If you have built or used a questionnaire, a supervisor or reviewer has almost certainly asked: "What is the Cronbach's alpha?" Alpha is the most widely reported reliability coefficient in the social and health sciences. It is also widely misunderstood.
This guide explains what alpha measures, what counts as a good value, how to improve a low alpha without cheating, when to report McDonald's omega instead, and exactly how to write it up.
In this guide
What Cronbach's alpha measures
Cronbach's alpha (α) estimates the internal consistency of a set of items that are meant to measure the same construct: how strongly the items "hang together". If people who score high on one item of a job-satisfaction scale also tend to score high on the others, alpha will be high.
Alpha depends on two things:
- The average correlation between items. Stronger correlations give a higher alpha.
- The number of items. More items give a higher alpha, even if the average correlation stays the same.
The standardized formula makes this explicit:
α = (k × r̄) ÷ (1 + (k − 1) × r̄)
where k is the number of items and r̄ is the mean inter-item correlation. With r̄ = .30, three items give α = .56, while ten items give α = .81.
A high alpha does not show that a scale is unidimensional (measures only one thing), and it says nothing about validity. Ten items about two different topics can still produce α = .80. Check dimensionality with exploratory factor analysis or CFA.
What is a good Cronbach's alpha?
The most quoted guideline comes from George and Mallery (2003), building on Nunnally (1978):
| Cronbach's alpha | Common interpretation |
|---|---|
| α ≥ .90 | Excellent (but check for redundant items) |
| .80 ≤ α < .90 | Good |
| .70 ≤ α < .80 | Acceptable |
| .60 ≤ α < .70 | Questionable |
| .50 ≤ α < .60 | Poor |
| α < .50 | Unacceptable |
In practice:
- .70 is the usual minimum for research scales.
- .60–.70 may be tolerated for new scales, exploratory research or very short scales (2–3 items).
- .80–.90 or higher is expected when a score is used to make decisions about individuals, for example clinical screening.
- Above .95 is not always good news. It often means some items are near-duplicates ("I like my job" / "I enjoy my job"), which wastes respondents' time.
How to calculate Cronbach's alpha
- Make sure every item is scored in the same direction. Reverse-score negatively worded items first (for a 1–5 scale, the new score is 6 minus the old score). The free data cleaning tool can do this in one step.
- Paste the item columns into the Cronbach's alpha calculator, one column per item and one row per respondent.
- Run it. You get alpha with a 95% confidence interval, standardized alpha, the mean inter-item correlation, corrected item-total correlations and alpha if item deleted.
Calculate alpha separately for each scale or subscale, never for the whole questionnaire at once.
Reading the item statistics
Corrected item-total correlation is the correlation between each item and the sum of the other items. Values below about .30 suggest the item does not fit well with the rest.
Alpha if item deleted shows what alpha would be without that item. If removing an item raises alpha noticeably, look closely at it.
Example output for a five-item engagement scale (α = .74):
| Item | Corrected item-total r | Alpha if deleted |
|---|---|---|
| ENG1 | .58 | .67 |
| ENG2 | .61 | .66 |
| ENG3 | .55 | .68 |
| ENG4 | .52 | .69 |
| ENG5 | .14 | .81 |
ENG5 stands out. Before deleting it, check whether it should have been reverse-scored, whether its wording is confusing, and whether it measures a different aspect of the construct. If you remove it, report that you did so and why.
What to do if your alpha is low
- Check reverse-scored items. Forgetting to reverse an item is the single most common cause of a low or even negative alpha.
- Look for data-entry errors and impossible values (a 9 on a 1–5 scale).
- Inspect item-total correlations and consider removing items that clearly do not belong, if theory supports it.
- Check dimensionality. If the scale covers two sub-dimensions, compute alpha for each.
- Consider the number of items. Two- or three-item scales rarely reach .70. For two-item scales, report the Spearman-Brown coefficient instead of alpha.
Do not delete items purely to push alpha over .70 when it costs content validity. Reviewers increasingly notice "alpha hacking".
Alpha vs McDonald's omega vs composite reliability
Alpha assumes every item is an equally good indicator of the construct (the "tau-equivalence" assumption). When loadings differ, which is the usual case, alpha tends to underestimate reliability. That is why many methodologists now recommend reporting McDonald's omega (ω) alongside or instead of alpha.
| Coefficient | Based on | Use when |
|---|---|---|
| Cronbach's alpha | Inter-item covariances | Quick check; expected by many journals |
| McDonald's omega | Factor loadings (one-factor model) | Items have unequal loadings; modern best practice |
| Composite reliability (CR) | Standardized loadings from CFA or SEM | Reporting a measurement model in CB-SEM or PLS-SEM |
You can calculate all three for free: McDonald's omega, composite reliability and AVE, or everything at once in the reliability and validity analysis tool.
How to report Cronbach's alpha in APA 7
Report alpha for each scale in the Method section (Measures), usually to two decimal places. Because alpha cannot exceed 1, omit the leading zero.
Work engagement was measured with five items rated on a 7-point scale (1 = strongly disagree, 7 = strongly agree). Internal consistency was acceptable (Cronbach's α = .74, 95% CI [.67, .80]). One item (ENG5) had a low corrected item-total correlation (r = .14); removing it increased reliability to α = .81, so the four-item version was used in all analyses.
In a table of several scales, add a column for α (and ω if you report it) next to the mean and SD of each scale.
Frequently asked questions
Is a Cronbach's alpha of 0.6 acceptable?
It is usually considered questionable. It can be tolerated for new or very short scales, or in exploratory research, but you should acknowledge it as a limitation. Most published research expects at least .70.
Can Cronbach's alpha be negative?
Yes. A negative alpha almost always means some items are negatively correlated with the others, usually because reverse-worded items were not reverse-scored.
Is a very high alpha (above .95) good?
Not necessarily. Very high values often mean that items are redundant, asking the same thing in slightly different words. Consider shortening the scale.
Should I report alpha or omega?
Reporting both is a safe choice. Omega is more accurate when items have unequal loadings, which is common, but many journals still expect alpha.
How many items do I need for a reliable scale?
There is no fixed number, but scales with fewer than three items rarely reach α = .70. Three to six good items per construct are common in survey research.