latent.stats.effect_size¶

Effect size calculations.

Delegates to scipy for core statistical computations.

Functions¶

cohens_d¶

cohens_d(scores_a: np.ndarray, scores_b: np.ndarray) -> float

Cohen's d effect size for two groups.

Derived from the t-statistic of an independent-samples t-test (equal variance assumed).

Reference: Cohen (1988), via scipy.stats.ttest_ind.

Args: scores_a: Scores from group A. scores_b: Scores from group B.

Returns: Cohen's d (positive means B > A).

cohens_d_paired¶

cohens_d_paired(scores_a: np.ndarray, scores_b: np.ndarray) -> float

Cohen's d_z for paired data.

Derived from the t-statistic of a paired-samples t-test.

Reference: Cohen (1988), via scipy.stats.ttest_rel.

Args: scores_a: Scores from group A. scores_b: Scores from group B (paired, same order).

Returns: Cohen's d_z (positive means B > A).

common_language_effect_size¶

common_language_effect_size(scores_a: np.ndarray, scores_b: np.ndarray) -> float

Common language effect size (probability of superiority).

Computes the probability that a randomly drawn score from B exceeds a randomly drawn score from A, using the Mann-Whitney U statistic (which handles ties as half-credit).

Reference: Vargha & Delaney (2000), via scipy.stats.mannwhitneyu.

Args: scores_a: Scores from group A. scores_b: Scores from group B.

Returns: Probability in [0, 1]. 0.5 means no difference.

odds_ratio¶

odds_ratio(a_success: int, a_total: int, b_success: int, b_total: int) -> float

Odds ratio comparing two groups with Haldane continuity correction.

Reference: Haldane-Anscombe correction. Uses the sample odds ratio formula (a * d) / (b * c) with 0.5 added to each cell to handle zero counts. scipy.stats.contingency.odds_ratio requires integer tables and cannot apply the Haldane correction directly.

Args: a_success: Successes in group A. a_total: Total in group A. b_success: Successes in group B. b_total: Total in group B.

Returns: Odds ratio (B relative to A). Values > 1 mean B has higher odds.

risk_ratio¶

risk_ratio(a_success: int, a_total: int, b_success: int, b_total: int) -> float

Risk ratio (relative risk) comparing two groups.

Reference: via scipy.stats.contingency.relative_risk.

Args: a_success: Successes in group A. a_total: Total in group A. b_success: Successes in group B. b_total: Total in group B.

Returns: Risk ratio (B relative to A). Values > 1 mean B has higher rate.