Great work, Martin. You see so many people just apply parametric stats to everything, so it’s nice that you illustrate the distinction.
Something I would find interesting, and possibly a future article idea for you, is to compare the actual table versus the xG table and see how they both correlate with the final table through each match week.
There are different ways of framing this, of course; the actual table does have the advantage that points accrued are used to determine the final standings! But you see some huge disparities between xG and league position at this stage of the season and it would be interesting to see which is closer to the “truth”, historically.
Other random thoughts that might be worth looking into.
What seasons/teams had less correlation and then why?
Would cumulatively points vs league average show something different when rank is determined by very minor differences. Especially in a season like the EPL this year with 3 points separating 3-10?
Questions I hadn’t thought of before reading this. Thanks for a great article!
Your comparison of different leagues made me wonder what leads to differences between them. Why, for example, do Serie A and the Bundesliga reach stability faster than Ligue 1? When you consider other leagues (as you suggested, and as I did), the differences become more marked.
For each of the top leagues in the main datasets on football-data.co.uk, the time-to-reach-80%-stability is in the 12-17 week range: for example, 12 weeks in Portugal and the Netherlands, 16 in Turkey, 17 in Scotland. But where lower-league data are available, stability takes much longer, especially when we compare within a country: 18 weeks vs 9 weeks for Bundesliga B vs A and for Serie B vs Serie A, 18 weeks vs 12 weeks for La Liga Segunda vs Premera, and 21 weeks vs 10 for Championship vs Premier League. (In Scotland it bobbles around more: for SPL/SC/Scottish L1/L2 it's 17/16/16/19 weeks; in England it goes 10/21/18/20/22 for PL to National League.)
In every country, lower-league football takes longer to stabilize than the corresponding top league. (Based on your 80% threshold; patterns seem broadly similar for 70% or 90%.) This gives us two questions: why is there variation across countries, and why is there variation between leagues in the same country (which always seems to follow the same pattern)?
Potential influences:
- level of variation in quality of teams within a league: leagues in which teams are similar may take longer to stabilize; top-tier leagues may have greater differences between the top mega-clubs and the bottom just-got-promoted teams. Could use estimates of team value from transfermarkt to estimate this.
- size of country; larger countries are likely to have more teams and ?more likelihood of mega-clubs emerging
- other things? Not sure. What other league-level differences, on which data are available, might influence time-to-stability?
I'll investigate this more as and when I get the time. Thanks for setting out your approach so clearly - I look forward to more!
Wow, thank you, Lang, for the kind words and for going the extra mile by testing the other leagues!
For me, the fact that league tables in lower tiers converge more slowly is simply because those leagues are more equal (teams-wise) and competitive — teams are generally closer to one another than in the upper tiers.
Nice article. It would be interesting to compare to points instead of position too. For example, this season the table is a lot tighter, the gap between 4th and 12th is only 3 points. So position at this point in this season would likely be less stable predictor of final positions
Great work, Martin. You see so many people just apply parametric stats to everything, so it’s nice that you illustrate the distinction.
Something I would find interesting, and possibly a future article idea for you, is to compare the actual table versus the xG table and see how they both correlate with the final table through each match week.
There are different ways of framing this, of course; the actual table does have the advantage that points accrued are used to determine the final standings! But you see some huge disparities between xG and league position at this stage of the season and it would be interesting to see which is closer to the “truth”, historically.
Thank you, John. Love the idea. Duly added to my to do list ✅
Other random thoughts that might be worth looking into.
What seasons/teams had less correlation and then why?
Would cumulatively points vs league average show something different when rank is determined by very minor differences. Especially in a season like the EPL this year with 3 points separating 3-10?
Questions I hadn’t thought of before reading this. Thanks for a great article!
Thank you for the suggestion, Chris. It turns out this remark about the points instead of ranks seems to be a popular follow-up :)
Love this!
This is great - nice work!
Your comparison of different leagues made me wonder what leads to differences between them. Why, for example, do Serie A and the Bundesliga reach stability faster than Ligue 1? When you consider other leagues (as you suggested, and as I did), the differences become more marked.
For each of the top leagues in the main datasets on football-data.co.uk, the time-to-reach-80%-stability is in the 12-17 week range: for example, 12 weeks in Portugal and the Netherlands, 16 in Turkey, 17 in Scotland. But where lower-league data are available, stability takes much longer, especially when we compare within a country: 18 weeks vs 9 weeks for Bundesliga B vs A and for Serie B vs Serie A, 18 weeks vs 12 weeks for La Liga Segunda vs Premera, and 21 weeks vs 10 for Championship vs Premier League. (In Scotland it bobbles around more: for SPL/SC/Scottish L1/L2 it's 17/16/16/19 weeks; in England it goes 10/21/18/20/22 for PL to National League.)
In every country, lower-league football takes longer to stabilize than the corresponding top league. (Based on your 80% threshold; patterns seem broadly similar for 70% or 90%.) This gives us two questions: why is there variation across countries, and why is there variation between leagues in the same country (which always seems to follow the same pattern)?
Potential influences:
- level of variation in quality of teams within a league: leagues in which teams are similar may take longer to stabilize; top-tier leagues may have greater differences between the top mega-clubs and the bottom just-got-promoted teams. Could use estimates of team value from transfermarkt to estimate this.
- size of country; larger countries are likely to have more teams and ?more likelihood of mega-clubs emerging
- other things? Not sure. What other league-level differences, on which data are available, might influence time-to-stability?
I'll investigate this more as and when I get the time. Thanks for setting out your approach so clearly - I look forward to more!
Wow, thank you, Lang, for the kind words and for going the extra mile by testing the other leagues!
For me, the fact that league tables in lower tiers converge more slowly is simply because those leagues are more equal (teams-wise) and competitive — teams are generally closer to one another than in the upper tiers.
Nice article. It would be interesting to compare to points instead of position too. For example, this season the table is a lot tighter, the gap between 4th and 12th is only 3 points. So position at this point in this season would likely be less stable predictor of final positions
Hi Ryan, a valid point indeed! Thank you for the suggestion.