About luckyduckstats

What this is

luckyduckstats is a set of statistical models for the NFL, the NBA, Major League Baseball, the NHL, PGA Tour golf, and ATP and WTA tennis, sitting on top of the ordinary record of what happened — scores, schedules, box scores, standings and rosters, all in the same shape across the six.

What the models do

Each sport gets its own model. They share no assumptions and are not one system with six settings, because the questions are not alike: a golf tournament is a hundred and fifty players over four rounds, a tennis match is two players and a surface, and a baseball game is two lineups against two pitching staffs.

They answer two kinds of question. The first is how likely each side of a game is — a number that can be set beside a bookmaker's price to see whether the two disagree. The second is what a particular player is likely to do: how many strikeouts, how many rushing yards, how many rebounds.

A projection here is a distribution, not a guess at one number. The useful output is not "about seven strikeouts" but the chance of five, six, seven or more, because that is the shape of the question anyone actually has. Golf is the one that is simulated rather than solved: the field plays the tournament through, round by round and cut included, tens of thousands of times over, and the finishes are counted.

One selection a day, in public

Each day one page carries a single selection: the model's probability, the price it was measured against, and the reasoning, written out with the numbers that produced it. Under it sits the record of the last thirty that have been graded, in the order they happened.

It is not advice. Nothing is staked here, a unit is only a way to compare one result with another, and what you do with any of it is your call and your risk.