If you've ever wondered what's actually happening behind a player prop projection, you're in the right place. At NFL Data Edge, we don't guess, and we don't peek at results after the fact. Here's a clear look at how our projections come together — from data to number.
We Only Use What Was Knowable Before Kickoff
The first rule of honest projections: no hindsight. Every projection we generate is built strictly from data available before the game starts. We never adjust a projection because we know how the game turned out — that would defeat the entire purpose of forecasting.
Our raw data comes from nflverse, an open and widely respected NFL data source. We're not repackaging someone else's paid picks or private algorithms — we're building our own models from public, verifiable play-by-play and game data.
What We Project — And How
Right now, NFL Data Edge covers five core markets:
- Receptions
- Receiving yards
- Rushing yards
- Passing yards
- Anytime touchdown
For each of these, we blend three key ingredients:
- Recent form — an exponentially weighted average of a player's recent games, so newer performances count more than older ones, but nothing is thrown away entirely.
- Usage — signals like target share that tell us how big a role a player has in the offense right now.
- Opponent-defense adjustment — accounting for the strength or weakness of the defense a player is facing that week.
Combining these three pieces gives us a projection that reflects both who the player has been lately and who they're about to play.
Matching the Math to the Market
Not every stat behaves the same way statistically, so we don't force one formula onto everything. Instead, we match each market to a distribution that fits its natural shape:
- Receptions use a Negative Binomial distribution, which handles the count-based, slightly clustered nature of catches.
- Yardage markets (receiving, rushing, passing) use a Gamma distribution, which suits continuous, positively-skewed outcomes like yards gained.
- Anytime touchdown uses a Poisson-based probability, well-suited to modeling rare, discrete scoring events.
Once those raw probabilities are generated, we run a calibration step to keep the over/under splits honest — making sure the probabilities we present actually reflect realistic likelihoods rather than skewed guesses.
The Takeaway
Good projections aren't magic — they're disciplined data work: recent form, usage, matchup context, the right statistical shape for each market, and a calibration check to keep everything grounded. That's the process behind every number you see from us.
Curious how it all looks in action? Explore our NFL projections and see the process for yourself.