Programs

Quantitative modeling in sport.

Analysts and researchers who want modeling work they can defend in a technical conversation, from clean analysis through full Bayesian workflows.

Ask about this program See the plan
Enrolling16 weeks · Foundations to Applied modeling to Capstone

No job is guaranteed. We do not promise placement, an interview, or an offer.

Is this you

Three things people say before they start.

If none of these are familiar, this is probably not the right program, and the call will tell you that faster than the page will.

  • You can fit a model. You have never had to defend one to somebody who does not trust it.
  • You have read about hierarchical models and never had a reason to build one.
  • Your work lives in a notebook and stops there.
Greg Matthews on modelingReading a cross-validation plot: which lambda to take, and why.

This is a real session, not a promotional cut. Modeling runs live alongside the program.

The plan

16 weeks, week by week.

Each phase names the skills you work on, what exists at the end of it, and what you can do by then.

Languages and tools across the program
PythonRStanPyMC
Weeks 1 to 4

Foundations

The analysis layer everything else sits on, and the working habits that make it reviewable by someone other than you.

 In these weeks
  • Python and pandas on real sports data
  • Reproducible project structure, not notebooks in a folder
  • End-to-end git workflow: branches, review, and QA gates before anything lands
  • Exploratory analysis that answers a stated question
What exists at the end
A reproducible analysisA reviewed pull request
By now you can

Take a messy sports dataset and produce an analysis another person can clone, run, and get the same numbers from.

Weeks 5 to 10

Applied modeling

Uncertainty done properly, coded rather than called, in both ecosystems the job descriptions name.

 In these weeks
  • Bayesian workflow in Stan and in PyMC
  • GLMs, then hierarchical models with real partial pooling
  • Prior predictive and posterior predictive checks
  • Model comparison, and knowing when to stop trusting a fit
  • Simulated technical interviews, recorded and reviewed
What exists at the end
A hierarchical model, fit and checkedA model criticism writeup
By now you can

Specify a hierarchical model, defend the priors you chose, and show the diagnostics that say whether it worked.

Weeks 11 to 16

Capstone

The build, then the two rooms it has to survive: a technical one and a non-technical one.

 In these weeks
  • Capstone model built end to end on a question you choose
  • Uncertainty-honest forecasts, intervals as the deliverable
  • Front-office presentation practice: the recommendation, and the pushback
  • AI orchestration training: planning work for delegation, delegated builds, browser QA
  • Simulated technical interviews, recorded and reviewed
What exists at the end
The capstone modelA presentation deckA portfolio writeup
By now you can

Present a model to a room that is not technical, and answer the pushback without retreating into jargon.

Not sure which phase you would start from? That is the first thing the call works out.

Ask about this program
Inside the program

The 6 sessions and courses it is built from.

Each one is on sale on its own, as a recording you buy once, on its own page. The program is the order, the weekly work around them and the capstone at the end.

Sessions 4

Async masterclass · 75 min HOT START OR REAL? Hierarchical Modeling using PyMC in Baseball Chris FonnesbeckPhD · Biostatistics @ Vanderbilt University Medical Center Available now After this you can rate every player on a roster, including the ones with almost no data, with partial pooling in PyMC.
Async masterclass · 75 min TRUST THE CHAINS MCMC in Practice Chris FonnesbeckPhD · Biostatistics @ Vanderbilt University Medical Center Available now After this you can tell a sampler that ran clean from a model that is actually right, and fix the three classic failures on your own data.
Async masterclass · 75 min SECONDS, NOT HOURS Variational Inference in Sports Chris FonnesbeckPhD · Biostatistics @ Vanderbilt University Medical Center Available now After this you can fit Bayesian models across a whole roster with variational inference and know when to go back to sampling.

Courses 2

Course · 80 lessons PRIOR. DATA. POSTERIOR. Becoming a BayeZian Scott SpencerPhD · Lecturer, Applied Analytics @ Columbia University Available now After this you can write a Bayesian model in Stan and say how sure you are, with numbers.
Course · 136 lessons MODEL ANYTHING. Becoming a BayeZian II Scott SpencerPhD · Lecturer, Applied Analytics @ Columbia University Available now After this you can put survival, Gaussian process and physics-constrained models into production Stan.
The Coaching Room

Practice with somebody on the other side of the table.

AI coaching is built into the practice sessions. The coach takes the stakeholder's part, and you work the problem in whatever medium the question actually calls for.

AI as copilot, not autopilot. You learn to direct the machine and defend the output. The scarce part is not producing an answer; it is knowing whether the answer is right and being able to say why in a room that will push back.

Simulated practice

A clock, a real task, and a room that pushes back.

How it actually runs, and the kind of question it runs on.

Practice is timed. You get a real task and a clock, not a quiz. The session is recorded, and the review afterwards is about the reasoning you showed rather than whether you reached the expected answer.

Scenarios are written the way the question actually arrives
  • A coach asks: our closer has thrown 9 innings in 6 days, what does the fatigue data say about tonight?
  • A director asks: this projection has us winning 4 more games than last year. How confident are you, and what would change your mind?
  • An executive asks: explain to me, without the word posterior, why this model disagrees with what our scouts think.
Questions

The ones people ask first.

Will I get a job at the end?

No job is guaranteed. We do not promise placement, an interview, or an offer. What the program promises is the work: a capstone you can defend, a portfolio with real builds in it, and interview practice that is recorded and reviewed.

What does it cost?

There is no public price and no checkout on this page. Cost, dates and where you would start are settled on a short call, and the honest answer is sometimes that a single session or course is the better start.

Which weeks would I start from?

That is the first thing the call works out. The plan on this page is the whole route; a person who already has the foundations starts further along it.

Not into sports?

Then take the method and leave the sport. Hierarchical modeling, calibrated uncertainty and honest inference are the methods; sport is where you practice them. The same model, with the nouns changed, answers questions in these fields:

  • Pharma Clinical trials are hierarchical modeling with the stakes raised
  • Finance Trading and risk work runs on calibrated uncertainty, not point estimates
  • Marketing Marketing mix modeling is partial pooling under another name
  • Polling Election modeling is the same problem: small samples, grouped structure
  • Insurance Pricing and reserving run on exactly this stack
What runs alongside the taught weeks?

Simulated technical interviews, recorded and reviewed. Front-office presentation practice, where a recommendation meets a room that pushes back. AI orchestration: planning work for delegation, delegated builds, and the checks that catch what a model got wrong.

Modeling is the live track running alongside the program, so the weekly sessions and the plan pull in the same direction.

Next step

Ask.

There is no public price and no checkout on this page. A short call settles cost, dates and where you would start, and if a single session is the better start we will say so.

No job is guaranteed. We do not promise placement, an interview, or an offer.

Ask about this program
Keep up

One email when something new arrives.

First name and email. One message when a session recording, a skill, a cheat sheet or a posting worth your time arrives here, and no other mail. Replies reach a person.