The League Through Time
80 seasons
See how the title race, scoring levels and competition between clubs have evolved over time.
League
Goals Journey — the whole league
Every goal scored in the league, added up after every matchround.
Season
Compare with
Matches without a goal — the other end of the story
When do the goals come?
Every match above six goals — the list breathes with the season, so a tie is never cut. Club names open that club’s own Goals Journey at this match.
Every match above six goals — the list breathes with the season, so a tie is never cut. Club names open that club’s own Goals Journey at this match.
Goals per match through the season
How long do the clubs remain in the title race?
Title race with matches to go — table
The closest title fights first — the gap from champion to runner-up, in points. Points compare only within one points system.
The podium, season by season
The champion and the two clubs behind them, newest season first, with the points each finished on.
These rows are in date order, not ranked. A rule marks every change of points system and each block says which system it was played under — points either side of a rule are not comparable quantities, so nothing here ranks one season against another.
How far apart do the clubs finish, and how much does the order look like last season? Two questions, one answer.
Two questions, deliberately kept apart. How far apart the clubs finish is a fact about one season. How much of the order stays the same is a fact about two. They are different questions and a league can move on one and not the other — which is exactly what happens here, and why they share a card rather than a score.
The verdict is computed from tested trends, not from the first decade against the last. Endpoints mislead in both directions: a league whose decade averages fall can have a rising fitted trend, and one whose averages rise can have no trend that survives testing. Each trend is fitted over every season and tested with a moving-block bootstrap, which resamples the residuals in blocks so that the season-to-season correlation survives the resample — an ordinary regression assumes each season is independent of the last, which for a league table is plainly false.
Four answers, not two. A trend is stated as fact below p = 0.01, stated with a caveat below 0.05, called unclear below 0.15, and otherwise reported as no trend. "Unclear" is not the same claim as "no trend": the first says the evidence could not settle it, the second says it did. The faint straight line on each chart is the line the test fitted, so what the words say and what the picture shows cannot come apart.
Both measures have real weaknesses, and both are written out in full under the two season-by-season tables below.
The verdict is computed from tested trends, not from the first decade against the last. Endpoints mislead in both directions: a league whose decade averages fall can have a rising fitted trend, and one whose averages rise can have no trend that survives testing. Each trend is fitted over every season and tested with a moving-block bootstrap, which resamples the residuals in blocks so that the season-to-season correlation survives the resample — an ordinary regression assumes each season is independent of the last, which for a league table is plainly false.
Four answers, not two. A trend is stated as fact below p = 0.01, stated with a caveat below 0.05, called unclear below 0.15, and otherwise reported as no trend. "Unclear" is not the same claim as "no trend": the first says the evidence could not settle it, the second says it did. The faint straight line on each chart is the line the test fitted, so what the words say and what the picture shows cannot come apart.
Both measures have real weaknesses, and both are written out in full under the two season-by-season tables below.
Every season pair, one by one
How much of one season's order was still there the next — the biggest shake-ups first.
In short: did the table look like last year's? 1.00 means an identical order, 0 means it bore no more resemblance than chance would give, and a negative number means it turned over. Only clubs that played in both seasons can be compared, so the figure counts what the two seasons had in common.
What this measures. Spearman's rank correlation between two consecutive final tables, over the clubs that played in both. It is the one figure on this card that looks BETWEEN seasons rather than inside one, which is why it exists: every other measure here would score a league of small margins with a permanent champion exactly like one where a different club wins every year. Rank correlation is used for this in the competitive-balance literature (Kringstad and Gerrard, 2007; Groot, 2008, prefers Kendall's tau; Humphreys, 2002, builds a related ratio).
Why some rows are marked. If next season's order were unrelated to this one, the figure would still not come out at zero. It has a margin of error that depends only on how many clubs are compared: the middle 90% runs to plus or minus 1.645 divided by the square root of one less than that number — plus or minus 0.38 for twenty clubs, plus or minus 0.50 for twelve. Rows inside that band are marked, because for those the honest answer is that the table was scrambled beyond anything this measure can distinguish from chance.
The weakness, and it is a real one. Only clubs present in both seasons can be compared, and the clubs that drop out are the ones that finished bottom. Cutting the bottom off a distribution always attenuates a correlation, so every figure here is systematically LOW — the true continuity of the competition is higher than the number shown. It was checked for whether it also distorts the movement over time, and it does not: the share of a field still present the following season is flat to slightly falling across this archive, which pushes modern figures down rather than up.
What else it cannot tell you. It reads order only, so a season in which the champion changed but nothing else did scores high. It is blind to margins: a title won by a point and one won by twenty look identical to it. And promotion and relegation are a design choice of the competition rather than something this measure can hold constant.
What it gets right, and no other table here does. It is built on ORDER, not points. That makes it completely untouched by the change from two points a win to three, so it is the only ranked table on this platform that puts the whole archive in a single list without grouping by era.
Why some rows are marked. If next season's order were unrelated to this one, the figure would still not come out at zero. It has a margin of error that depends only on how many clubs are compared: the middle 90% runs to plus or minus 1.645 divided by the square root of one less than that number — plus or minus 0.38 for twenty clubs, plus or minus 0.50 for twelve. Rows inside that band are marked, because for those the honest answer is that the table was scrambled beyond anything this measure can distinguish from chance.
The weakness, and it is a real one. Only clubs present in both seasons can be compared, and the clubs that drop out are the ones that finished bottom. Cutting the bottom off a distribution always attenuates a correlation, so every figure here is systematically LOW — the true continuity of the competition is higher than the number shown. It was checked for whether it also distorts the movement over time, and it does not: the share of a field still present the following season is flat to slightly falling across this archive, which pushes modern figures down rather than up.
What else it cannot tell you. It reads order only, so a season in which the champion changed but nothing else did scores high. It is blind to margins: a title won by a point and one won by twenty look identical to it. And promotion and relegation are a design choice of the competition rather than something this measure can hold constant.
What it gets right, and no other table here does. It is built on ORDER, not points. That makes it completely untouched by the change from two points a win to three, so it is the only ranked table on this platform that puts the whole archive in a single list without grouping by era.
Every season, one by one
How far apart the clubs finished — the closest seasons first.
In short: how far apart were the clubs, once ordinary luck is taken out? A bigger number means the strong and the weak finished further apart. The spread is in win-rate points — 7 means the typical club sat about seven points of win rate from average. The ratio beside it is the classical measure, where 1.0 means a season no more spread out than chance alone would make it.
What this measures. Each season's clubs are compared with an idealised league in which every club is equally strong and every match is settled by chance at that season's own draw rate. The classical figure — the Noll–Scully ratio, from Roger Noll (1988) and Gerald Scully (1989), standard since Quirk and Fort's Pay Dirt (1992) — is the actual spread of results divided by that idealised spread. A win counts one, a draw a half, a loss nothing, measured at the deepest round every club had reached. The draw-corrected form of the idealised spread is Owen's (Journal of Sports Economics, 2012).
Why the ranking uses a different figure. The classical ratio rises with the number of matches played even when the clubs are equally matched. Simulated with club strength held fixed, it climbs from 1.76 over eighteen matches to 2.47 over forty-two — the same league scoring forty per cent less even for playing more football. This archive holds seasons of both lengths, so ranking by the ratio would order the table partly by calendar. The ranking instead uses the spread that remains once the chance component is subtracted rather than divided out, which does not depend on season length. That the ratio is length-sensitive is documented in Owen (2010), Owen and King (Economic Inquiry, 2015) and Doria and Nalebuff (2021); the subtraction is Tango's (2006) and Mauboussin's (2012), with a formally bias-corrected version in Lee, Kim and Kim (2019). Ours is the plain form, biased a little low by the square root and a little high by the clamp at zero.
Why some seasons are marked. The comparison carries a margin of error that depends on how many clubs played, and on nothing else — not season length, not the draw rate. Even a perfectly balanced ten-club league scores anywhere between 0.63 and 1.35, and the middle of that range is 0.98, so under perfect balance about half of all seasons would come in below one. A value below one is therefore neither impossible nor an error. Seasons whose ratio falls inside that band are marked, because for those the honest answer is that the clubs could not be told apart.
What it cannot tell you. It looks at one season at a time, so it cannot see whether the same club won every year — a league of small margins with a permanent champion scores like one where a different club wins each time. It treats a draw as half a win, so it cannot separate a side that drew twenty from one that won ten and lost ten. It is blind to shape: one runaway leader among fifteen equals scores much like a smooth gradient. And the idealised league it measures against cannot be achieved by any real competition, so the number orders seasons well and means little as an absolute score.
One thing it does get right. It is built on win ratio rather than points share, deliberately. The classical ratio is not invariant to the points system — under three points for a win the idealised spread is a different expression entirely — so a measure built on points would jump when a league moved from two points to three, for reasons having nothing to do with balance. This one does not move at all.
Why the ranking uses a different figure. The classical ratio rises with the number of matches played even when the clubs are equally matched. Simulated with club strength held fixed, it climbs from 1.76 over eighteen matches to 2.47 over forty-two — the same league scoring forty per cent less even for playing more football. This archive holds seasons of both lengths, so ranking by the ratio would order the table partly by calendar. The ranking instead uses the spread that remains once the chance component is subtracted rather than divided out, which does not depend on season length. That the ratio is length-sensitive is documented in Owen (2010), Owen and King (Economic Inquiry, 2015) and Doria and Nalebuff (2021); the subtraction is Tango's (2006) and Mauboussin's (2012), with a formally bias-corrected version in Lee, Kim and Kim (2019). Ours is the plain form, biased a little low by the square root and a little high by the clamp at zero.
Why some seasons are marked. The comparison carries a margin of error that depends on how many clubs played, and on nothing else — not season length, not the draw rate. Even a perfectly balanced ten-club league scores anywhere between 0.63 and 1.35, and the middle of that range is 0.98, so under perfect balance about half of all seasons would come in below one. A value below one is therefore neither impossible nor an error. Seasons whose ratio falls inside that band are marked, because for those the honest answer is that the clubs could not be told apart.
What it cannot tell you. It looks at one season at a time, so it cannot see whether the same club won every year — a league of small margins with a permanent champion scores like one where a different club wins each time. It treats a draw as half a win, so it cannot separate a side that drew twenty from one that won ten and lost ten. It is blind to shape: one runaway leader among fifteen equals scores much like a smooth gradient. And the idealised league it measures against cannot be achieved by any real competition, so the number orders seasons well and means little as an absolute score.
One thing it does get right. It is built on win ratio rather than points share, deliberately. The classical ratio is not invariant to the points system — under three points for a win the idealised spread is a different expression entirely — so a measure built on points would jump when a league moved from two points to three, for reasons having nothing to do with balance. This one does not move at all.
One period at a time: every club that played in it, ranked by the points it took.
Turn your phone for the full width