Cronbach's α measures the internal-consistency reliability of a multi-item scale, reporting α-if-item-deleted so weak items stand out.
α answers 'do these items measure the same underlying construct consistently?' by comparing the variance of the total scale score to the sum of item variances. Mathematically α = (k/(k−1)) · (1 − Σσᵢ² / σ_total²); it equals the average split-half reliability across all possible splits.
Conventional benchmarks: α ≥ 0.9 excellent (but maybe redundant items), 0.8–0.9 good, 0.7–0.8 acceptable for research scales, 0.6–0.7 questionable, < 0.6 poor. Clinical decision tools usually demand α ≥ 0.9.
α is *not* a measure of unidimensionality — it assumes it. A scale measuring two distinct factors can still hit α = 0.8. Check unidimensionality separately (factor analysis or McDonald's ω) before relying on α as evidence of construct coherence.
≥ 2 numeric item columns (one row per respondent).
α + mean inter-item correlation + item variances + per-item α-if-deleted.
A single item where α-if-deleted exceeds the overall α flags an item that's hurting the scale — usually because it's measuring something different or was reverse-coded but not recoded.