wwu-mmll / confound_corrected_cpm / 28516526735

85%

main: 85%

push

github

Add confound-inflation simulation and demonstration

Extend the SEM simulation generator to support three interpretable edge
classes (pure-signal, mixed, confound-only) built from two orthogonal
latents, with exact closed-form R2 targets controlled by a spurious-fraction
knob (kappa). Add simulate_data_given_kappa and generate_confound_grid.

Add examples/confound_inflation_demo.py, which sweeps a 3x6 grid of
brain-outcome strength x confound strength and shows that raw CPM is
maximally inflated, partial-correlation edge selection only partially
deconfounds (it drops confound-only edges but keeps confound-carrying mixed
edges), and residualizing the connectome recovers the true association.

Note in mediator_simulation that its c is an outcome proxy, not a
common-cause confound.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

54 of 76 new or added lines in 1 file covered. (71.05%)

2128 of 2973 relevant lines covered (71.58%)

0.72 hits per line