Statistics¶
This section explains the statistical methods PyFLASH uses in plots, pipelines, model outputs, and structured report records. The pages focus on what PyFLASH computes, where the results appear, and what can and cannot be concluded from the output.
| Page | Purpose |
|---|---|
| Group Comparisons | Tests for comparing numeric measurements across groups. |
| Effect Sizes | Magnitude estimates, confidence intervals, and interpretation labels. |
| Multiple Testing | Bonferroni, Holm, Sidak, false discovery rate, and PyFLASH gate behavior. |
| Correlation | Pearson, Spearman, Kendall, matrix outputs, and correlation gates. |
| Linear Models | Ordinary least squares models, covariates, coefficients, and adjusted means. |
| Classification | Cross-validation metrics and model-sweep interpretation. |
| Rhythm | Cosinor, circular phase summaries, and rhythm group tests. |
| Power | Achieved power, required sample size, and minimum detectable effects. |
| Structured Results | Report records, manifests, and describe coverage. |
General Reading Order¶
Start with the function or pipeline page for the command you ran, then use these statistics pages to interpret the columns and decisions in the output tables. For example, group_comparison explains how to run the pipeline, while Group Comparisons explains the tests and result columns it uses.
Shared Conventions¶
PyFLASH reports p-values as evidence against a null model, not as effect size or biological importance. Corrected p-values or q-values appear only where the function enables a correction step. Effect sizes and confidence intervals should be read alongside p-values, group sizes, missing-data handling, and the design of the experiment.
Most pipeline outputs are written as CSV files plus a manifest.json file. Some
functions also emit structured report records when the report collector is
active; see Structured Results.