Cronbach's alpha
Cronbach's alpha summarizes how items vary together in a summed score; its interval shows sampling uncertainty, not proof of validity.
Fit
When to use it, and when not to
Use it when
- You want internal consistency for a fixed set of items meant to measure one construct.
- You have item-level responses from the population where the score will be used.
Not when
- You need stability over time or agreement between raters. Use a reliability intraclass correlation with a suitable measurement design. Read: Intraclass correlation for reliability
- You need evidence about dimensions. Use factor analysis and separate evidence about validity. Read: Factor analysis
Before you start
What you need
- The fixed item set and scoring ruleState reverse coding, item weights and whether alpha uses raw covariances or standardized correlations.
- Item responses and intended populationKeep each person’s responses together. State handling of missing items and the interval method.
- Items are intended to reflect one construct.How to check: Check content and dimensional structure separately. Alpha alone cannot establish either.
- Reliability interpretation needs essentially tau-equivalent items and uncorrelated errors.How to check: Items should share the same true-score component, up to constants. Unequal loadings or shared wording can break these conditions.
- People are independent draws from the target population.How to check: Resample whole people for a bootstrap interval. Use a design-aware resampling scheme for clustered samples.
- Missing item responses have a stated assumption.How to check: Complete-case analysis needs justification. Under missing at random (MAR), consider suitable imputation; assess sensitivity when missing not at random (MNAR) is plausible.
Plain language
How it works
Raw alpha compares the sum of item variances with the variance of their total score. For k items, calculate k/(k−1) times one minus that variance ratio. Standardized alpha instead uses correlations and answers a different scoring question. Reliability interpretation needs the stated measurement assumptions.
A percentile bootstrap resamples whole people, recomputes the same alpha, then takes the chosen quantiles. It reflects sampling uncertainty for fixed items. Small samples can give unstable coverage. Sijtsma explains why alpha cannot establish dimensionality or validity.
- Person 1: (0, 0, 1, 0).
- Person 2: (0, 0, 1, 0).
- Person 3: (2, 1, 1, 3).
- Person 4: (4, 2, 3, 2).
- Person 5: (4, 3, 3, 4).
- Person 6: (0, 1, 0, 0).
- Person 7: (0, 1, 1, 2).
- Person 8: (2, 1, 2, 2).
- Person 9: (2, 3, 3, 2).
- Person 10: (3, 3, 4, 4).
- Person 11: (1, 0, 1, 1).
- Person 12: (1, 1, 1, 1).
- Person 13: (2, 1, 2, 3).
- Person 14: (3, 4, 3, 2).
- Person 15: (3, 4, 4, 4).
- Person 16: (0, 1, 1, 0).
- Person 17: (0, 1, 1, 1).
- Person 18: (3, 3, 1, 1).
- Person 19: (3, 4, 2, 2).
- Person 20: (4, 4, 4, 4).
Worked example
A worked example
- Setting
- An invented pilot has 20 independent people and four equally weighted symptom items, scored from 0 to 4.
- Question
- How consistent is their summed score within this sample?
- Approach
- Compute raw-score alpha from the item covariance matrix. Resample whole response rows 10,000 times and use the 2.5th and 97.5th percentiles.
- Result
- Raw-score alpha is 0.925. Its 95% percentile bootstrap confidence interval is 0.854 to 0.962.
- What it means
- The interval concerns alpha for this fixed item set. This small pilot cannot establish one dimension, validity or stability over time.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A high alpha does not prove one dimension, validity, responsiveness or stability over time. Redundant items can raise alpha. Interpreting it as score reliability requires essentially tau-equivalent items and uncorrelated errors; report an interval for the alpha estimator used.
- Using a high alpha as proof that a scale is valid.Instead: Check content, structure and relationships with other measures for the intended use.
- Deleting items solely to raise alpha.Instead: Protect content coverage. More or redundant items can raise alpha without improving measurement.
- Treating alpha as a proportion of correct responses.Instead: Report the coefficient and its matching interval. Alpha can be negative and is not a binomial proportion.
Inside Mallard
How Mallard uses this
Mallard includes this method in Learn search and the Atlas. It has no automatic plan cue, dedicated calculator or execution route for this method.
analysis.cronbach_alpha, the id a plan cue records.How to report it
Report the item set, population, scoring and missing-item rules, sample size, raw or standardized alpha, and the interval method and level. Keep dimensionality and validity evidence separate. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.
Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Coefficient alpha and the internal structure of tests (opens in a new tab) · Psychometrika 16:297–334
- On the Use, the Misuse, and the Very Limited Usefulness of Cronbach's Alpha. (opens in a new tab) · Psychometrika · doi 10.1007/s11336-008-9101-0
- Bootstrap Methods for Standard Errors, Confidence Intervals, and Other Measures of Statistical Accuracy (opens in a new tab) · Statistical Science 1:54–75