Pomeroy on steroids?

[Bornorange]

Something obviously wrong with the man.
Wisconsin with 5 losses still the number two rated team by him. Syracuse hangs at number 5.
At what point does he ask himself, "Do I enjoy looking this ignorant?"
He holds onto a screwed up algorithm with as much determination as JB does to the Zone.
Difference of course is JB is succesful while Doctor Pomeroy refuses to admit a cancer in his numbers.
He fails to break into my top ten computer polls.

 

[OrlandoCuse]

Something obviously wrong with the man.
Wisconsin with 5 losses still the number two rated team by him. Syracuse hangs at number 5.
At what point does he ask himself, "Do I enjoy looking this ignorant?"
He holds onto a screwed up algorithm with as much determination as JB does to the Zone.
Difference of course is JB is succesful while Doctor Pomeroy refuses to admit a cancer in his numbers.
He fails to break into my top ten computer polls.

I don't have a clue as to how Pomeroy does his rankings, but I'm assuming that he places equal weight on both defensive and offensive statistics. That might be the flaw in his formula.

I'm not saying defensive stats aren't important, but they probably aren't as important as offensive stats. IMO, a team with a great offense and an average defense is typically better than a team with a great defense and an average offense. Look at the teams that typically win the national championship; they're doing it with great offense and average to good defense.

If he weighted the offensive statistics a little more than the defensive stats (like offense x's 1.2, defense x's 0.8), he'd probably get a more accurate model.

 

[Cusefannotindc]

I don't have a clue as to how Pomeroy does his rankings, but I'm assuming that he places equal weight on both defensive and offensive statistics. That might be the flaw in his formula.

I'm not saying defensive stats aren't important, but they probably aren't as important as offensive stats. IMO, a team with a great offense and an average defense is typically better than a team with a great defense and an average offense. Look at the teams that typically win the national championship; they're doing it with great offense and average to good defense.

If he weighted the offensive statistics a little more than the defensive stats (like offense x's 1.2, defense x's 0.8), he'd probably get a more accurate model.

The model is 1/(1+(Defensive PPP/Offensive PPP)^10.25 at its base, he weights based on recent results and strength of opponent. The formula is the same used for baseball and professional basketball to predict strength. Obviously with fewer teams and more games, there is better information on the teams playing and weighting is less important.

The formula rewards defense slightly more than offense, a one point increase in PPP on offense is worth about .0254 for your PYTH and a decrease of a point in PPP on defense is worth about .0257 for you PYTH.

 

[Cusefannotindc]

Something obviously wrong with the man.
Wisconsin with 5 losses still the number two rated team by him. Syracuse hangs at number 5.
At what point does he ask himself, "Do I enjoy looking this ignorant?"
He holds onto a screwed up algorithm with as much determination as JB does to the Zone.
Difference of course is JB is succesful while Doctor Pomeroy refuses to admit a cancer in his numbers.
He fails to break into my top ten computer polls.

Where is your ranking system? It is easy to point one mistake you don't like. He uses objective statistics based on offensive and defensive efficiency to rank teams, similar to what is done in sabermetrics. It doesn't include any individual biases like a voter ranking SU 4th because he hasn't seen them play enough.

 

[OrlandoCuse]

The model is 1/(1+(Defensive PPP/Offensive PPP)^10.25 at its base, he weights based on recent results and strength of opponent. The formula is the same used for baseball and professional basketball to predict strength. Obviously with fewer teams and more games, there is better information on the teams playing and weighting is less important.

The formula rewards defense slightly more than offense, a one point increase in PPP on offense is worth about .0254 for your PYTH and a decrease of a point in PPP on defense is worth about .0257 for you PYTH.

There's the flaw in his formula. While in a professional sport (like NBA and MLB), defense (which would include pitching in baseball), is probably more important, I do not think it has the same importance in college basketball.

He should tweak his formula to either give equal weight to offense and defense, or skew it to offense slightly.

You have to take everything Pomeroy does with a grain of salt, because he's really just a weatherman with a time consuming hobby.

 

[Knicks411]

Interesting idea, sorta like the home/road thing with RPI, though usually I feel like the champion is pretty awesome at both offense and defense.

But the formula is essentially the pythagorean formula, with adjustments for opponent strength, as well as home/road, which I believe does weight offense more heavily? (Not entirely sure, but it has offense in the numerator and denominator) I think you want to have some ratings that go against general consensus; what would the point be otherwise? (For example, he had Duke as the #1 team in the country in 2010. They won the title, that doesn't "prove" anything, of course, but Duke was considered by a lot of people to be overrated that year).

But I'm having a little trouble believing Wisconsin is the second best team in the country. The rating is basically being driven by their defense, they rank second in the country in adjusted defensive efficiency, 3 points per 100 better than #3. I think the issue is they have some unbelievable defensive performance against some really bad teams; they allowed .53 points per possession vs Kennesaw St (313th), .6 against Wofford (200th), .5 against Missouri Kansas City (280th), .67 vs Bradley (230th), .63 vs Savannah St (250th), and .66 vs Miss Valley State (263). That's over 1/3 of their schedule. Granted, there are opponent adjustments, but I guess in this case, they aren't working out.

I'm having trouble exactly recreating his win%, using the exponent he claims he is using, 11.5. It looks like he is using 10.25, instead of 11.5. Anyway, instead of having the #2 defense in the country, what if Wisconsin was #10? Then, their ranking would drop to #11, which some may still think is a little high, but seems more reasonable than #2.

I think Wisconsin's defense isn't the second best in the country. They've allowed 194 points in 187 possessions in Big 10 play this year, a far cry from their adjusted efficiency of .80 for the season. Have they been a little unlucky? Yeah, probably. They lost by 3 to UNC, they lost by 3 in OT to Mich State, and they haven't really won any close games. But I don't think they're #2 in the country, I think that defensive rating is being propped up by some really good efforts against terrible teams.

Edit: He's not reinventing the wheel with his formula; I'm sure there have been various tests done to determine the right exponent. I'd say the bigger issue wouldn't be the exponent so much as the adjustments to the efficiency numbers.

There's the flaw in his formula. While in a professional sport (like NBA and MLB), defense (which would include pitching in baseball), is probably more important, I do not think it has the same importance in college basketball.

Do you have any evidence for this? Not saying you are wrong, but people who use the pythagorean formula's in sports usually do a lot of research to determine the right exponent. I'm not saying they are always right, but I'm not sure offense is any more important than D. I'm not sure it isn't, either.

 

[OrlandoCuse]

Do you have any evidence for this? Not saying you are wrong, but people who use the pythagorean formula's in sports usually do a lot of research to determine the right exponent. I'm not saying they are always right, but I'm not sure offense is any more important than D. I'm not sure it isn't, either.

Just my opinion on the differences between college basketball and the pro sports where he derived his formula from. I'm just thinking back to the teams that typically win it all or make deep runs in tourney, and they are usually great on offense. A few teams in recent years have bucked that tradition, like Butler. Of course, they didn't win it all.

I just think taking a formula from one sport and applying it to another sport (without adjustments) can skew your results.

 

[Knicks411]

Just my opinion on the differences between college basketball and the pro sports where he derived his formula from. I'm just thinking back to the teams that typically win it all or make deep runs in tourney, and they are usually great on offense. A few teams in recent years have bucked that tradition, like Butler. Of course, they didn't win it all.

I just think taking a formula from one sport and applying it to another sport (without adjustments) can skew your results.

The formula is absolutely adjusted though. In baseball the exponent is 1.83. In the NBA it's like 14. Pomeroy is using 10.25, it looks like.

 

[Cusefannotindc]

Do you have any evidence for this? Not saying you are wrong, but people who use the pythagorean formula's in sports usually do a lot of research to determine the right exponent. I'm not saying they are always right, but I'm not sure offense is any more important than D. I'm not sure it isn't, either.

Sometimes people can be right by accident. I ran last years winning % against Pyth, Offensive Efficiency and Defensive Effeciency. Putting OE and DE in a regression, I got coefficients that would weight Offense at .518 and Defense at .488.

 

[OrlandoCuse]

The formula is absolutely adjusted though. In baseball the exponent is 1.83. In the NBA it's like 14. Pomeroy is using 10.25, it looks like.

As I said, I don't have his formulas, but in my opinion offense should outweigh defense in college basketball. Pomeroy seems to have adjusted it the other way around, based off of pro sports adjustments. He's obviously adjusted it down from the NBA, but maybe he hasn't gone far enough.

I'd have to see all of his formulas and data and really play with it to develop an educated opinion, though.

 

[OrlandoCuse]

Sometimes people can be right by accident. I ran last years winning % against Pyth, Offensive Efficiency and Defensive Effeciency. Putting OE and DE in a regression, I got coefficients that would weight Offense at .518 and Defense at .488.

Do you still have your regression model from last year? If so, do you have the P-values of your OE and DE variables? That will tell you more about their influence on the response, more so than the coefficients.

 

[Knicks411]

As I said, I don't have his formulas, but in my opinion offense should outweigh defense in college basketball. Pomeroy seems to have adjusted it the other way around, based off of pro sports adjustments. He's obviously adjusted it down from the NBA, but maybe he hasn't gone far enough.

I'd have to see all of his formulas and data and really play with it to develop an educated opinion, though.

And I'm not trying to say I know enough to say you are wrong. But the numbers definitely have been adjusted; whether or not in the most perfect way, I of course have no way of saying with any kind of accuracy.

To bring it back to Wisconsin, btw, I'm sure you could weight defense at 48% or whatever and they'd still be #2, it's not like it's going to change that much.

 

[OrlandoCuse]

And I'm not trying to say I know enough to say you are wrong. But the numbers definitely have been adjusted; whether or not in the most perfect way, I of course have no way of saying with any kind of accuracy.

To bring it back to Wisconsin, btw, I'm sure you could weight defense at 48% or whatever and they'd still be #2, it's not like it's going to change that much.

You'd be surprised at how small adjustments in a formula can changed the end result rankings. 0.48 to 0.52 is really a 4% spread, which is pretty significant in statistics (and mathematics).

 

[Cusefannotindc]

Do you still have your regression model from last year? If so, do you have the P-values of your OE and DE variables? That will tell you more about their influence on the response, more so than the coefficients.

Offense is more influential, the t-stat is 13.25 and defense's is -10.

 

[Cusefannotindc]

And I'm not trying to say I know enough to say you are wrong. But the numbers definitely have been adjusted; whether or not in the most perfect way, I of course have no way of saying with any kind of accuracy.

To bring it back to Wisconsin, btw, I'm sure you could weight defense at 48% or whatever and they'd still be #2, it's not like it's going to change that much.

If you applied the data from my regression, the rankings would look like this:


1 Ohio St.
2 Syracuse
3 Kentucky
4 Kansas
5 Wisconsin
6 North Carolina
7 Indiana
8 Missouri
9 Duke
10 Michigan St.

 

[Knicks411]

You'd be surprised at how small adjustments in a formula can changed the end result rankings. 0.48 to 0.52 is really a 4% spread, which is pretty significant in statistics (and mathematics).

Fair enough, and I was thinking of earlier, when Wisconsin had a much larger lead in the ratings.

I'm not even sure exactly how it would work in terms of this; just adjust their offensive efficiency upwards by 2%? Doesn't seem right.

Edit: I was originalyl going to say Wisconsin would stay in the top 5, but I decided to double down and say they'd stay at 2. Was probably pretty stupid.

 

[Cusefannotindc]

You have to take everything Pomeroy does with a grain of salt, because he's really just a weatherman with a time consuming hobby.

Wasn't Bill James working as a night security guard when he first stated publishing his abstract?

 

[Knicks411]

Btw, just noticed he has a blogpost about Wisconsin today. Haven't read it yet.

 

[OrlandoCuse]

Offense is more influential, the t-stat is 13.25 and defense's is -10.

Okay, so there you go, my assumption has some merit and it appears Pomeroy's formula is slightly flawed having a bias toward defense (as many suspected).

 

[Cusefannotindc]

Fair enough, and I was thinking of earlier, when Wisconsin had a much larger lead in the ratings.

I'm not even sure exactly how it would work in terms of this; just adjust their offensive efficiency upwards by 2%? Doesn't seem right.

Edit: I was originalyl going to say Wisconsin would stay in the top 5, but I decided to double down and say they'd stay at 2. Was probably pretty stupid.

.44+(OE*.011)+(DE*-.0103)

It comes out with significantly lower winning % than Pyth does.

 

[OrlandoCuse]

Wasn't Bill James working as a night security guard when he first stated publishing his abstract?

I think he did have an unconventional background.

My point about Pomeroy is that many people take his rankings as a "word of god" type of view. His background is basically in weather. While I'm sure atmospheric science does require a lot of analytical thinking, I'm not sure there is a tremendous correlation between the study of weather and the study of sports statistics. Weathermen, after all, are not held to a high level of accuracy. He started the whole thing as a hobby and it just kind of snowballed (no pun intended) into something bigger.

Maybe Pomeroy's rise to the top of college basketball statistics gives credence to the demand for good basketball statistics? Stat and math guys at the graduate level, at least the one's I've dealt with, typically aren't too into researching sports. Not much funding for it, so their research interests go elsewhere.

 

[moqui]

I don't have a clue as to how Pomeroy does his rankings, but I'm assuming that he places equal weight on both defensive and offensive statistics. That might be the flaw in his formula.

I'm not saying defensive stats aren't important, but they probably aren't as important as offensive stats. IMO, a team with a great offense and an average defense is typically better than a team with a great defense and an average offense. Look at the teams that typically win the national championship; they're doing it with great offense and average to good defense.

If he weighted the offensive statistics a little more than the defensive stats (like offense x's 1.2, defense x's 0.8), he'd probably get a more accurate model.

I think you are on the right track, SC. Not certain what the weighting should be, but there is no doubt that the ability to score is the most important attribute in basketball. In fact, there are two aspects to that, as well - I believe that the ability to efficiently score in half court sets is more important than the ability to score in transition, and that becomes more pronounced in tournament settings where a single loss is fatal, while games tend to become more and more of a half court contest (as you advance each round, you are more likely to face a quality opponent who will limit your transition opportunities). That is what limits Syracuse to the 3rd round except in those rare cases where it has an exceptional offensive player who is able to consistently get buckets or get to the line in half court situations.

 

[Knicks411]

I think those are fair points on Pomeroy; I visit the site probably every day, but I don't try and use them as the gospel, so to speak. Team A isn't better than Team B cause Team A is #4 and Team B is #5.

 

[NineOneSeven]

Where's fanfanclubclub when you need him for some linear regression and statistical models?

 

[Cusefannotindc]

I think you are on the right track, SC. Not certain what the weighting should be, but there is no doubt that the ability to score is the most important attribute in basketball. In fact, there are two aspects to that, as well - I believe that the ability to efficiently score in half court sets is more important than the ability to score in transition, and that becomes more pronounced in tournament settings where a single loss is fatal, while games tend to become more and more of a half court contest (as you advance each round, you are more likely to face a quality opponent who will limit your transition opportunities). That is what limits Syracuse to the 3rd round except in those rare cases where it has an exceptional offensive player who is able to consistently get buckets or get to the line in half court situations.

The problem is that that particular stat seems pretty difficult to measure. It can't be easily pulled out of publicly available box scores. I looked through play by play data the other day when you mentioned this and there wasn't an apparent way to tell the difference between half court and transition offense.