Methodology & Projections
How forecasting and win-probability models are built, why they disagree, and what the data really says about predicting basketball.
Forecasting is where basketball statistics stops describing and starts committing. These pieces are about the machinery: what a projection system is really fitting, why two reputable systems hand you different numbers for the same player, and how much of a win-probability curve is model rather than scoreboard. Two of them are cautionary — the plus-minus piece on why a stint-level metric needs thousands of games before it settles, and the preseason piece testing whether October results carry any signal at all into the regular season. Pythagorean expectation is the counterweight: two season totals, one exponent, and an expected record that landed within four wins of reality for 27 of 30 teams in 2023-24. Read them as a set and the pattern repeats: a method is only as good as the uncertainty it admits to.
-
The Handoff: Six Games In, This NBA Season Outranks the Last
Last season's margin is the better guide to an NBA team for five games. At game six this season's takes over; after game eight last season adds almost nothing.
-
The October Glass: Preseason Rebounding Is Real. Its Shooting Is Not.
Added to last season's numbers, NBA preseason offensive rebounding doubles what can be explained, 19.2% to 40.5%. Preseason shooting adds nothing.
-
The Giveback: Last Season's Luck Is Worth a Fifth of a Win
A win an NBA team's margin earned carries 0.63 of a win into the next season; a win beyond it carries 0.21. What that says about the Lakers and Hornets.
-
The Continuity Dividend: What Bringing a Roster Back Is Worth
Every ten points of last season's minutes an NBA team brings back is worth 0.48 points of margin, about 1.2 wins, on top of its margin. Age is a separate lever.
-
Pythagorean Expectation: Predicting the NBA Standings From Two Numbers
Points for and against alone reproduced the 2023-24 NBA standings at r=0.981 — and the teams the formula missed (Lakers, Pistons) missed on close-game record.
-
Why Plus-Minus Needs a Thousand Games: On-Off, RAPM, and Noise
Raw plus-minus is the noisiest number on the stat sheet. How adjusted plus-minus and RAPM try to fix it, and why even they need years of data.
-
How NBA Win-Probability Models Work (And Why They Disagree)
That live win-probability number is a model output, not a fact. Here's what goes into one, and why two good models can disagree by 20 points.
-
Does Preseason Performance Predict Anything? What the Data Says
Every October we read into preseason results. We pulled preseason and regular-season net ratings to see whether October tells us anything at all.
-
Why Player Projection Systems Disagree: A Look Under the Hood
Feed two projection systems the same player and they'll often hand back different futures. The disagreement is structural — here's where it comes from.