---

name: glm-analysis description: "Build Nilearn first-level and second-level GLM workflows, including design matrices, contrasts, BIDS helpers, thresholding, and GLM reports." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

GLM Analysis

Use this sub-skill when a task asks for Nilearn statistical modeling of fMRI activation with first-level or second-level general linear models.

Read First

• For signatures, fitted attributes, methods, and contrast outputs, read API Reference.
• For practical recipes and checklists, read Workflows.
• For common errors and debugging paths, read Troubleshooting.
• For a local no-network design-matrix sanity check, run python scripts/make_design_matrix_smoke.py --help.

Route Here

• Create or validate first-level design matrices from frame times, events, HRFs, drifts, FIR delays, or extra regressors.
• Fit FirstLevelModel, inspect design_matrices_, and compute t/F contrasts from expressions or vectors, including fixed effects over multiple runs.
• Use first_level_from_bids after BIDS derivatives are already present and a task needs model objects plus fit inputs for each subject.
• Fit SecondLevelModel, create second-level design matrices, compute group contrasts, or run non_parametric_inference.
• Apply GLM thresholding with threshold_stats_img or cluster_level_inference, then produce cluster tables or GLM reports.

Route Elsewhere

• Use ../data-io-signal/SKILL.md for image loading, masking, resampling, confound cleaning, signal extraction, and preprocessing before GLM fitting.
• Use ../datasets-interfaces/SKILL.md for dataset fetchers, BIDS queries, fMRIPrep confound strategy selection, and derivative discovery details.
• Use ../plotting-reporting/SKILL.md for figure styling, interactive views, report rendering details, and optional plotting dependency setup.
• Do not use this sub-skill for decoding, searchlight, classification, regression prediction, connectivity, or connectome workflows.

Fast Operating Rules

• Prefer explicit t_r, slice_time_ref, mask_img, and per-run events or design matrices; route upstream data preparation to the sibling skills.
• Inspect model.design_matrices_ before contrasts; expression contrasts are safer than numeric vectors when runs have different column orders or names.
• Set minimize_memory=False only when voxelwise residuals, predictions, R-squared, or MSE are needed after fitting.
• Keep BIDS workflows local: first_level_from_bids reads an existing dataset and derivatives tree; it does not download datasets.
• Treat thresholding choices as statistical decisions: document whether the workflow uses FPR, FDR, Bonferroni, explicit thresholds, cluster extent, permutation inference, TFCE, one-sided, or two-sided tests.