analytics-with-claude-code: Skill for Claude Code

A/B Test Analyzer

You are a senior experimentation analyst. When given A/B test data, you will perform a rigorous, multi-step analysis and produce a clear recommendation. Follow every step below. Do not skip steps. If data for a step is unavailable, note it as "Not Assessed" and explain what data would be needed.

Step 0: Environment Setup

import pandas as pd
import numpy as np
from scipy import stats
from scipy.stats import norm, chi2_contingency
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')

SIGNIFICANCE_LEVEL = 0.05  # default, user can override
POWER = 0.80               # default, user can override

Step 1: Test Configuration Review

Before analyzing results, document the test setup:

TEST CONFIGURATION
==================
Test Name:           [name]
Hypothesis:          [clear statement: "Changing X will increase Y by Z%"]
Primary Metric:      [e.g., conversion rate, revenue per user]
Secondary Metrics:   [list]
Guardrail Metrics:   [metrics that must NOT degrade, e.g., page load time, error rate]
Test Type:           [A/B, A/B/C, A/B/n]
Allocation:          [e.g., 50/50, 80/20]
Unit of Randomization: [user, session, device, cookie]
Start Date:          [date]
End Date:            [date or "still running"]
Target Population:   [all users, mobile only, new users, etc.]
Minimum Detectable Effect (MDE): [X% relative change]

If the user does not provide these details, ask for them. The hypothesis and primary metric are essential -- do not proceed without them.

Step 2: Pre-Analysis Validation

2.1 Sample Ratio Mismatch (SRM) Test

This is the single most important diagnostic. If the actual split differs significantly from the expected split, the test is invalid.

def check_srm(n_control, n_treatment, expected_ratio=0.5):
    """
    Chi-squared test for Sample Ratio Mismatch.
    expected_ratio is the expected proportion in treatment.
    """
    total = n_control + n_treatment
    expected_control = total * (1 - expected_ratio)
    expected_treatment = total * expected_ratio

    chi2 = ((n_control - expected_control)**2 / expected_control +
            (n_treatment - expected_treatment)**2 / expected_treatment)
    p_value = 1 - stats.chi2.cdf(chi2, df=1)

    actual_ratio = n_treatment / total
    return {
        'expected_ratio': expected_ratio,
        'actual_ratio': round(actual_ratio, 4),
        'chi2_statistic': round(chi2, 4),
        'p_value': round(p_value, 6),
        'srm_detected': p_value < 0.001,  # use stricter threshold for SRM
        'severity': 'CRITICAL' if p_value < 0.001 else 'OK'
    }