When tipsters agree, does it pay?

The side more tipsters back is usually the favourite, and favourites win more often at shorter prices. That is what a short price means, not evidence the crowd called it right — see the strike-rate/yield pair for the general case. Yield is the number that pays, so the verdict reads from it. Win rate stays visible because the gap between the two is itself the point.

Both are measured over advertised prices only. Some sources publish a probability rather than a price; a price derived from one carries no bookmaker margin and describes a bet nobody could have placed, so those tips are excluded from these four figures entirely.

Two tips are the same fixture when they share an event, or name the same two teams on the same UTC day. A crowd needs three or more tips on the same market, and one selection has to beat every other individually — a tie has no winner to name.

"Most-backed" is a plurality, not a majority: on a three-way market a selection only has to beat each rival on its own. A split of 5 home / 4 draw / 3 away reads as 5 against 7. The other side is a mixture of whatever different bets those tipsters made, never one shared opposing view.

These groups are backed by 801 distinct tipsters across 13 distinct sources.

These figures speak for 53.0% of the settled corpus — 24,407 of 46,011 rows carry a fixture identity we can group on. The rest carry neither an event nor a verified kickoff, and are excluded rather than guessed at.

Read the Crowds column before the yields. These buckets hold very different numbers of crowds, and a striking figure over a handful of them is a small sample, not a finding about that market. A row marked within noise is one whose two yields cannot be told apart: the 95% interval on the difference between them still contains zero, so whatever gap the columns show is inside what ordinary variance produces on its own.

Bigger crowds also gather on bigger matches, and bigger matches carry shorter prices. A difference across the crowd-size buckets can be the prices talking rather than the crowd.

These four axes are around twenty comparisons in total, each at 95% confidence. On that many, roughly one in twenty would clear the bar by chance alone even if nothing here were real, so a single separable row is weaker evidence than it looks. Two rows pointing in opposite directions on the same axis, as the crowd-composition split does, is the pattern that is hard to get by accident.

Every crowd on this page has already played and been graded, because every figure on it is a result — a yield, a strike rate, a comparison against the rest of the field. None of those exist for a fixture that has not happened, so an unplayed crowd has nothing to contribute here.

Crowds on fixtures still ahead are on the upcoming board instead, which shows the split and the tipsters and stops there. Read this page first: what the most-backed selection has historically returned is the only thing that gives a number over there any meaning. The methodology sets out which surfaces publish before an event and which do not.

Advertised prices only. Some sources publish a probability instead of a price; deriving one from it (1/p) gives a number carrying no bookmaker margin, which flatters every return computed from it. Those tips still count towards who is in a crowd — an algorithm's tip is a real tip — but never towards a figure about money.

This page reads the live corpus, so its figures move as tips settle. The corpus study closes its data at the start of each month so a number quoted from it is still there when someone checks. The two will report different figures for the same measurement, and neither is wrong. Cite the study; read this page.