It's never too early to make wrong predictions, so here goes:
Two days ago the X account Pythagoras in Boots posted a thread arguing with data that the best single predictor of how a team will do next year is the Points minus xPoints difference this year. It's better than xG, or xGF, Actual Points or Table Place. They looked at 5 years of Premier League results to reach this conclusion. Based on Pts-xPts, teams with large Plus or Minus variances tend to regress to the mean. You don't necessarily overshoot, but you tend to finish with a Point total closer to your xPoints in year 2 if the delta was very big in Year 1, and teams tend to drop or rise table places in the same direction.
They also noted that xPoint performance has smaller swings than Point swings year to year.
So I looked at MLS data from ASA from 2020-2025. For 2020 I prorated all figures to a 34 game season. My question was: how likely is a team that exceeded the median Pts-xPts amount in Year 1 to rise or fall in the table in Year 2?
First I calculated the medians. The median positive xPt differential in the data set was +5.87. The negative median was -6.75. I simplified and limited the data set on both sides to teams who had a Point total that differed from the xPoint total by more than 6 in either direction. That left 37 teams from 2020 to 2024 with a negative points variance greater than 6 and 32 with a positive variance greater than 6.
Of the 37 teams with a negative points variance of 6 or more, 23, or 62%, moved up in the table the following year.
Of the 32 teams with a positive points variance of 6 or more, 25, or 78%, dropped in the table the following year.
Then I checked the teams with differentials of less than 6 in either direction:
Of the 32 teams with a negative points variance of < 6, 17, or 53%, moved up in the table the following year.
Of the 38 teams with a positive points variance of < 6, 20, or 53%, dropped in the table the following year.
So, the pattern holds at even lower variances, though by much smaller percentages, as you would expect. If those percentages hold over greater time periods, I think the tendencies at variances greater than 6 are rather substantial.
With that noted, seven MLS teams in 2025 had a negative variance of Points minus xPoints of greater than 6.0:
Six MLS Teams had a positive variance of Points minus xPoints greater than 6.0 in 2025:
I did not distinguish between a small change in table status and larger moves. The 62% and 78% figures include moves of just 1 spot in the expected direction, but not teams that did not move up or down at all.
I did not do a calculated breakdown to fine tune the Pt-xPt variances further. Eyeballing the data, more extreme variances (e.g. +/- 10) do appear to be more predictive than ones between 6 and 10, but the 6 to 10 range still seems rather likely to predict a rise or drop. But yes, all things equal I believe Miami and NYC are less likely to fall in the table than Cincinnati or Charlotte, but more likely than clubs with a variance of <6. Also Miami seems to break all the models.
Note also that all table ranking data is based on the combined Supporters Shield table rather than conference tables.
Finally, this is not to claim that things like roster or coaching changes, injuries, player development and aging are irrelevant. But a predictive success rate of 62-78% seems rather meaningful before taking into account all of those confounding factors.
Two days ago the X account Pythagoras in Boots posted a thread arguing with data that the best single predictor of how a team will do next year is the Points minus xPoints difference this year. It's better than xG, or xGF, Actual Points or Table Place. They looked at 5 years of Premier League results to reach this conclusion. Based on Pts-xPts, teams with large Plus or Minus variances tend to regress to the mean. You don't necessarily overshoot, but you tend to finish with a Point total closer to your xPoints in year 2 if the delta was very big in Year 1, and teams tend to drop or rise table places in the same direction.
They also noted that xPoint performance has smaller swings than Point swings year to year.
So I looked at MLS data from ASA from 2020-2025. For 2020 I prorated all figures to a 34 game season. My question was: how likely is a team that exceeded the median Pts-xPts amount in Year 1 to rise or fall in the table in Year 2?
First I calculated the medians. The median positive xPt differential in the data set was +5.87. The negative median was -6.75. I simplified and limited the data set on both sides to teams who had a Point total that differed from the xPoint total by more than 6 in either direction. That left 37 teams from 2020 to 2024 with a negative points variance greater than 6 and 32 with a positive variance greater than 6.
Of the 37 teams with a negative points variance of 6 or more, 23, or 62%, moved up in the table the following year.
Of the 32 teams with a positive points variance of 6 or more, 25, or 78%, dropped in the table the following year.
Then I checked the teams with differentials of less than 6 in either direction:
Of the 32 teams with a negative points variance of < 6, 17, or 53%, moved up in the table the following year.
Of the 38 teams with a positive points variance of < 6, 20, or 53%, dropped in the table the following year.
So, the pattern holds at even lower variances, though by much smaller percentages, as you would expect. If those percentages hold over greater time periods, I think the tendencies at variances greater than 6 are rather substantial.
With that noted, seven MLS teams in 2025 had a negative variance of Points minus xPoints of greater than 6.0:
| Atlanta | -15.15 |
| DC United | -14.49 |
| Houston | -7.14 |
| LA Galaxy | -11.23 |
| Montreal | -11.24 |
| San Jose | -12.30 |
| St Louis | -10.99 |
Six MLS Teams had a positive variance of Points minus xPoints greater than 6.0 in 2025:
| Cincinnati | +22.52 |
| Charlotte | +15.98 |
| Miami | +7.40 |
| Minnesota | +9.29 |
| NYCFC | +7.46 |
| San Diego | +10.33 |
I did not distinguish between a small change in table status and larger moves. The 62% and 78% figures include moves of just 1 spot in the expected direction, but not teams that did not move up or down at all.
I did not do a calculated breakdown to fine tune the Pt-xPt variances further. Eyeballing the data, more extreme variances (e.g. +/- 10) do appear to be more predictive than ones between 6 and 10, but the 6 to 10 range still seems rather likely to predict a rise or drop. But yes, all things equal I believe Miami and NYC are less likely to fall in the table than Cincinnati or Charlotte, but more likely than clubs with a variance of <6. Also Miami seems to break all the models.
Note also that all table ranking data is based on the combined Supporters Shield table rather than conference tables.
Finally, this is not to claim that things like roster or coaching changes, injuries, player development and aging are irrelevant. But a predictive success rate of 62-78% seems rather meaningful before taking into account all of those confounding factors.
