Does anyone here understand the pythagorean winning percentage?

[curious]

as discussed at this site: http://kenpom.com/stats.php ? I'm trying to figure out how to factor in the spread.

[MrX]

Pythagorean Wins is an attempt to predict winning percentage based on points scored vs points allowed. Done correctly, it has more predictive value than past winning percentage. That, of course, could be really handy for handicapping.

The Pythagorean wins on that site wont tell you anything directly about the percentage likelihood of covering a spread, but they could give you some information regarding which teams may be over/under valued by the betting market.

[Destroyer]

The Pythagorean Winning Percentage developed by Bill James is intended to calculate an estimate on a team's winning percentage based on points scored and points allowed. The Pythagorean Winning Percentage isn't intended to estimate the winning percentage against the spread. Hope this helps.

[curious]

The Pythagorean Winning Percentage developed by Bill James is intended to calculate an estimate on a team's winning percentage based on points scored and points allowed. The Pythagorean Winning Percentage isn't intended to estimate the winning percentage against the spread. Hope this helps.

Thanks. I have been using it to find mismatches. If I find two teams playing each other with a huge discrepency between PWP but the spread isn't too terribly large, I take the favorite.

Another thing I found that is working pretty good is to find two teams that are close in PWP but one of the teams is a medium size to big underdog according to the spread. Yesterday both Maryland and USC fit this description.

Today Penn State, Weber State and Oregon fit this description. Penn State already covered, Weber looks like they may cover, Oregon yet to play.

[Ganchrow]

I'd recommend an abundance of caution when using so-called "Pythagorean" expectation or winning percentage in the manner you've described.

A precondition for using a statistic predictively should be to be demonstrate that it can be useful descriptively. For NCAA basketball, especially in the early season, I'm not convinced that even so weak a requirement has ever been met.

Still, even were we to grant that PWP serves an adequate descriptor of expected prior win percentage conditioned on an observed numbers of points for and points against, we'd still need to account for strength of schedule. In NCAA BB, especially near the beginning of the conference season, PWPs will be based heavily on performance versus non-conference opponents, the relative strengths of which may or may not be comparable across teams.

It doesn't really do much for us quantitatively to be able to state, "Team X should have gone 11-4 against its last 15 conference and non-conference opponents but really went 9-6, while its opponent Team Y should have gone 9-6 against its last 15 conference and non-conference opponents but really went 11-4" when X and Y have few opponents in common. Now sure one might try to use PWP to qualitatively determine which teams might be under or overvalued by the market, but that's a long way from using PWP to create objective forecasts of future win probabilities.

[Wheell]

Ganchrow is correct. If you were able to give teams a long period of time their pythagorian wins would eventually come to equal their actual winning %, as well as their predictive winning %, but that period of time is far longer than any season I know of. See Diamondback, Arizona for a good example. KenPom is a great analytical tool for understanding past results, but... well, let's just say it had West Virginia as the best team in the country for quite a while this season. Use it with a bottle of salt.

[RickySteve]

I'd recommend an abundance of caution when using so-called "Pythagorean" expectation or winning percentage in the manner you've described.

A precondition for using a statistic predictively should be to be demonstrate that it can be useful descriptively. For NCAA basketball, especially in the early season, I'm not convinced that even so weak a requirement has ever been met.

Still, even were we to grant that PWP serves an adequate descriptor of expected prior win percentage conditioned on an observed numbers of points for and points against, we'd still need to account for strength of schedule. In NCAA BB, especially near the beginning of the conference season, PWPs will be based heavily on performance versus non-conference opponents, the relative strengths of which may or may not be comparable across teams.

It doesn't really do much for us quantitatively to be able to state, "Team X should have gone 11-4 against its last 15 conference and non-conference opponents but really went 9-6, while its opponent Team Y should have gone 9-6 against its last 15 conference and non-conference opponents but really went 11-4" when X and Y have few opponents in common. Now sure one might try to use PWP to qualitatively determine which teams might be under or overvalued by the market, but that's a long way from using PWP to create objective forecasts of future win probabilities.

The numbers he's using are SoS-adjusted.

[RickySteve]

Ganchrow is correct. If you were able to give teams a long period of time their pythagorian wins would eventually come to equal their actual winning %, as well as their predictive winning %, but that period of time is far longer than any season I know of. See Diamondback, Arizona for a good example. KenPom is a great analytical tool for understanding past results, but... well, let's just say it had West Virginia as the best team in the country for quite a while this season. Use it with a bottle of salt.

I love seeing the argument "Maybe it's true in the long run, but that doesn't mean anything now." It saves me a lot of time winnowing out the people I need to pay attention to. Tell me, are you the one keeping charts at the roulette table?

[curious]

I'd recommend an abundance of caution when using so-called "Pythagorean" expectation or winning percentage in the manner you've described.

A precondition for using a statistic predictively should be to be demonstrate that it can be useful descriptively. For NCAA basketball, especially in the early season, I'm not convinced that even so weak a requirement has ever been met.

Still, even were we to grant that PWP serves an adequate descriptor of expected prior win percentage conditioned on an observed numbers of points for and points against, we'd still need to account for strength of schedule. In NCAA BB, especially near the beginning of the conference season, PWPs will be based heavily on performance versus non-conference opponents, the relative strengths of which may or may not be comparable across teams.

It doesn't really do much for us quantitatively to be able to state, "Team X should have gone 11-4 against its last 15 conference and non-conference opponents but really went 9-6, while its opponent Team Y should have gone 9-6 against its last 15 conference and non-conference opponents but really went 11-4" when X and Y have few opponents in common. Now sure one might try to use PWP to qualitatively determine which teams might be under or overvalued by the market, but that's a long way from using PWP to create objective forecasts of future win probabilities.

the pythagorean formula I use doesn't use points scored for/against. It uses points per possession for/against. It also factors in strength of schedule. So far it is working out great. I'm not betting the farm on it, just doing an experiment.

Well, I think finding two teams with very similar win percentages but one of them is a huge dog is finding teams that are undervalued by the market.

[curious]

I love seeing the argument "Maybe it's true in the long run, but that doesn't mean anything now." It saves me a lot of time winnowing out the people I need to pay attention to. Tell me, are you the one keeping charts at the roulette table?

Well, I think as an analytical tool, taking points per possession for/against, strength of schedule and home/away to make a relative ranking is fairly solid. Using that as a starting point to look at a line to see if there is value in the line seems like a good approach to me.

Today I like San Diego, Syracuse, and St Peters. San Diego because its ranking is twice as high as Portland and the line is pickem, and Syracuse and St Peters because their rankings are similar to their opponents yet they are getting lots of points.

I would agree, trying to use any analytical tool blindly is stupid.