Aaron Goodin Causal science and marketing measurement

I measure the difference between what happened and what would have happened anyway.

I design incrementality tests, calibrate attribution and media mix models to lift results, and build honest baselines when an A/B test isn't possible. The projects below are how I show the work: open source, running, and free to use.

See the projectsEmail me

Nine years in data. Meta Certified Marketing Science Professional. Python, R, SQL, and NetLogo.

What happened versus what would have happened anywayInterventionWhat happenedWhat would havehappened anywayImpact

Things I've built

Each one answers a measurement question that teams usually settle by guessing.

do-attribution

Python library, published on PyPI under the MIT license

Multi-touch attribution you can audit. Closed-form absorbing Markov chains with four removal-effect definitions, exact Shapley values, changepoint-detected time epochs, and a calibration layer that reconciles channel credit with lift tests.

Platform reports hand credit to whoever touched the customer last. This library shows its math, gives the same answer on every run, and adjusts to your experiment results instead of ignoring them.

pip install do-attribution
View on PyPIRead the source

Impact Analysis Tools

Web app on Hugging Face Spaces, built in Python and Gradio

Upload a dataset, pick a method, and read the result. Standard and matched difference-in-differences, interrupted time series, Granger causality, propensity score matching with nearest-neighbor, caliper, stratification and Love plots, and forecasting with Auto-ARIMA, ETS, Prophet, and SARIMAX.

These are the methods analysts reach for when randomization wasn't possible. The app lets you run them on your own data before writing any code.

Open the app

fashion-market

Agent-based simulation that runs in your browser, built in R with webR and NetLogoR

A full fashion retail season as a digital twin. 24,000 shopper agents roll up to store P&L, with stockouts, returns, markdowns, e-commerce cannibalization, and daily ARIMA demand forecasts.

Set the strategy, offers, and assortment, then watch the market push back before you commit real inventory. Nothing to install: the R code runs in the page and every number on every tab comes from it.

Run the simulationRead the source

Interactive guides

Nineteen working references for the questions that come up in measurement reviews. Change the inputs and see what moves.

Marketing measurement

9 tools

  • Measurement plan
  • The attribution trap
  • Advanced attribution
  • MTA calibration
  • Saturation and decay
  • Incrementality
  • Surveys
  • Response models
  • Budget decision
Open marketing measurement

Customer insights

4 tools

  • Customer lifetime value and churn
  • Loyalty program lift
  • Market basket analysis
  • Share of wallet
Open customer insights

Math primer

6 tools

  • P-hacking simulator
  • P-values
  • Reading regressions
  • Sample size and power
  • t-distribution table
  • F-distribution table
Open the math primer

Bring me a measurement question

Something your dashboard can't settle, like whether a promotion pays for itself or which channel is really doing the work.

[email protected]Call 8139-CAUSAL