cattell, cf. “bloated specifics”). All that to say that looking for the original source of this or that convention might be of some historical interest but none of this is terribly useful to inform psychological measurement.

• Thank you. Indeed, I am interested for historical interest. I was asked to cite the use of the word "excellent", and I didn't feel like citing an SPSS manual. I looked further, without much luck. – Behacad Sep 17 '13 at 18:57
• I would look at a later edition of Nunnally's book I think. – Gala Sep 17 '13 at 19:06
• (+1) I have other references in these personal notes. – chl Sep 17 '13 at 20:55
• Nunnally and Bernstein provide quite a long discussion of when different values of alpha might be acceptable. Generations of psychologists summarize this as "above 0.7 is OK." – Jeremy Miles Sep 18 '13 at 1:04

Wikipedia cites the sources as

• George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update (4th ed.). Boston: Allyn & Bacon.
• Kline, P. (1999). The handbook of psychological testing (2nd ed.). London: Routledge

I would follow up those references to see if they cite additional, primary sources. However, as a rule of thumb, these value descriptions may not have a primary source.

• Yes, I saw these on wikipedia. I am looking for any original research. I believe the answer is merely "convention", but I was hoping for something a bit more concrete. – Behacad Sep 17 '13 at 17:03
• Also, it would seem to me that the convention started somewhere... – Behacad Sep 17 '13 at 17:42
• @Behacad right, but that somewhere need not be a primary research article. It may have been in software, or decided upon in a working group or workshop for example, for a particular project and just stuck. – Gavin Simpson Sep 17 '13 at 17:45
• Yes, and I am wondering what that particular project was, if there was one. At one point, someone said .90 is excellent (or similar), and it stuck. Who was it? – Behacad Sep 17 '13 at 17:56

Lance, C. E., Butts, M. M., & Michels, L. C. (2006). The sources of four commonly reported cutoff criteria what did they really say?. Organizational research methods, 9(2), 202-220.

"Comparing this section to citations to it, we note several things. First, we suspect that most authors who cite Nunnally’s .70 reliability criterion would not agree that they are trying to save time and energy in an early stage of research by using measures that have only modest reliabilities. Rather, we suspect that most researchers would claim to be conducting basic (or perhaps applied) research, for which purpose Nunnally clearly recommended a reliability standard of .80. Carmines and Zeller (1979) made a similar recommendation: “As a general rule, we believe that reliabilities should not be below.80 for widely used scales” (p. 51). Thus, our second point is that .80, and not .70 as has been attributed, appears to be Nunnally’s recommended reliability standard for the majority of purposes cited in organizational research."