Ranking the Ranker

[Wrestleknownothing]

I was asked on 34 and Counting about how accurate the pre-season rankings are. Given that lineups are not yet known for every team until somewhere in the middle third of the season, injuries happen, and freshmen are generally not ranked in the pre-season (and then take some time to find their true ranking during the season), I do not know how worthwhile the exercise would be. And it would be a large amount of work, so I am not doing that.

Instead I will give myself a score.

In 2025 my expected points model had an r-squared of 0.956. Not bad. Because my numbers are based on past results they generally handle over and under performance well at summary levels (team scores, seed performance across all ten weights), but there is always a lot of volatility at the individual wrestler level.

The big outlier was the #5 seed which performed historically bad with only 3 AA's from the group of ten. You have to go back to 1980 to find the #5 seed perform at that level.

 

[AgSurfer]

Good work, WKN. Here's a suggestion, just to keep things interesting. Looking at your chart, you see pretty good correlation in the higher seeded wrestlers but much more scatter among the lower seeds. How about taking the same data and show it without the top 4 seeds and then again without the top 8 seeds and then without the top 12 seeds. Maybe I'm taking things too much to the nerd level, but it's interesting to see how the level of uncertainty increases as you get to lower and lower seeds. This might be a good supporting argument for guys like Ben Askren who think that seeding beyond a certain level is not a good idea since it's almost like just rolling dice.

 

[Col]

I was asked on 34 and Counting about how accurate the pre-season rankings are. Given that lineups are not yet known for every team until somewhere in the middle third of the season, injuries happen, and freshmen are generally not ranked in the pre-season (and then take some time to find their true ranking during the season), I do not know how worthwhile the exercise would be. And it would be a large amount of work, so I am not doing that.

Instead I will give myself a score.

In 2025 my expected points model had an r-squared of 0.956. Not bad. Because my numbers are based on past results they generally handle over and under performance well at summary levels (team scores, seed performance across all ten weights), but there is always a lot of volatility at the individual wrestler level.

The big outlier was the #5 seed which performed historically bad with only 3 AA's from the group of ten. You have to go back to 1980 to find the #5 seed perform at that level.

Interesting a linear function... I would of guessed an exponential one. I agree 5th seed is an outlier. Is there a data point in that group that you could ignore in order to bring the S squared into line (for those not following s2 is the variance from the expected standard deviation... the lower standard deviation the easier to use as a predictor while R2 measures how good the line fits the data), I add this to ensure that my analysis is not compared to a used diaper filled with Indian food but at least makes it to the turd covered in burnt hair level.

 

[Dogwelder]

Interesting a linear function... I would of guessed an exponential one. I agree 5th seed is an outlier. Is there a data point in that group that you could ignore in order to bring the S squared into line (for those not following s2 is the variance from the expected standard deviation... the lower standard deviation the easier to use as a predictor while R2 measures how good the line fits the data), I add this to ensure that my analysis is not compared to a use diaper filled with Indian but at least makes it to the turd covered in burnt hair level.

Atom: As we travel through the villi, the CP4 barium sonar wave will indicate any monoclinic phylosilicates in our vicinity.
Aquaman: English, man, English!
Atom: Uhh, this little doohickey beeps when the cooties are near.
Aquaman: Ahh, I... cooties, yes.

 

[Wrestleknownothing]

Good work, WKN. Here's a suggestion, just to keep things interesting. Looking at your chart, you see pretty good correlation in the higher seeded wrestlers but much more scatter among the lower seeds. How about taking the same data and show it without the top 4 seeds and then again without the top 8 seeds and then without the top 12 seeds. Maybe I'm taking things too much to the nerd level, but it's interesting to see how the level of uncertainty increases as you get to lower and lower seeds. This might be a good supporting argument for guys like Ben Askren who think that seeding beyond a certain level is not a good idea since it's almost like just rolling dice.

I will play around with that. But also there is a scaling issue that I will discuss ion the next post.

 

[Wrestleknownothing]

Interesting a linear function... I would of guessed an exponential one. I agree 5th seed is an outlier. Is there a data point in that group that you could ignore in order to bring the S squared into line (for those not following s2 is the variance from the expected standard deviation... the lower standard deviation the easier to use as a predictor while R2 measures how good the line fits the data), I add this to ensure that my analysis is not compared to a used diaper filled with Indian food but at least makes it to the turd covered in burnt hair level.

It appears linear because the variable expand at nearly the same rate causing the distance between points to grow on a straight line. If you graph these on a seed scale the points expand faster than the seed changes giving you the exponential function. So this is kind of, sort of, an exponential axis to make the line look straight.

 

[BWFight]

It appears linear because the variable expand at nearly the same rate causing the distance between points to grow on a straight line. If you graph these on a seed scale the points expand faster than the seed changes giving you the exponential function. So this is kind of, sort of, an exponential axis to make the line look straight.

Just as I suspected

 

[Tryingtodoitright]

And to realize that I got lost after 4th grade math...

 

[Aardvark86]

I love it when you guys talk dirty.

 

[Wrestleknownothing]

Good work, WKN. Here's a suggestion, just to keep things interesting. Looking at your chart, you see pretty good correlation in the higher seeded wrestlers but much more scatter among the lower seeds. How about taking the same data and show it without the top 4 seeds and then again without the top 8 seeds and then without the top 12 seeds. Maybe I'm taking things too much to the nerd level, but it's interesting to see how the level of uncertainty increases as you get to lower and lower seeds. This might be a good supporting argument for guys like Ben Askren who think that seeding beyond a certain level is not a good idea since it's almost like just rolling dice.

After playing around with this a bit some things occurred to me. First, while the r-squared is interesting, a better indicator of fit is the slope of the fitted line. Ideally it would be 1 if the fit was near perfect. The original line had a slope of 1.013, pretty good. But as you suggest the ends of the data do not look the same.

Breaking the seeds into 1 - 8, 9 - 16, and 17 - 33 produces slopes of 1.010, 1.083, and 0.853. It gets harder to predict as the seed goes lower. This certainly supports some of what Askren is saying. On the other hand the largest point deviation in the 17 - 33 group was only 1.18 points, so it is not as though there are large deviations. And the results in this one year do not look random at all (0.853 is still a pretty good fit), suggesting that the seeds were generally right rather than a crap shoot.

 

[Cstroke]

 

[BWFight]

After playing around with this a bit some things occurred to me. First, while the r-squared is interesting, a better indicator of fit is the slope of the fitted line. Ideally it would be 1 if the fit was near perfect. The original line had a slope of 1.013, pretty good. But as you suggest the ends of the data do not look the same.

Breaking the seeds into 1 - 8, 9 - 16, and 17 - 33 produces slopes of 1.010, 1.083, and 0.853. It gets harder to predict as the seed goes lower. This certainly supports some of what Askren is saying. On the other hand the largest point deviation in the 17 - 33 group was only 1.18 points, so it is not as though there are large deviations. And the results in this one year do not look random at all (0.853 is still a pretty good fit), suggesting that the seeds were generally right rather than a crap shoot.

 

[BWFight]

View attachment 991136

I can’t get enough of this stuff.