SciTeX Stats (scitex-stats)

Publication-ready statistical testing with 23 tests, effect sizes, power analysis, and APA formatting

Full Documentation · pip install scitex-stats

---

Problem

Statistical testing in Python is fragmented across scipy, statsmodels, and pingouin — each with different interfaces and output conventions. Getting publication-ready results requires substantial manual work: computing effect sizes, running power analysis, formatting to APA or journal standards. AI agents face a further barrier: they cannot call Python libraries directly and need structured, tool-based access.

Solution

scitex-stats provides a unified interface that covers the full statistical workflow:

• 23 statistical tests with automatic recommendation based on data characteristics
• Built-in effect sizes (Cohen's d, Cliff's delta, eta squared), power analysis, and APA-formatted output
• Three interfaces — Python API, CLI, and MCP server — so human researchers and AI agents use the same engine

flowchart LR
    A[Raw Data] --> B{Recommend Test}
    B --> C[Run Test]
    C --> D[Effect Size]
    C --> E[Power Analysis]
    D --> F[APA Format]
    E --> F
    F --> G[Publication-Ready Result]

    style A fill:#4a90d9,stroke:#2c3e50,color:#fff
    style B fill:#f5a623,stroke:#2c3e50,color:#fff
    style C fill:#27ae60,stroke:#2c3e50,color:#fff
    style D fill:#8e44ad,stroke:#2c3e50,color:#fff
    style E fill:#8e44ad,stroke:#2c3e50,color:#fff
    style F fill:#e74c3c,stroke:#2c3e50,color:#fff
    style G fill:#2c3e50,stroke:#1a252f,color:#fff

Loading

Figure 1. Statistical testing workflow. scitex-stats automates the full pipeline from raw data to publication-ready results: test recommendation based on data characteristics, test execution with effect size and power analysis, and APA-formatted output.

Every test returns a unified result dictionary with consistent keys:

{
  "test_method": "Student's t-test (independent)",
  "statistic": -3.210,
  "stat_symbol": "t",
  "alternative": "two-sided",
  "n_x": 30,
  "n_y": 30,
  "pvalue": 0.0022,
  "stars": "**",
  "alpha": 0.05,
  "significant": true,
  "effect_size": -0.829,
  "effect_size_metric": "Cohen's d",
  "effect_size_interpretation": "large",
  "power": 0.884,
  "H0": "μ(x) = μ(y)",
  "formatted": "t = -3.210, p = 0.0022, Cohen's d = -0.829, **"
}

Table 3. Unified result format. All 23 tests return the same dictionary structure with test statistics, p-value, effect size with interpretation, statistical power, and APA-formatted string.

Installation

Requires Python >= 3.10.

pip install scitex-stats

# With MCP server for AI agents
pip install scitex-stats[mcp]

# Everything
pip install scitex-stats[all]

SciTeX users: pip install scitex already includes Stats. Use import scitex then scitex.stats.

Quickstart

import scitex_stats as ss

# Get test recommendation
ctx = ss.StatContext(n_groups=2, sample_sizes=[30, 30], outcome_type="continuous", design="between", paired=False)
recs = ss.recommend_tests(ctx)

# Run a test
result = ss.run_test("ttest_ind", data=group1, data2=group2)

# APA-formatted output
print(result["formatted"])

Three Interfaces

Python API

import scitex_stats as ss

# Automatic test recommendation
ctx = ss.StatContext(n_groups=2, sample_sizes=[30, 30], outcome_type="continuous", design="between", paired=False)
recs = ss.recommend_tests(ctx)

# Run a test
result = ss.run_test("ttest_ind", data=group1, data2=group2)

# Effect sizes
from scitex_stats import effect_sizes
d = effect_sizes.cohens_d(group1, group2)

# Power analysis
from scitex_stats import power
n = power.sample_size_ttest(effect_size=0.5, alpha=0.05, power=0.8)

# Multiple comparison correction
from scitex_stats import correct
corrected = correct.correct_fdr(results)

# Post-hoc tests
from scitex_stats import posthoc
results = posthoc.posthoc_tukey(groups)

Full API reference

CLI Commands

scitex-stats --help-recursive                # Show all commands
scitex-stats list-python-apis                # List Python API tree
scitex-stats list-python-apis -v             # With docstrings
scitex-stats mcp list-tools                  # List MCP tools
scitex-stats mcp doctor                      # Check server health
scitex-stats mcp start                       # Start MCP server

Full CLI reference

MCP Server — for AI Agents

AI agents can run statistical tests and format publication-ready results autonomously.

Tool	Description
recommend_tests	Recommend appropriate tests based on data characteristics
run_test	Execute a statistical test on provided data
format_results	Format results in journal style (APA, Nature, etc.)
power_analysis	Calculate statistical power or required sample size
correct_pvalues	Apply multiple comparison correction
describe	Calculate descriptive statistics
effect_size	Calculate effect size between groups
normality_test	Test whether data follows normal distribution
posthoc_test	Run post-hoc pairwise comparisons
p_to_stars	Convert p-value to significance stars

Table 1. MCP tools available for AI agent integration via scitex-stats mcp start.

scitex-stats mcp start

Full MCP specification

Choosing the Right Test

Figure 2. Decision flowchart for choosing a statistical test. Start with your data type, then follow the branches based on number of groups and study design. Brunner-Munzel is recommended as the default for two-group comparisons due to its robustness to unequal variances and non-normality.

Available Tests

Category	Tests
Parametric	t-test (ind, paired, 1-sample), ANOVA (1-way, RM, 2-way)
Nonparametric	Mann-Whitney U, Wilcoxon, Kruskal-Wallis, Friedman, Brunner-Munzel
Correlation	Pearson, Spearman, Kendall, Theil-Sen
Categorical	Chi-squared, Fisher exact, McNemar, Cochran's Q
Normality	Shapiro-Wilk, Kolmogorov-Smirnov (1-sample, 2-sample)

Table 2. All 23 statistical tests organized by category.

Lint Rules

Detected by scitex-linter when this package is installed.

Rule	Severity	Message
STX-ST001	warning	scipy.stats.ttest_ind() — use stx.stats.ttest_ind() for auto effect size + CI
STX-ST002	warning	scipy.stats.mannwhitneyu() — use stx.stats.mannwhitneyu() for auto effect size
STX-ST003	warning	scipy.stats.pearsonr() — use stx.stats.pearsonr() for auto CI + power
STX-ST004	warning	scipy.stats.f_oneway() — use stx.stats.anova_oneway() for post-hoc + effect sizes
STX-ST005	warning	scipy.stats.wilcoxon() — use stx.stats.wilcoxon() for auto effect size
STX-ST006	warning	scipy.stats.kruskal() — use stx.stats.kruskal() for post-hoc + effect sizes

Part of SciTeX

SciTeX Stats is part of SciTeX. When used inside the SciTeX framework, statistical testing integrates with the full pipeline — from data loading through analysis to publication-ready figures:

import scitex

@scitex.session
def main(CONFIG=scitex.INJECTED, plt=scitex.INJECTED):
    # Load data
    data = scitex.io.load("measurements.csv")

    # Run statistical test
    result = scitex.stats.run_test("ttest_ind", data=group1, data2=group2)
    scitex.io.save(result, "stats_result.csv")

    # Visualize with figrecipe (scitex.plt)
    fig, ax = scitex.plt.subplots()
    ax.plot_box([group1, group2], labels=["Control", "Treatment"])
    ax.set_xyt("Group", "Value", f"p = {result['pvalue']:.4f} {result['stars']}")
    scitex.io.save(fig, "comparison.png")  # Saves plot + CSV data

    return 0

Figure 3. Example output combining scitex.stats (statistical test) with scitex.plt (publication-ready figure). The box plot shows group comparison with individual data points, significance bracket, p-value, and effect size — all generated from the unified result dictionary.

The ecosystem modules work together:

Module	Package	Role
scitex.stats	scitex-stats	Statistical testing, effect sizes, power analysis
scitex.plt	figrecipe	Publication-ready figures with auto CSV export
scitex.io	scitex-io	Universal file I/O (30+ formats)
scitex.clew	scitex-clew	Reproducibility verification via hash DAGs

The SciTeX system follows the Four Freedoms for Research below, inspired by the Free Software Definition:

Four Freedoms for Research

0. The freedom to run your research anywhere — your machine, your terms.
1. The freedom to study how every step works — from raw data to final manuscript.
2. The freedom to redistribute your workflows, not just your papers.
3. The freedom to modify any module and share improvements with the community.

AGPL-3.0 — because we believe research infrastructure deserves the same freedoms as the software it runs on.

---