Understanding what the numbers mean and how to use them for smarter analysis
Last updated: January 15, 2026
OddsRx tracks how accurate our NBA player stat projections have been across five key metrics: Points (PTS), Rebounds (REB), Assists (AST), 3-Pointers Made (3PM), and Minutes (MIN).
The goal is simple: show how close our projections are to the real box score, and make it easy to compare accuracy across teams and players.
Each stat category includes a quick performance snapshot. Here's what each metric means and why it matters:
How many player projections are included in the evaluation. Bigger n means more stable results. Smaller n (like early-season 3PM samples) can swing week to week, so treat those as "still stabilizing."
Mean Absolute Error — "On average, how many units off were we?" If Points MAE is around 5, the typical projection is about five points away from the actual result. MAE is one of the best overall accuracy summaries because it measures the size of the miss without caring whether it was high or low.
MAE after applying calibration — a feedback adjustment that helps reduce systematic mistakes. If calibrated MAE is lower than raw MAE, that's a good sign the projection system is improving as more results come in.
Whether the model is consistently too high or too low. Positive bias = projections run high. Negative bias = projections run low. You can have decent MAE but still lean in one direction — and that directional lean matters when comparing projections to prop lines.
Root Mean Squared Error — similar to MAE, but penalizes big misses more heavily. If RMSE is much higher than MAE, it usually means there are occasional large outliers from foul trouble, injuries, blowouts, or unexpected rotation changes.
Under each stat category you'll see Hit Rates — the percentage of projections that landed within a certain range of the actual result (±1, ±2, ±3, ±5, etc.).
This answers a very practical question: "How often are we close?"
Hit rates are especially useful because they reflect real-world volatility. Points and 3PM are naturally higher-variance categories, so you should expect tighter hit rates for assists and rebounds than for points.
Points are naturally swingy, but the model is generally in the ballpark. Watch whether bias stays near zero — if points projections run consistently high or low, that creates a predictable lean you can account for when comparing to market lines.
Rebounds tend to be more stable than points, so accuracy usually tightens faster. Positive bias suggests the model may be slightly overestimating rebounding opportunity — often tied to minutes or matchup assumptions.
Often one of the most consistent projection categories when role and usage are stable. Strong hit rates within small ranges indicate the model is reading ball-handling responsibilities and rotation patterns well.
3PM is inherently streaky, so you'll typically see wider error bands than assists. Early results can look great or rough depending on a relatively small sample — best interpreted with sample size in mind.
The toughest stat to project. Minutes swing wildly based on blowouts, foul trouble, injuries, rest, coaching decisions, and rotation experiments. Minutes matter because they affect everything downstream — if minutes are off, points/rebounds/assists projections will usually drift too.
The By Team tables break out accuracy stats by team, helping identify which teams are easier or harder to project.
Some teams are stable: predictable rotations, consistent starters, steady minutes. Others are volatile: deep benches, frequent lineup changes, shifting roles.
Use the team view as a quick reliability check — when a team's MAE is consistently higher and hit rates are weaker, props involving that team may carry more uncertainty.
The By Player section is where this becomes most practical.
Some players are easier to project: stable minutes, stable usage, predictable rotation. Others are harder to project: bench scorers, matchup-dependent roles, injury-managed situations, or players who swing with coaching decisions.
The point isn't just "best vs worst" — it's helping you identify which players are dependable and which come with more variance.
Assists and rebounds often grade as the most consistent categories, while minutes remains the hardest to nail down.
The biggest long-term improvement lever is simple: get minutes closer. When minutes are right, everything else becomes easier to project accurately.
This page is a transparent performance tracker showing average error (MAE/RMSE), directional lean (bias), and real-world closeness (hit rates).
It is not a guarantee that tonight's projection will hit. NBA outcomes are noisy even with strong models.
The page is designed to answer: When the model is wrong, how wrong is it — and does it lean high or low?
