Data science in sport

Data science for better decisions in sport. Statistical modelling, simulation and decision support, explained in plain language with code in R and Python.

Data science for better decisions in sport.

I spent seven years working directly with athletes and coaches in professional rugby. I now work as a data scientist, using statistical modelling and simulation to tackle the kinds of problems coaches, managers and administrators face.

I’m working towards decision-support tools that can handle complexity underneath without putting that complexity on the person making the decision.

Here I share the work in plain language, along with the data and code (usually in both R and Python).

Read the posts About and experience

Background: posterior draws from a Gaussian process fitted to nine simulated observations. R code · Python code

Portrait of Mitch Henderson

Selected work

NFL player positions with an estimated 85% probability of a successful short pass play

Sometimes interpretability doesn't cost predictive power

A model built around football concepts predicted short pass success as well as a neural network. It also made it easier to explain why the probability changed as a play unfolded.

  • Bayesian logistic regression
  • R
  • Python
Chart of estimated strength curves for three athletes, each with a shaded band showing uncertainty

Making 1 rep max estimates more accurate and honest

Standard formulas treat every athlete the same and give a single number. This model learns each athlete’s strength profile from their training history and says how confident it is in the estimate.

  • Bayesian multilevel model
  • R
  • Python
Dot plot of every NRL team's season over 10 years by matches won, coloured by whether the team made the finals

How many wins do NRL teams need to make the finals?

Probably 13. Ten thousand simulated seasons show why the finals cut-off moves from year to year, and how likely a team is to qualify at each win total.

  • Monte Carlo simulation
  • R
  • Python

What I work on

Statistical modelling

Build models that capture the important parts of the problem and quantify what we know and what we don’t.

Bayesian and multilevel models · Stan, brms

Simulation and decision support

Use models and simulation to explore what could happen, compare options, and understand the consequences of different decisions.

Monte Carlo methods · probabilistic forecasting

Making it useful

Turn the complexity underneath into results and tools that are clear and useful for the people making the decisions.

Visualisation · interactive tools · technical communication

Day to day I work in R, Python and SQL.

Earlier writing

Tutorials from 2020 and 2021 on working with GPS and wearable data in R are kept in the archive.