Skip to content

Statistics Options

Summary

Statistics options control group-comparison tests, annotations, p-value corrections, and significance gates. These options do not define the groups; they act on the groups created by conditions, factors, split_by, and row filters.

Used By

Accepted Values

Parameter Accepted values Behavior
comparisons One-based strings such as "1-2"; some matrix-difference functions also accept explicit pairs. Selects group contrasts in the current group order.
force_nonparametric True or False. Forces the shared multi-group path to use non-parametric testing. In the current two-group bar-statistics path, very small groups fall back to Mann-Whitney U, otherwise the independent t-test path is used.
multiple_comparison Usually "One-Way" or "Two-Way". In the shared bar statistics path, "One-Way" selects one-way ANOVA when groups are normal; other values use the two-way ANOVA path.
posthoc "Conover" or "Dunn"; common text variants such as "Dunn's test" normalize to "Dunn". Selects the Kruskal-Wallis post-hoc test.
posthoc_correction "auto", "Bonferroni", "Uncorrected", booleans, or common yes/no synonyms. "auto" applies Bonferroni only when there are more than three comparisons.
ns Usually "ns" or "p". Text for non-significant annotations. "p" prints a rounded p-value instead.
alpha Float, commonly 0.05. Significance threshold for annotations, FDR rejection, gates, and matrices.
gate Usually "p" or "fdr" in screening pipelines. Correlation-style gates also accept FDR aliases such as "q", "q_value", "q-value", "fdr_bh", and "bh". Chooses whether raw p-values or adjusted q-values define significance.
fdr_method / correction method Any method accepted by statsmodels.stats.multitest.multipletests, such as "fdr_bh", "fdr_by", "holm", "bonferroni", or "sidak". Used by FDR and multiple-testing helpers.
test / tests / correlation "pearsonr", "spearmanr", "kendalltau" and aliases "pearson"/"p", "spearman"/"s", "kendall"/"k". Correlation method selection.

Examples

Force non-parametric mean-bar tests:

from PyFLASH.plotting import plot_mean_bars

plot_mean_bars(
    batch,
    data_cols=["GFAP_Count"],
    force_nonparametric=True,
    posthoc="Dunn",
    posthoc_correction="Bonferroni",
    comparisons=["1-2", "1-3", "2-3"],
)

Use an FDR gate in a correlation pipeline:

from PyFLASH import correlation

result = correlation(
    batch,
    data_cols=["GFAP_Count", "Iba1_Count", "NeuN_Count"],
    tests=("pearsonr", "spearmanr"),
    gate="fdr",
    alpha=0.05,
)

Screen group comparisons with FDR:

from PyFLASH import group_comparison

result = group_comparison(
    batch,
    data_cols=["GFAP_Count", "Iba1_Count"],
    control="Control",
    screen=True,
    gate="fdr",
    alpha=0.05,
)

Interactions

The shared bar statistics engine first drops empty groups. If fewer than two groups remain, no comparison is made.

For two valid groups, the current bar-statistics engine uses an independent t-test when both groups have at least two observations. It falls back to Mann-Whitney U only when a group is too small for the t-test path.

For three or more valid groups, the engine uses Kruskal-Wallis when normality fails, when any group is too small, or when force_nonparametric=True. If the normality path is allowed, multiple_comparison="One-Way" selects one-way ANOVA; otherwise the two-way ANOVA path is used.

gate="fdr" requires an FDR table. In group_comparison, data_overview significance audit, and rhythm, PyFLASH requires screen=True before using an FDR gate because q-values are only computed during a screen.

comparisons=None usually means "use planned comparisons from the condition list when present; otherwise use default pairwise comparisons for the valid groups."

Common Errors

  • Passing zero-based comparisons such as "0-1". Comparison strings are one-based.
  • Requesting gate="fdr" without enabling the screen that computes q-values.
  • Supplying posthoc="Tukey" for Kruskal-Wallis. The supported post-hoc names are Conover and Dunn.
  • Treating alpha as a visual transparency option in statistical contexts. In these functions it is a significance threshold.
  • Reusing comparison strings after changing group order.

See Also