Technical posts on A/B testing, hypothesis testing, regression, time series and related methods.

CUPED — making your A/B tests more sensitive

How pre-experiment data can help you detect smaller effects with fewer users.

Is this Simpson's Paradox?

A common misdiagnosis in A/B test results, and how to tell the difference.

Sequential testing — making decisions before your experiment ends

Why peeking at results inflates your false positive rate, and how sequential testing fixes it.

Way too much about linear regression.

There's a lot of details

AB testing Google Ads Bidding strategies

Hypothesis testing with dependent samples.

Kernel-MMD-Hypothesis Testing without assumptions

What to do if you can't assume anything about your distributions.

Hypothesis Testing by Betting

A simpler alternative to $p$-values.

A short introduction to Martingales

One way to think about dependent random variables.

How wide should your confidence interval be?

A short, hands-on, guide.

What p-values really mean

Most people get $p$-values wrong. This is how to understand and apply them correctly.

An introduction to Compressive Sensing.

Short, fat matrices are useful, actually.

A Grab Bag of Approaches to Frequentist Multiple Testing.

Pick your poison.

Must we adjust p-values if we test multiple hypotheses?

Yes, yes you must

Bayesian Sequential Hypothesis Testing

Frequentist algorithms are often Bayesian.

Sequential Hypothesis Testing

Hypothesis testing step-by-step.

Running Conda environments from Jupyter Notebook.

An introduction to Gradient Descent

How to optimise basically any function.

Time Series with GAMs

Time series without time.

Structural Time Series in PyMC!

Same ideas, different framework.

Understanding the Guts of Generalized Additive Models (GAMs) with Hands-on Examples.

Generalised Additive Models look harder than they actually are.

Neural Networks from scratch.

NNs are easier than you think

Bayesian Structural Time Series in pystan.

Adding time varying components to time varying regression

More Complex (Linear) Regressions

We extend our ideas of how to do regression to a more complex class of functions.

Local Linear Trend models for time series

Simple time-varying regression

Mixed Models

How to handle struture in your regressions