scitex 2.10.3__py3-none-any.whl → 2.13.0__py3-none-any.whl

scitex/stats/_mcp_handlers.py

@@ -0,0 +1,1191 @@
+#!/usr/bin/env python3
+# Timestamp: 2026-01-08
+# File: src/scitex/stats/_mcp_handlers.py
+# ----------------------------------------
+
+"""Handler implementations for the scitex-stats MCP server."""
+
+from __future__ import annotations
+
+import asyncio
+from datetime import datetime
+
+import numpy as np
+
+__all__ = [
+    "recommend_tests_handler",
+    "run_test_handler",
+    "format_results_handler",
+    "power_analysis_handler",
+    "correct_pvalues_handler",
+    "describe_handler",
+    "effect_size_handler",
+    "normality_test_handler",
+    "posthoc_test_handler",
+    "p_to_stars_handler",
+]
+
+
+async def recommend_tests_handler(
+    n_groups: int = 2,
+    sample_sizes: list[int] | None = None,
+    outcome_type: str = "continuous",
+    design: str = "between",
+    paired: bool = False,
+    has_control_group: bool = False,
+    top_k: int = 3,
+) -> dict:
+    """Recommend appropriate statistical tests based on data characteristics."""
+    try:
+        from scitex.stats.auto import StatContext, recommend_tests
+
+        loop = asyncio.get_event_loop()
+
+        def do_recommend():
+            ctx = StatContext(
+                n_groups=n_groups,
+                sample_sizes=sample_sizes or [30] * n_groups,
+                outcome_type=outcome_type,
+                design=design,
+                paired=paired,
+                has_control_group=has_control_group,
+                n_factors=1,
+            )
+            tests = recommend_tests(ctx, top_k=top_k)
+
+            # Get details about each recommended test
+            from scitex.stats.auto._rules import TEST_RULES
+
+            recommendations = []
+            for test_name in tests:
+                rule = TEST_RULES.get(test_name)
+                if rule:
+                    recommendations.append(
+                        {
+                            "name": test_name,
+                            "family": rule.family,
+                            "priority": rule.priority,
+                            "needs_normality": rule.needs_normality,
+                            "needs_equal_variance": rule.needs_equal_variance,
+                            "rationale": _get_test_rationale(test_name),
+                        }
+                    )
+
+            return recommendations
+
+        recommendations = await loop.run_in_executor(None, do_recommend)
+
+        return {
+            "success": True,
+            "context": {
+                "n_groups": n_groups,
+                "sample_sizes": sample_sizes,
+                "outcome_type": outcome_type,
+                "design": design,
+                "paired": paired,
+                "has_control_group": has_control_group,
+            },
+            "recommendations": recommendations,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+def _get_test_rationale(test_name: str) -> str:
+    """Get rationale for recommending a specific test."""
+    rationales = {
+        "brunner_munzel": "Robust nonparametric test - no normality/equal variance assumptions",
+        "ttest_ind": "Classic parametric test for comparing two independent groups",
+        "ttest_paired": "Parametric test for paired/matched samples",
+        "ttest_1samp": "One-sample t-test for comparing to a population mean",
+        "mannwhitneyu": "Nonparametric alternative to independent t-test",
+        "wilcoxon": "Nonparametric alternative to paired t-test",
+        "anova": "Parametric test for comparing 3+ groups",
+        "kruskal": "Nonparametric alternative to one-way ANOVA",
+        "chi2": "Test for independence in contingency tables",
+        "fisher_exact": "Exact test for small sample contingency tables",
+        "pearson": "Parametric correlation coefficient",
+        "spearman": "Nonparametric rank correlation",
+        "kendall": "Robust nonparametric correlation for ordinal data",
+    }
+    return rationales.get(test_name, "Applicable to the given context")
+
+
+async def run_test_handler(
+    test_name: str,
+    data: list[list[float]],
+    alternative: str = "two-sided",
+) -> dict:
+    """Execute a statistical test on provided data."""
+    try:
+        from scipy import stats as scipy_stats
+
+        loop = asyncio.get_event_loop()
+
+        def do_test():
+            # Convert data to numpy arrays
+            groups = [np.array(g, dtype=float) for g in data]
+
+            result = {}
+
+            # Run the appropriate test
+            if test_name == "ttest_ind":
+                if len(groups) != 2:
+                    raise ValueError("t-test requires exactly 2 groups")
+                stat, p_value = scipy_stats.ttest_ind(
+                    groups[0], groups[1], alternative=alternative
+                )
+                df = len(groups[0]) + len(groups[1]) - 2
+                result = {
+                    "test": "Independent t-test",
+                    "statistic": float(stat),
+                    "statistic_name": "t",
+                    "p_value": float(p_value),
+                    "df": df,
+                }
+
+            elif test_name == "ttest_paired":
+                if len(groups) != 2:
+                    raise ValueError("Paired t-test requires exactly 2 groups")
+                stat, p_value = scipy_stats.ttest_rel(
+                    groups[0], groups[1], alternative=alternative
+                )
+                df = len(groups[0]) - 1
+                result = {
+                    "test": "Paired t-test",
+                    "statistic": float(stat),
+                    "statistic_name": "t",
+                    "p_value": float(p_value),
+                    "df": df,
+                }
+
+            elif test_name == "ttest_1samp":
+                if len(groups) != 1:
+                    raise ValueError("One-sample t-test requires exactly 1 group")
+                stat, p_value = scipy_stats.ttest_1samp(
+                    groups[0], 0, alternative=alternative
+                )
+                df = len(groups[0]) - 1
+                result = {
+                    "test": "One-sample t-test",
+                    "statistic": float(stat),
+                    "statistic_name": "t",
+                    "p_value": float(p_value),
+                    "df": df,
+                }
+
+            elif test_name == "brunner_munzel":
+                if len(groups) != 2:
+                    raise ValueError("Brunner-Munzel requires exactly 2 groups")
+                res = scipy_stats.brunnermunzel(
+                    groups[0], groups[1], alternative=alternative
+                )
+                result = {
+                    "test": "Brunner-Munzel test",
+                    "statistic": float(res.statistic),
+                    "statistic_name": "BM",
+                    "p_value": float(res.pvalue),
+                }
+
+            elif test_name == "mannwhitneyu":
+                if len(groups) != 2:
+                    raise ValueError("Mann-Whitney U requires exactly 2 groups")
+                stat, p_value = scipy_stats.mannwhitneyu(
+                    groups[0], groups[1], alternative=alternative
+                )
+                result = {
+                    "test": "Mann-Whitney U test",
+                    "statistic": float(stat),
+                    "statistic_name": "U",
+                    "p_value": float(p_value),
+                }
+
+            elif test_name == "wilcoxon":
+                if len(groups) != 2:
+                    raise ValueError("Wilcoxon requires exactly 2 paired groups")
+                stat, p_value = scipy_stats.wilcoxon(
+                    groups[0], groups[1], alternative=alternative
+                )
+                result = {
+                    "test": "Wilcoxon signed-rank test",
+                    "statistic": float(stat),
+                    "statistic_name": "W",
+                    "p_value": float(p_value),
+                }
+
+            elif test_name == "anova":
+                if len(groups) < 2:
+                    raise ValueError("ANOVA requires at least 2 groups")
+                stat, p_value = scipy_stats.f_oneway(*groups)
+                df_between = len(groups) - 1
+                df_within = sum(len(g) for g in groups) - len(groups)
+                result = {
+                    "test": "One-way ANOVA",
+                    "statistic": float(stat),
+                    "statistic_name": "F",
+                    "p_value": float(p_value),
+                    "df_between": df_between,
+                    "df_within": df_within,
+                }
+
+            elif test_name == "kruskal":
+                if len(groups) < 2:
+                    raise ValueError("Kruskal-Wallis requires at least 2 groups")
+                stat, p_value = scipy_stats.kruskal(*groups)
+                result = {
+                    "test": "Kruskal-Wallis H test",
+                    "statistic": float(stat),
+                    "statistic_name": "H",
+                    "p_value": float(p_value),
+                    "df": len(groups) - 1,
+                }
+
+            elif test_name == "chi2":
+                # Expects contingency table as data
+                table = np.array(data)
+                chi2, p_value, dof, expected = scipy_stats.chi2_contingency(table)
+                result = {
+                    "test": "Chi-square test of independence",
+                    "statistic": float(chi2),
+                    "statistic_name": "chi2",
+                    "p_value": float(p_value),
+                    "df": int(dof),
+                    "expected_frequencies": expected.tolist(),
+                }
+
+            elif test_name == "fisher_exact":
+                # Expects 2x2 contingency table
+                table = np.array(data)
+                if table.shape != (2, 2):
+                    raise ValueError("Fisher's exact test requires a 2x2 table")
+                odds_ratio, p_value = scipy_stats.fisher_exact(
+                    table, alternative=alternative
+                )
+                result = {
+                    "test": "Fisher's exact test",
+                    "statistic": float(odds_ratio),
+                    "statistic_name": "odds_ratio",
+                    "p_value": float(p_value),
+                }
+
+            elif test_name == "pearson":
+                if len(groups) != 2:
+                    raise ValueError("Pearson correlation requires exactly 2 variables")
+                r, p_value = scipy_stats.pearsonr(groups[0], groups[1])
+                result = {
+                    "test": "Pearson correlation",
+                    "statistic": float(r),
+                    "statistic_name": "r",
+                    "p_value": float(p_value),
+                }
+
+            elif test_name == "spearman":
+                if len(groups) != 2:
+                    raise ValueError(
+                        "Spearman correlation requires exactly 2 variables"
+                    )
+                r, p_value = scipy_stats.spearmanr(groups[0], groups[1])
+                result = {
+                    "test": "Spearman correlation",
+                    "statistic": float(r),
+                    "statistic_name": "rho",
+                    "p_value": float(p_value),
+                }
+
+            elif test_name == "kendall":
+                if len(groups) != 2:
+                    raise ValueError("Kendall correlation requires exactly 2 variables")
+                tau, p_value = scipy_stats.kendalltau(groups[0], groups[1])
+                result = {
+                    "test": "Kendall tau correlation",
+                    "statistic": float(tau),
+                    "statistic_name": "tau",
+                    "p_value": float(p_value),
+                }
+
+            else:
+                raise ValueError(f"Unknown test: {test_name}")
+
+            # Calculate effect size if applicable
+            if test_name in [
+                "ttest_ind",
+                "ttest_paired",
+                "brunner_munzel",
+                "mannwhitneyu",
+            ]:
+                from scitex.stats.effect_sizes import cliffs_delta, cohens_d
+
+                if len(groups) == 2:
+                    d = cohens_d(groups[0], groups[1])
+                    delta = cliffs_delta(groups[0], groups[1])
+                    result["effect_size"] = {
+                        "cohens_d": float(d),
+                        "cliffs_delta": float(delta),
+                    }
+
+            # Add significance determination
+            alpha = 0.05
+            result["significant"] = result["p_value"] < alpha
+            result["alpha"] = alpha
+
+            return result
+
+        result = await loop.run_in_executor(None, do_test)
+
+        return {
+            "success": True,
+            "test_name": test_name,
+            "alternative": alternative,
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def format_results_handler(
+    test_name: str,
+    statistic: float,
+    p_value: float,
+    df: float | None = None,
+    effect_size: float | None = None,
+    effect_size_name: str | None = None,
+    style: str = "apa",
+    ci_lower: float | None = None,
+    ci_upper: float | None = None,
+) -> dict:
+    """Format statistical results in journal style."""
+    try:
+        loop = asyncio.get_event_loop()
+
+        def do_format():
+            from scitex.stats.auto import format_test_line, p_to_stars
+            from scitex.stats.auto._formatting import EffectResultDict, TestResultDict
+
+            # Build test result dict
+            test_result: TestResultDict = {
+                "test_name": test_name,
+                "stat": statistic,
+                "p_raw": p_value,
+            }
+            if df is not None:
+                test_result["df"] = df
+
+            # Build effect result if provided
+            effects = None
+            if effect_size is not None:
+                effects = [
+                    EffectResultDict(
+                        name=effect_size_name or "d",
+                        label=effect_size_name or "Cohen's d",
+                        value=effect_size,
+                        ci_lower=ci_lower,
+                        ci_upper=ci_upper,
+                    )
+                ]
+
+            # Map style names
+            style_map = {
+                "apa": "apa_latex",
+                "nature": "nature",
+                "science": "science",
+                "brief": "brief",
+            }
+            style_id = style_map.get(style, "apa_latex")
+
+            # Format the line
+            formatted = format_test_line(
+                test_result,
+                effects=effects,
+                style=style_id,
+                include_n=False,
+            )
+
+            # Get stars representation
+            stars = p_to_stars(p_value)
+
+            return {
+                "formatted": formatted,
+                "stars": stars,
+            }
+
+        result = await loop.run_in_executor(None, do_format)
+
+        return {
+            "success": True,
+            "style": style,
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def power_analysis_handler(
+    test_type: str = "ttest",
+    effect_size: float | None = None,
+    alpha: float = 0.05,
+    power: float = 0.8,
+    n: int | None = None,
+    n_groups: int = 2,
+    ratio: float = 1.0,
+) -> dict:
+    """Calculate statistical power or required sample size."""
+    try:
+        loop = asyncio.get_event_loop()
+
+        def do_power():
+            from scitex.stats.power._power import power_ttest, sample_size_ttest
+
+            result = {}
+
+            if test_type == "ttest":
+                if n is not None and effect_size is not None:
+                    # Calculate power given n and effect size
+                    calculated_power = power_ttest(
+                        effect_size=effect_size,
+                        n1=n,
+                        n2=int(n * ratio),
+                        alpha=alpha,
+                        test_type="two-sample",
+                    )
+                    result = {
+                        "mode": "power_calculation",
+                        "power": calculated_power,
+                        "n1": n,
+                        "n2": int(n * ratio),
+                        "effect_size": effect_size,
+                        "alpha": alpha,
+                    }
+                elif effect_size is not None:
+                    # Calculate required sample size
+                    n1, n2 = sample_size_ttest(
+                        effect_size=effect_size,
+                        power=power,
+                        alpha=alpha,
+                        ratio=ratio,
+                    )
+                    result = {
+                        "mode": "sample_size_calculation",
+                        "required_n1": n1,
+                        "required_n2": n2,
+                        "total_n": n1 + n2,
+                        "effect_size": effect_size,
+                        "target_power": power,
+                        "alpha": alpha,
+                    }
+                else:
+                    raise ValueError("Either n or effect_size must be provided")
+
+            elif test_type == "anova":
+                # Simplified ANOVA power (using f = d * sqrt(k-1) / sqrt(2k))
+                if effect_size is None:
+                    raise ValueError("effect_size required for ANOVA power")
+
+                # Convert Cohen's f to d for approximation
+                # This is a simplified calculation
+                from scipy import stats as scipy_stats
+
+                if n is not None:
+                    df1 = n_groups - 1
+                    df2 = n_groups * n - n_groups
+                    nc = effect_size**2 * n * n_groups
+                    f_crit = scipy_stats.f.ppf(1 - alpha, df1, df2)
+                    power_val = 1 - scipy_stats.ncf.cdf(f_crit, df1, df2, nc)
+                    result = {
+                        "mode": "power_calculation",
+                        "power": power_val,
+                        "n_per_group": n,
+                        "n_groups": n_groups,
+                        "effect_size_f": effect_size,
+                        "alpha": alpha,
+                    }
+                else:
+                    # Binary search for n
+                    n_min, n_max = 2, 1000
+                    while n_max - n_min > 1:
+                        n_mid = (n_min + n_max) // 2
+                        df1 = n_groups - 1
+                        df2 = n_groups * n_mid - n_groups
+                        nc = effect_size**2 * n_mid * n_groups
+                        f_crit = scipy_stats.f.ppf(1 - alpha, df1, df2)
+                        power_val = 1 - scipy_stats.ncf.cdf(f_crit, df1, df2, nc)
+                        if power_val < power:
+                            n_min = n_mid
+                        else:
+                            n_max = n_mid
+
+                    result = {
+                        "mode": "sample_size_calculation",
+                        "required_n_per_group": n_max,
+                        "total_n": n_max * n_groups,
+                        "n_groups": n_groups,
+                        "effect_size_f": effect_size,
+                        "target_power": power,
+                        "alpha": alpha,
+                    }
+
+            elif test_type == "correlation":
+                # Power for correlation coefficient
+                from scipy import stats as scipy_stats
+
+                if effect_size is None:
+                    raise ValueError("effect_size (r) required for correlation power")
+
+                if n is not None:
+                    # Calculate power
+                    z = 0.5 * np.log((1 + effect_size) / (1 - effect_size))
+                    se = 1 / np.sqrt(n - 3)
+                    z_crit = scipy_stats.norm.ppf(1 - alpha / 2)
+                    power_val = (
+                        1
+                        - scipy_stats.norm.cdf(z_crit - z / se)
+                        + scipy_stats.norm.cdf(-z_crit - z / se)
+                    )
+                    result = {
+                        "mode": "power_calculation",
+                        "power": power_val,
+                        "n": n,
+                        "effect_size_r": effect_size,
+                        "alpha": alpha,
+                    }
+                else:
+                    # Calculate required n (binary search)
+                    z = 0.5 * np.log((1 + effect_size) / (1 - effect_size))
+                    z_crit = scipy_stats.norm.ppf(1 - alpha / 2)
+                    z_power = scipy_stats.norm.ppf(power)
+                    required_n = int(np.ceil(((z_crit + z_power) / z) ** 2 + 3))
+                    result = {
+                        "mode": "sample_size_calculation",
+                        "required_n": required_n,
+                        "effect_size_r": effect_size,
+                        "target_power": power,
+                        "alpha": alpha,
+                    }
+
+            elif test_type == "chi2":
+                # Chi-square power (simplified)
+                from scipy import stats as scipy_stats
+
+                if effect_size is None:
+                    raise ValueError("effect_size (w) required for chi2 power")
+
+                df = n_groups - 1  # Simplified: using n_groups as number of cells
+
+                if n is not None:
+                    nc = effect_size**2 * n
+                    chi2_crit = scipy_stats.chi2.ppf(1 - alpha, df)
+                    power_val = 1 - scipy_stats.ncx2.cdf(chi2_crit, df, nc)
+                    result = {
+                        "mode": "power_calculation",
+                        "power": power_val,
+                        "n": n,
+                        "df": df,
+                        "effect_size_w": effect_size,
+                        "alpha": alpha,
+                    }
+                else:
+                    # Binary search for n
+                    n_min, n_max = 10, 10000
+                    while n_max - n_min > 1:
+                        n_mid = (n_min + n_max) // 2
+                        nc = effect_size**2 * n_mid
+                        chi2_crit = scipy_stats.chi2.ppf(1 - alpha, df)
+                        power_val = 1 - scipy_stats.ncx2.cdf(chi2_crit, df, nc)
+                        if power_val < power:
+                            n_min = n_mid
+                        else:
+                            n_max = n_mid
+
+                    result = {
+                        "mode": "sample_size_calculation",
+                        "required_n": n_max,
+                        "df": df,
+                        "effect_size_w": effect_size,
+                        "target_power": power,
+                        "alpha": alpha,
+                    }
+
+            else:
+                raise ValueError(f"Unknown test_type: {test_type}")
+
+            return result
+
+        result = await loop.run_in_executor(None, do_power)
+
+        return {
+            "success": True,
+            "test_type": test_type,
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def correct_pvalues_handler(
+    pvalues: list[float],
+    method: str = "fdr_bh",
+    alpha: float = 0.05,
+) -> dict:
+    """Apply multiple comparison correction to p-values."""
+    try:
+        loop = asyncio.get_event_loop()
+
+        def do_correct():
+            from statsmodels.stats.multitest import multipletests
+
+            # Map method names
+            method_map = {
+                "bonferroni": "bonferroni",
+                "fdr_bh": "fdr_bh",
+                "fdr_by": "fdr_by",
+                "holm": "holm",
+                "sidak": "sidak",
+            }
+            sm_method = method_map.get(method, "fdr_bh")
+
+            pvals = np.array(pvalues)
+            reject, pvals_corrected, _, _ = multipletests(
+                pvals, alpha=alpha, method=sm_method
+            )
+
+            return {
+                "original_pvalues": pvalues,
+                "corrected_pvalues": pvals_corrected.tolist(),
+                "reject_null": reject.tolist(),
+                "n_significant": int(reject.sum()),
+                "n_tests": len(pvalues),
+            }
+
+        result = await loop.run_in_executor(None, do_correct)
+
+        return {
+            "success": True,
+            "method": method,
+            "alpha": alpha,
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except ImportError:
+        # Fallback implementation without statsmodels
+        try:
+            n = len(pvalues)
+            pvals = np.array(pvalues)
+
+            if method == "bonferroni":
+                corrected = np.minimum(pvals * n, 1.0)
+            elif method == "holm":
+                sorted_idx = np.argsort(pvals)
+                corrected = np.empty(n)
+                cummax = 0.0
+                for rank, idx in enumerate(sorted_idx, start=1):
+                    adj = min((n - rank + 1) * pvals[idx], 1.0)
+                    adj = max(adj, cummax)
+                    corrected[idx] = adj
+                    cummax = adj
+            elif method == "fdr_bh":
+                sorted_idx = np.argsort(pvals)
+                corrected = np.empty(n)
+                prev = 1.0
+                for rank in range(n, 0, -1):
+                    idx = sorted_idx[rank - 1]
+                    bh = pvals[idx] * n / rank
+                    val = min(bh, prev, 1.0)
+                    corrected[idx] = val
+                    prev = val
+            elif method == "sidak":
+                corrected = 1 - (1 - pvals) ** n
+            else:
+                corrected = pvals
+
+            return {
+                "success": True,
+                "method": method,
+                "alpha": alpha,
+                "original_pvalues": pvalues,
+                "corrected_pvalues": corrected.tolist(),
+                "reject_null": (corrected < alpha).tolist(),
+                "n_significant": int((corrected < alpha).sum()),
+                "n_tests": n,
+                "timestamp": datetime.now().isoformat(),
+            }
+
+        except Exception as e:
+            return {"success": False, "error": str(e)}
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def describe_handler(
+    data: list[float],
+    percentiles: list[float] | None = None,
+) -> dict:
+    """Calculate descriptive statistics for data."""
+    try:
+        loop = asyncio.get_event_loop()
+
+        def do_describe():
+            arr = np.array(data, dtype=float)
+            arr = arr[~np.isnan(arr)]  # Remove NaN
+
+            if len(arr) == 0:
+                return {"error": "No valid data points"}
+
+            percs = percentiles or [25, 50, 75]
+            percentile_values = np.percentile(arr, percs)
+
+            result = {
+                "n": int(len(arr)),
+                "mean": float(np.mean(arr)),
+                "std": float(np.std(arr, ddof=1)) if len(arr) > 1 else 0.0,
+                "var": float(np.var(arr, ddof=1)) if len(arr) > 1 else 0.0,
+                "sem": (
+                    float(np.std(arr, ddof=1) / np.sqrt(len(arr)))
+                    if len(arr) > 1
+                    else 0.0
+                ),
+                "min": float(np.min(arr)),
+                "max": float(np.max(arr)),
+                "range": float(np.max(arr) - np.min(arr)),
+                "median": float(np.median(arr)),
+                "percentiles": {
+                    str(int(p)): float(v) for p, v in zip(percs, percentile_values)
+                },
+                "iqr": float(np.percentile(arr, 75) - np.percentile(arr, 25)),
+            }
+
+            # Add skewness and kurtosis if scipy available
+            try:
+                from scipy import stats as scipy_stats
+
+                result["skewness"] = float(scipy_stats.skew(arr))
+                result["kurtosis"] = float(scipy_stats.kurtosis(arr))
+            except ImportError:
+                pass
+
+            return result
+
+        result = await loop.run_in_executor(None, do_describe)
+
+        return {
+            "success": True,
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def effect_size_handler(
+    group1: list[float],
+    group2: list[float],
+    measure: str = "cohens_d",
+    pooled: bool = True,
+) -> dict:
+    """Calculate effect size between groups."""
+    try:
+        from scitex.stats.effect_sizes import (
+            cliffs_delta,
+            cohens_d,
+            interpret_cliffs_delta,
+            interpret_cohens_d,
+        )
+
+        loop = asyncio.get_event_loop()
+
+        def do_effect_size():
+            g1 = np.array(group1, dtype=float)
+            g2 = np.array(group2, dtype=float)
+
+            result = {}
+
+            if measure == "cohens_d":
+                d = cohens_d(g1, g2)
+                result = {
+                    "measure": "Cohen's d",
+                    "value": float(d),
+                    "interpretation": interpret_cohens_d(d),
+                }
+
+            elif measure == "hedges_g":
+                # Hedges' g is Cohen's d with bias correction
+                d = cohens_d(g1, g2)
+                n1, n2 = len(g1), len(g2)
+                correction = 1 - (3 / (4 * (n1 + n2) - 9))
+                g = d * correction
+                result = {
+                    "measure": "Hedges' g",
+                    "value": float(g),
+                    "interpretation": interpret_cohens_d(g),  # Same thresholds
+                }
+
+            elif measure == "glass_delta":
+                # Glass's delta uses only control group std
+                mean_diff = np.mean(g1) - np.mean(g2)
+                delta = mean_diff / np.std(g2, ddof=1)
+                result = {
+                    "measure": "Glass's delta",
+                    "value": float(delta),
+                    "interpretation": interpret_cohens_d(delta),
+                }
+
+            elif measure == "cliffs_delta":
+                delta = cliffs_delta(g1, g2)
+                result = {
+                    "measure": "Cliff's delta",
+                    "value": float(delta),
+                    "interpretation": interpret_cliffs_delta(delta),
+                }
+
+            else:
+                raise ValueError(f"Unknown measure: {measure}")
+
+            # Add confidence interval approximation for Cohen's d
+            if measure in ["cohens_d", "hedges_g", "glass_delta"]:
+                n1, n2 = len(g1), len(g2)
+                se = np.sqrt(
+                    (n1 + n2) / (n1 * n2) + result["value"] ** 2 / (2 * (n1 + n2))
+                )
+                result["ci_lower"] = float(result["value"] - 1.96 * se)
+                result["ci_upper"] = float(result["value"] + 1.96 * se)
+
+            return result
+
+        result = await loop.run_in_executor(None, do_effect_size)
+
+        return {
+            "success": True,
+            "group1_n": len(group1),
+            "group2_n": len(group2),
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def normality_test_handler(
+    data: list[float],
+    method: str = "shapiro",
+) -> dict:
+    """Test whether data follows a normal distribution."""
+    try:
+        from scipy import stats as scipy_stats
+
+        loop = asyncio.get_event_loop()
+
+        def do_normality():
+            arr = np.array(data, dtype=float)
+            arr = arr[~np.isnan(arr)]
+
+            if len(arr) < 3:
+                return {"error": "Need at least 3 data points"}
+
+            result = {}
+
+            if method == "shapiro":
+                stat, p_value = scipy_stats.shapiro(arr)
+                result = {
+                    "test": "Shapiro-Wilk",
+                    "statistic": float(stat),
+                    "statistic_name": "W",
+                    "p_value": float(p_value),
+                }
+
+            elif method == "dagostino":
+                if len(arr) < 8:
+                    return {"error": "D'Agostino test requires at least 8 samples"}
+                stat, p_value = scipy_stats.normaltest(arr)
+                result = {
+                    "test": "D'Agostino-Pearson",
+                    "statistic": float(stat),
+                    "statistic_name": "K2",
+                    "p_value": float(p_value),
+                }
+
+            elif method == "anderson":
+                res = scipy_stats.anderson(arr, dist="norm")
+                # Use 5% significance level
+                idx = 2  # Index for 5% level
+                result = {
+                    "test": "Anderson-Darling",
+                    "statistic": float(res.statistic),
+                    "statistic_name": "A2",
+                    "critical_value_5pct": float(res.critical_values[idx]),
+                    "normal": bool(res.statistic < res.critical_values[idx]),
+                }
+
+            elif method == "lilliefors":
+                try:
+                    from statsmodels.stats.diagnostic import lilliefors
+
+                    stat, p_value = lilliefors(arr, dist="norm")
+                    result = {
+                        "test": "Lilliefors",
+                        "statistic": float(stat),
+                        "statistic_name": "D",
+                        "p_value": float(p_value),
+                    }
+                except ImportError:
+                    return {"error": "statsmodels required for Lilliefors test"}
+
+            else:
+                raise ValueError(f"Unknown method: {method}")
+
+            # Add interpretation
+            if "p_value" in result:
+                result["is_normal"] = result["p_value"] >= 0.05
+                result["interpretation"] = (
+                    "Data appears normally distributed (p >= 0.05)"
+                    if result["is_normal"]
+                    else "Data deviates from normal distribution (p < 0.05)"
+                )
+
+            return result
+
+        result = await loop.run_in_executor(None, do_normality)
+
+        return {
+            "success": True,
+            "method": method,
+            "n": len(data),
+            **result,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def posthoc_test_handler(
+    groups: list[list[float]],
+    group_names: list[str] | None = None,
+    method: str = "tukey",
+    control_group: int = 0,
+) -> dict:
+    """Run post-hoc pairwise comparisons."""
+    try:
+        loop = asyncio.get_event_loop()
+
+        def do_posthoc():
+            group_arrays = [np.array(g, dtype=float) for g in groups]
+            names = group_names or [f"Group_{i + 1}" for i in range(len(groups))]
+
+            comparisons = []
+
+            if method == "tukey":
+                from scipy import stats as scipy_stats
+
+                # All pairwise comparisons with Tukey HSD approximation
+                all_data = np.concatenate(group_arrays)
+                group_labels = np.concatenate(
+                    [[names[i]] * len(g) for i, g in enumerate(group_arrays)]
+                )
+
+                # Use statsmodels if available, otherwise manual calculation
+                try:
+                    from statsmodels.stats.multicomp import pairwise_tukeyhsd
+
+                    tukey = pairwise_tukeyhsd(all_data, group_labels)
+
+                    for i in range(len(tukey.summary().data) - 1):
+                        row = tukey.summary().data[i + 1]
+                        comparisons.append(
+                            {
+                                "group1": str(row[0]),
+                                "group2": str(row[1]),
+                                "mean_diff": float(row[2]),
+                                "p_adj": float(row[3]),
+                                "ci_lower": float(row[4]),
+                                "ci_upper": float(row[5]),
+                                "reject": bool(row[6]),
+                            }
+                        )
+                except ImportError:
+                    # Fallback: Bonferroni-corrected t-tests
+                    n_comparisons = len(groups) * (len(groups) - 1) // 2
+                    for i in range(len(groups)):
+                        for j in range(i + 1, len(groups)):
+                            stat, p = scipy_stats.ttest_ind(
+                                group_arrays[i], group_arrays[j]
+                            )
+                            p_adj = min(p * n_comparisons, 1.0)
+                            comparisons.append(
+                                {
+                                    "group1": names[i],
+                                    "group2": names[j],
+                                    "mean_diff": float(
+                                        np.mean(group_arrays[i])
+                                        - np.mean(group_arrays[j])
+                                    ),
+                                    "t_statistic": float(stat),
+                                    "p_value": float(p),
+                                    "p_adj": float(p_adj),
+                                    "reject": p_adj < 0.05,
+                                }
+                            )
+
+            elif method == "dunnett":
+                from scipy import stats as scipy_stats
+
+                # Compare all groups to control
+                control = group_arrays[control_group]
+                n_comparisons = len(groups) - 1
+
+                for i, (name, group) in enumerate(zip(names, group_arrays)):
+                    if i == control_group:
+                        continue
+                    stat, p = scipy_stats.ttest_ind(group, control)
+                    p_adj = min(p * n_comparisons, 1.0)
+                    comparisons.append(
+                        {
+                            "group": name,
+                            "vs_control": names[control_group],
+                            "mean_diff": float(np.mean(group) - np.mean(control)),
+                            "t_statistic": float(stat),
+                            "p_value": float(p),
+                            "p_adj": float(p_adj),
+                            "reject": p_adj < 0.05,
+                        }
+                    )
+
+            elif method == "games_howell":
+                from scipy import stats as scipy_stats
+
+                # Games-Howell doesn't assume equal variances
+                for i in range(len(groups)):
+                    for j in range(i + 1, len(groups)):
+                        g1, g2 = group_arrays[i], group_arrays[j]
+                        n1, n2 = len(g1), len(g2)
+                        m1, m2 = np.mean(g1), np.mean(g2)
+                        v1, v2 = np.var(g1, ddof=1), np.var(g2, ddof=1)
+
+                        se = np.sqrt(v1 / n1 + v2 / n2)
+                        t_stat = (m1 - m2) / se
+
+                        # Welch-Satterthwaite df
+                        df = (v1 / n1 + v2 / n2) ** 2 / (
+                            (v1 / n1) ** 2 / (n1 - 1) + (v2 / n2) ** 2 / (n2 - 1)
+                        )
+
+                        p = 2 * (1 - scipy_stats.t.cdf(abs(t_stat), df))
+                        n_comparisons = len(groups) * (len(groups) - 1) // 2
+                        p_adj = min(p * n_comparisons, 1.0)
+
+                        comparisons.append(
+                            {
+                                "group1": names[i],
+                                "group2": names[j],
+                                "mean_diff": float(m1 - m2),
+                                "t_statistic": float(t_stat),
+                                "df": float(df),
+                                "p_value": float(p),
+                                "p_adj": float(p_adj),
+                                "reject": p_adj < 0.05,
+                            }
+                        )
+
+            elif method == "dunn":
+                from scipy import stats as scipy_stats
+
+                # Dunn's test for Kruskal-Wallis post-hoc
+                all_data = np.concatenate(group_arrays)
+                ranks = scipy_stats.rankdata(all_data)
+
+                # Assign ranks to groups
+                idx = 0
+                group_ranks = []
+                for g in group_arrays:
+                    group_ranks.append(ranks[idx : idx + len(g)])
+                    idx += len(g)
+
+                n_total = len(all_data)
+                n_comparisons = len(groups) * (len(groups) - 1) // 2
+
+                for i in range(len(groups)):
+                    for j in range(i + 1, len(groups)):
+                        n_i, n_j = len(group_arrays[i]), len(group_arrays[j])
+                        r_i, r_j = np.mean(group_ranks[i]), np.mean(group_ranks[j])
+
+                        se = np.sqrt(n_total * (n_total + 1) / 12 * (1 / n_i + 1 / n_j))
+                        z = (r_i - r_j) / se
+                        p = 2 * (1 - scipy_stats.norm.cdf(abs(z)))
+                        p_adj = min(p * n_comparisons, 1.0)
+
+                        comparisons.append(
+                            {
+                                "group1": names[i],
+                                "group2": names[j],
+                                "mean_rank_diff": float(r_i - r_j),
+                                "z_statistic": float(z),
+                                "p_value": float(p),
+                                "p_adj": float(p_adj),
+                                "reject": p_adj < 0.05,
+                            }
+                        )
+
+            else:
+                raise ValueError(f"Unknown method: {method}")
+
+            return comparisons
+
+        comparisons = await loop.run_in_executor(None, do_posthoc)
+
+        return {
+            "success": True,
+            "method": method,
+            "n_groups": len(groups),
+            "n_comparisons": len(comparisons),
+            "comparisons": comparisons,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+async def p_to_stars_handler(
+    p_value: float,
+    thresholds: list[float] | None = None,
+) -> dict:
+    """Convert p-value to significance stars."""
+    try:
+        thresh = thresholds or [0.001, 0.01, 0.05]
+
+        if p_value < thresh[0]:
+            stars = "***"
+            significance = f"p < {thresh[0]}"
+        elif p_value < thresh[1]:
+            stars = "**"
+            significance = f"p < {thresh[1]}"
+        elif p_value < thresh[2]:
+            stars = "*"
+            significance = f"p < {thresh[2]}"
+        else:
+            stars = "ns"
+            significance = f"p >= {thresh[2]} (not significant)"
+
+        return {
+            "success": True,
+            "p_value": p_value,
+            "stars": stars,
+            "significance": significance,
+            "thresholds": thresh,
+            "timestamp": datetime.now().isoformat(),
+        }
+
+    except Exception as e:
+        return {"success": False, "error": str(e)}
+
+
+# EOF