Scale reliability checker
Check the internal consistency of a questionnaire scale with Cronbach's alpha, its confidence interval, alpha if each item is deleted and corrected item-total correlations. Reverse-worded items are recoded before analysis.
Runs locally in your browser; nothing is uploaded. Paste up to 1,000,000 characters of CSV with at most 100 columns in total (including any ID column), 2 to 100 item columns, 3 to 10,000 respondents and at most 50,000 item scores in total. Blank lines at the end are ignored. Scores are plain numbers inside the declared minimum and maximum. Downloads: a JSON report plus CSV item statistics and an inter-item correlation matrix. One header row of item names, then one row per respondent with numeric scores. Up to 1,000,000 characters. Review the settings before running.
Your result
How to use Scale reliability checker
- Paste CSV with one column per item and one row per respondent.
- List any reverse-worded items (none are reversed by default) and give the scale minimum and maximum.
- Choose how blank cells are handled and the confidence level.
- Calculate, review weak items, and download the report and CSV tables.
Useful next steps
What alpha measures
Cronbach's alpha (Cronbach 1951, Psychometrika 16:297) compares the sum of item variances with the variance of the total score. The interval uses Feldt's F-distribution method (Feldt 1965, Psychometrika 30:357). Standardized alpha uses the mean inter-item correlation. Alpha if item deleted and the corrected item-total correlation (each item against the total of the other items) show which items fit the scale least.
Reverse coding
A reverse-worded item is recoded as minimum plus maximum minus the score before anything is computed. Forgetting this step can make alpha small or negative, so check the item-total correlations for negative values.
Assumptions and limits
Alpha assumes the items measure one construct with similar loadings. It is not evidence of validity, and a high value can come from many similar items. Blank cells either stop the analysis or drop the whole respondent.