8 of the last 10 is closer to 6 of 10, in every sport

Why?

Our first big piece ran this test on six weeks of AFL bookmaker lines and ended by promising the full-season version in every code. I wanted to know whether footy was the odd one out, or whether 8 of 10 shrinks the same way everywhere.

Every player-game, one season per code, got a line, set the way a bookmaker sets one: from the longer record, not the last few weeks. The line is the first half-point at or above the medianThe middle game once the 30 are sorted. of the player's previous 30 games, so 25 disposals in the middle of those 30 makes the line 25.5. Run metres use 99.5, 109.5 and so on.

Then two questions. Had the player gone over that line in exactly 8 of the last 10? And did the next game go over?

Five sports, one answer

Across the five codes, 1,594 of 2,622 next games went over, 60.8%. No sport broke the pattern: every code landed between 59.6% (NBA) and 63.5% (NRL).

Inside a code, the stats spread further than the codes did:

AFL tackles were the lowest of any stat with more than 20 lines. Disposals, the AFL market most multis are built from, and NBA points, the two biggest samples, both sat just above 63%. The smallest samples, like 5 lines of WNBA assists, say nothing on their own; they are in the table so nothing is hidden.

Sixty per cent is not nothing. The same lines were cleared under half the time across all games. But it is a long way from the 80% the phrase sounds like, and it is the same distance in every sport.

The more of the last 10, the more came in

Running the same test at every count, from none of the last 10 to all 10, shows how much a last-10 hit rate is really worth.

After 8 of 10, 60.8% came in. After 5 of 10, 46.7%. After none of 10, 22.8%. Most steps up the chart are worth three to five points, not the ten each extra game sounds like. Only a perfect 10 of 10 came in about as often as it sounds, 111 of 138, and that is the smallest group on the chart.

Why it shrinks

Ten games is a small sample, and a player who has cleared a line in 8 of them has usually had a good run. Good runs are real, but they end, and the longer record the line was set from pulls the next game back towards normal. It is the same reason the number next to a hit rate matters as much as the rate.

The first test used the bookmakers' own lines and found 68.9%. These stand-in lines found 60.8%. Different lines, different weeks, and the same answer: 8 of 10 turns into somewhere around 6 or 7 of 10.

Would real lines change it?

These numbers show what happened across one season per code. They can't say what the next game will do, and nothing can. The lines here are stand-ins built from each player's own record, not the prices anyone was offered. I want to run it again on the bookmakers' own lines once Punt holds a full season of them.