quant-scale-reliabilitylisted

# quant-scale-reliability ## 1. Objective Run this skill when the user wants to assess whether a set of survey items reliably measures a single underlying construct before computing a composite score. **Primary outputs:** - Cronbach's α (raw + standardized) with 95% bootstrap CI - McDonald's ω_total (more robust to tau-inequivalence) - Item-total correlations (corrected, with α-if-deleted) - Average inter-item correlation - Subscale α (if user specifies groupings) - Assumption flags (🟢/🟡/🔴) for internal consistency, item discrimination, missing data - Interactive HTML report + 3 PNG plots + `results.json` **When NOT to use:** - If the user wants to *compute* a composite score — that is downstream work - If the user wants dimensionality/factor analysis — use a dedicated EFA skill - For test–retest reliability or inter-rater reliability — different methods needed --- ## 2. Pre-flight checks Before running analysis, verify: | Check | Threshold | |---|---| | Minimum items | ≥ 3 (α is unreliable with fewer) | | Minimum N | ≥ 50 complete cases (warn at 30–49) | | Variable type | All selected columns must be numeric or ordinal integer | | Missing data | Report rate per item; listwise deletion applied | --- ## 3. AskUserQuestion protocol Ask questions **sequentially** — do not batch. ### Q1 — Input file ``` Which data file should I use? ○ output/data/cleaned_data.csv (output from quant-data-prep — Recommended) ○ Specify a different path ``` ### Q2 — Scale item column