Are Games Between Near Equal Clubs More Likely To Produce Draws?

The form momentum charts I have been using for MLS can be used to study this, as the resistance is a calculation of position differential as if the teams were in a single table. If you'd like me to share any raw data with you, I'd be more than happy to.

 

Great question. I decided to try to figure that out myself, but I used a slightly different method than you.

I used the number of points at the end of the season for each team as to calculate the difference in quality as the absolute value of the home team's points - the away teams points. Then for each match, a win by either team was coded as 0 and a draw as 1.

I then took two approaches to answering the question. First, I just did a simple t-test to compare the means of point differences between games that ended in draws and games that didn't. The mean for draws was 17, the mean for wins was 22, and the difference was significant (p = .002). So that's one piece of evidence that says draws are more likely between teams closer in points.

Then I fit a logit model to see the probability of a win at each point difference, and the result is even more conclusive. The probability of a draw between teams with equal points is .36, the probability for teams with a 58 point difference (the max in the 2009-10 epl season if you don't count Portsmouth's 9 point penalty) is .11. The coefficient is significant, and the relationship between points and probability of a draw is linear, and actually slightly curvilinear. I don't think I can post pictures otherwise I'd post the graph. Instead I'll just approximate what it looks like:


Prob .36 | *
| *
| *
| *
| *
.11 |_______________*
Point dif. 0 58


This may not exactly be the same question you posed, which might have had more to do with probability of a tie given current form or relative standing compared to their final standing. That is obviously a little more complicated because teams start from the same place, so games at the beginning of the year all appear to be between equal teams. I'll be interested to see what you come up with using ranks.

 

Great question. I decided to try to figure that out myself, but I used a slightly different method than you.

I used the number of points at the end of the season for each team to calculate the difference in quality as the absolute value of the home team's points - the away teams points. Then for each match, a win by either team was coded as 0 and a draw as 1.

I then took two approaches to answering the question. First, I just did a simple t-test to compare the means of point differences between games that ended in draws and games that didn't. The mean for draws was 17, the mean for wins was 22, and the difference was significant (p = .002). So that's one piece of evidence that says draws are more likely between teams closer in points.

Then I fit a logit model to see the probability of a win at each point difference, and the result is even more conclusive. The probability of a draw between teams with equal points is .36, the probability for teams with a 58 point difference (the max in the 2009-10 epl season if you don't count Portsmouth's 9 point penalty) is .11. The coefficient is significant, and the relationship between points and probability of a draw is linear, and actually slightly curvilinear. I don't think I can post pictures otherwise I'd post the graph. Instead I'll just approximate what it looks like:


Prob .36 | *
| *
| *
| *
| *
.11 |_______________*
Point dif. 0 58


This may not exactly be the same question you posed, which might have had more to do with probability of a tie given current form or relative standing compared to their final standing. That is obviously a little more complicated because teams start from the same place, so games at the beginning of the year all appear to be between equal teams. I'll be interested to see what you come up with using ranks.


Edit- it seems bigsoccer doesn't like the whitespace in my graph, so it's not really helpful. But the basic idea can be inferred that its more or less a straight line from the top left corner to the bottom right, with probability as the y-axis and point difference as the x-axis.

 

Is this based on their rank at the end of the season, or at the time the game was played? If it's the time the game was played, how did you determine the rankings for the first week of the season?

 

This seems to me closer to my prior than sokol's curvilinear relationship. I expect a bit of an inverted U-curve. If you're VERY close you might still play an open game to go for the win. If you're close but a bit below, you'd be happier with a draw and you have the quality to push the percentage up. If you're far, you'd be happy with a draw but don't have the quality to get a high enough percentage.

It would be nice to see the win percentages for those groups.

 

Not really, its more dependant on form and style of play.

 

I remember looking into something similar many years ago, but based on which quarter of the league table each team was in, when the games were played.

While hardly a conclusive study, it did show that draws were most frequent when a team played at home to a team in the quarter below them - in other words the common belief that home advantage leveling things out and making draws more likely was wrong.