A Cronbach's alpha of .70 or higher is the most widely accepted threshold for adequate internal consistency, a figure that goes back to Nunnally (1978), who recommended .70 for early-stage research. Values between .80 and .90 are generally regarded as good, and values above .95 can signal redundant items rather than an excellent scale (Streiner, 2003). Values between .60 and .70 are sometimes accepted in exploratory work (Hair et al., 2019), but you need to say so and justify it.
Where the .70 cut-off comes from
Cronbach (1951) introduced coefficient alpha as an estimate of how consistently a set of items measures the same thing. He did not set a cut-off. The .70 figure that appears in almost every thesis is taken from the second edition of Nunnally's Psychometric Theory (1978), where .70 is described as sufficient for the early stages of research, while higher values (.80 and above, and as high as .90 to .95) are needed when scores are used to make decisions about individuals. In other words, .70 was never meant as a universal pass mark. It is a minimum for research that compares groups or tests relationships.
Taber (2018) reviewed how alpha is used in science education journals and found that authors describe the same values with very different adjectives ("acceptable", "satisfactory", "high"), often without a source. The safest approach in a thesis is to state one threshold, cite where it comes from, and apply it consistently to every scale.
A practical interpretation table
| Alpha | Usual reading | Source to cite |
|---|---|---|
| ≥ .95 | Possibly too high: items may be near-duplicates | Streiner (2003) |
| .80 – .95 | Good reliability for research purposes | Nunnally (1978) |
| .70 – .80 | Acceptable | Nunnally (1978); Hair et al. (2019) |
| .60 – .70 | Acceptable only in exploratory research | Hair et al. (2019) |
| < .60 | Poor: investigate the items | Hair et al. (2019) |
You will also see the rules of thumb from George and Mallery (2003): above .9 excellent, above .8 good, above .7 acceptable, above .6 questionable, above .5 poor and below .5 unacceptable. These labels are popular in student work, but they are a textbook convenience rather than a derivation, so pair them with Nunnally if your examiner is strict about sources.
Worked example
Suppose a four-item scale for Perceived Usefulness, answered on a 5-point Likert scale by 200 respondents. The item variances are 1.10, 0.95, 1.20 and 1.05, so their sum is 4.30. The variance of the total score (the sum of the four items) is 12.60.
Alpha = k / (k − 1) × (1 − sum of item variances / total variance) = 4 / 3 × (1 − 4.30 / 12.60) = 1.333 × .659 = .88.
An alpha of .88 is above the .70 threshold, so you would write: "Cronbach's alpha for Perceived Usefulness was .88, above the .70 threshold recommended by Nunnally (1978)." Next, check the item-level output. If any Corrected Item-Total Correlation falls below .30 (Field, 2018), or the "Cronbach's alpha if item deleted" column shows a clear rise when one item is dropped, look at that item's wording before deciding anything. Remember that alpha rises with the number of items, so a ten-item scale can reach .80 even when the items are only loosely related (Streiner, 2003).
What to do when alpha is below .70
- Check reverse-coded items first. A negatively worded item that was not recoded pulls alpha down sharply and often shows a negative item-total correlation.
- Look at item-total correlations. Items below .30 are candidates for removal, but only if removal does not change what the construct means.
- Consider the number of items. Two- and three-item scales often have modest alpha even when they are fine; report composite reliability as well.
- Report honestly. If you keep a scale at .65, cite the exploratory threshold and list it as a limitation.
If you are running PLS-SEM, alpha is treated as a lower bound and is read alongside composite reliability (rho_c) and rho_A, with .70 to .95 as the desired range for both (Hair et al., 2019; Dijkstra & Henseler, 2015).
Run it free
Paste your item columns into the free Cronbach's alpha calculator. It reports alpha, Corrected Item-Total Correlation and alpha if item deleted, flags each value against its sourced threshold and writes an APA results sentence. Results are checked against R.
FAQ
Is a Cronbach's alpha of .60 acceptable?
Only for exploratory research, and you should cite the exception (Hair et al., 2019). For established scales, .70 is expected.
Can Cronbach's alpha be too high?
Yes. Values above about .95 often mean items repeat the same question in slightly different words (Streiner, 2003).
Should I delete items to increase alpha?
Only if the item is weak on several indicators (low item-total correlation, poor wording) and the construct still covers its full meaning without it. Report every deletion.
Does alpha prove my scale is valid?
No. Alpha measures internal consistency, not whether the scale measures the right thing. You still need factor analysis and validity evidence.
References
- Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests. Psychometrika, 16(3), 297–334. https://doi.org/10.1007/BF02310555
- Dijkstra, T. K., & Henseler, J. (2015). Consistent partial least squares path modeling. MIS Quarterly, 39(2), 297–316.
- Field, A. (2018). Discovering statistics using IBM SPSS Statistics (5th ed.). Sage.
- George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update (4th ed.). Allyn & Bacon.
- Hair, J. F., Risher, J. J., Sarstedt, M., & Ringle, C. M. (2019). When to use and how to report the results of PLS-SEM. European Business Review, 31(1), 2–24. https://doi.org/10.1108/EBR-11-2018-0203
- Nunnally, J. C. (1978). Psychometric theory (2nd ed.). McGraw-Hill.
- Streiner, D. L. (2003). Starting at the beginning: An introduction to coefficient alpha and internal consistency. Journal of Personality Assessment, 80(1), 99–103. https://doi.org/10.1207/S15327752JPA8001_18
- Taber, K. S. (2018). The use of Cronbach's alpha when developing and reporting research instruments in science education. Research in Science Education, 48(6), 1273–1296. https://doi.org/10.1007/s11165-016-9602-2