Cronbach's Alpha Calculator

What Cronbach's alpha actually measures

Alpha summarizes how consistently a set of items rank people the same way. It rises when items correlate with each other and when there are more of them. That second part matters more than most people realize: a 40-item scale of mediocre items will out-score a 5-item scale of excellent ones, because alpha is a function of test length as much as item quality.

What alpha is not is a measure of whether your scale captures one thing. That is a common misreading. A scale with two distinct clusters of items can still produce a high alpha, and reporting it will not reveal the split. If you need evidence that your items measure a single construct, you need factor analysis, not alpha.

What counts as acceptable

The usual benchmarks are .70 for exploratory work, .80 for established scales, and .90 when scores drive decisions about individuals. Treat them as conventions rather than thresholds. Alpha above .95 is often a warning rather than a triumph, because it usually means several items are near-duplicates of each other and the scale is longer than it needs to be.

Sample size matters too. Alpha estimated on 20 respondents is unstable, which is why this calculator reports a confidence interval alongside the point estimate. A scale reporting alpha of .82 with an interval running from .61 to .93 has not really demonstrated .82.

Reading the item-total table

The corrected item-total correlation is the correlation between one item and the sum of the remaining items. Below about .30, the item is not measuring what the rest of the scale measures. Check first whether it should have been reverse-scored, since a negative value almost always means a reversal was missed.

Alpha if item deleted shows what alpha would be with that item removed. When a value here exceeds the overall alpha, the item is dragging the scale down. Do not delete it on that basis alone. An item can be statistically weak and still be the only one covering part of your construct, and dropping items to chase a number is how scales quietly lose their content coverage.

Reverse-scored items

If your scale mixes positively and negatively worded items, the negative ones must be reversed before alpha means anything. Enter their column numbers above along with the scale maximum, and this calculator will reverse them for you using the standard formula: new score equals minimum plus maximum minus old score.

Forgetting this is the single most common cause of a surprisingly low alpha. If your result comes back near zero or negative, check reversals before you conclude anything about the scale.

The assumptions alpha depends on

Alpha is a lower bound on reliability, and it equals reliability only when all items are essentially tau-equivalent, meaning every item relates to the underlying construct with equal strength. Real scales rarely satisfy that. When they do not, alpha understates reliability, sometimes noticeably.

McDonald's omega drops the equal-loadings requirement and is the better estimate for most scales, which is why methodologists have been recommending it for years. Alpha survives mostly because reviewers still expect it. Report both when you can. Omega, along with the factor model it depends on, is available in Quanta.

How to report alpha in APA style

Report alpha to two decimals with no leading zero, name the scale and the number of items, and give the sample it was computed on. Alpha is a property of a set of scores, not a permanent property of the instrument, so a value borrowed from a published paper does not describe your data. Compute it on your own sample every time.

If your alpha is low and you are not sure whether the problem is the items, the sample, or the construct, a fixed-price method review will tell you which before you rewrite the scale.