StatsTest Blog
Experimental design, data analysis, and statistical tooling for modern teams. No hype, just the math.
Bootstrap for Heavy-Tailed Metrics: Best Practices and Gotchas
How to use bootstrap correctly for revenue, engagement, and other heavy-tailed metrics. Learn about BCa intervals, when bootstrap fails, and how many resamples you actually need.
Comparing ARPU and ARPPU: Segmentation vs. Modeling Approaches
How to properly analyze revenue per user metrics in A/B tests. Learn the statistical pitfalls of ARPU vs. ARPPU, when to segment, and how to avoid Simpson's paradox.
Dealing with Zeros: Zero-Inflated and Two-Part Models
How to handle metrics with many zeros—revenue from non-purchasers, engagement from inactive users, events that didn't happen. Learn when to use zero-inflated models, two-part models, and simpler alternatives.
Delta Method vs. Bootstrap: When Each Is Appropriate
A practical guide to choosing between delta method and bootstrap for variance estimation. Learn when each approach excels, their assumptions, and how to implement both.
Metric Distributions in Product Analytics: Heavy Tails, Skew, and What to Do
A comprehensive guide to handling real-world metric distributions in product analytics. Learn why revenue is hard, how to deal with zeros, when to transform vs. use robust methods, and how to communicate results on skewed data.
Percentiles and Latency: Comparing P50, P95, P99 Correctly
How to properly compare percentile metrics like latency P50, P95, and P99 across groups. Learn about bootstrap inference, quantile regression, and the pitfalls of naive percentile comparisons.
Ratio Metrics (CTR, Conversion): Why They're Tricky and Stable Alternatives
Why ratio metrics like CTR and conversion rates require special statistical treatment. Learn about variance estimation, the delta method, and when to use alternative approaches.
Why Revenue Is Hard: Log-Normal Distributions and Heavy Tails
A deep dive into why revenue metrics are statistically challenging. Learn about log-normal distributions, heavy tails, whale effects, and practical approaches to analyzing revenue in A/B tests.