In individual/correlation.ipynb (cells 3–5), QQ2 and QQHRV are merged to ~question-level rows for a single subject, pooled across all three sessions, then used to compute Spearman correlations with significance stars (cells 3/4) and an OLS fit with t-statistics and p-values using df = n_obs - n_feat - 1 (cell 5).
All rows come from one participant measured repeatedly, so they aren't independent observations. The reported p-values, FDR "significant" flags, and regression R²/coefficient significance are invalid (pseudoreplication — no session or random-effect term). Effect estimates may be fine as description, but the inferential claims aren't supported.
Fix: treat these as exploratory/descriptive only, or model the repeated-measures structure (mixed-effects with a session/subject random effect, or per-session correlations aggregated properly) instead of pooling question-rows and applying independent-sample tests.
In
individual/correlation.ipynb(cells 3–5), QQ2 and QQHRV are merged to ~question-level rows for a single subject, pooled across all three sessions, then used to compute Spearman correlations with significance stars (cells 3/4) and an OLS fit with t-statistics and p-values usingdf = n_obs - n_feat - 1(cell 5).All rows come from one participant measured repeatedly, so they aren't independent observations. The reported p-values, FDR "significant" flags, and regression R²/coefficient significance are invalid (pseudoreplication — no session or random-effect term). Effect estimates may be fine as description, but the inferential claims aren't supported.
Fix: treat these as exploratory/descriptive only, or model the repeated-measures structure (mixed-effects with a session/subject random effect, or per-session correlations aggregated properly) instead of pooling question-rows and applying independent-sample tests.