The Problem with Sports Analytics Isn't the Math

I like analytics. Let me get that out of the way. Expected goals, win probability, player tracking data — it's all useful. It tells you things your eyes can't, especially at scale.

The problem isn't analytics. It's analytics culture. The discourse around sports data has produced a certain type of fan who treats numbers as arguments-enders rather than arguments-starters. You say a player looked great tonight. They pull up a stat you've never heard of and tell you he actually had a negative expected contribution above replacement per 36 minutes adjusted for pace and altitude.

Cool. He also made a play in the fourth quarter that made the entire arena stand up, but I guess the altitude-adjusted number says otherwise.

What numbers miss

Analytics are great at measuring what happened. They're bad at measuring why. A quarterback's passer rating on third down is a useful number. It doesn't tell you that he changed the protection at the line because he saw a linebacker creep forward, bought an extra half-second, and found a receiver who adjusted his route on the fly. That sequence is the sport. The number is a shadow of it.

The best analysts — the ones actually working in front offices — know this. They use data as one input among many. They watch film. They talk to coaches. They understand that a player's value includes things that don't show up in a box score: leadership, the way he changes a locker room, the way opponents game-plan around him.

The worst analysts are the ones on Twitter with a model and an attitude. They've decided that if something can't be measured, it doesn't exist. That's not empiricism. That's the opposite — it's assuming your tools capture everything worth knowing.

The eye test isn't dead

Here's my take: the best way to evaluate a player is to watch them play and then check the numbers. Start with observation. Form an impression. Then see if the data supports it, contradicts it, or adds context you missed.

Starting with the numbers and working backward is how you end up confused when a player who grades out poorly by the metrics keeps winning. Sometimes the model is wrong. Sometimes the model is right but incomplete. Sometimes a player is doing things the model isn't designed to capture.

Data tells you what.
Film tells you how.
Context tells you why.

You need all three. The analytics-first crowd acts like the first one is sufficient. It's not.