PWR modification for mid-season only

[Numbers]

Could the stats gurus and programmers help me with this:

Right now, when you look at the "Current pairwise" on this site (or any other that keeps that), the only thing that is considered is games already played. It is as if the remainder of the schedule did not exist. That's OK. But, i have been wondering if it is possible to do some thing like this:

RPI = 25% own win % + 54% opp win % + 21% opp's opp's win%. (Is that right?)

What if the last two pieces were calculated, including results from games already played, for teams remaining on a schedule?

For example, I am a Gopher fan. We have Notre Dame on our schedule next month. Win or lose, that game is going to help our strength of schedule. Can we add NoDame's present results (and all other games yet to be played as well)?

To give a better idea of the concept, right now RPI can be thought of as win% + Strength of Schedule. In the middle of the season, that only pertains to games already played.
What I want is: win% + Strength of entire schedule for the year. I think that would help remove some of the crazy fluctuations that seem to occur between now and the end of January.

Can anyone else comment or help? Thanks.

 

[Patman]

Re: PWR modification for mid-season only

The RPI only takes into consideration games played up to the point of creating the ranking... to that end there is no official RPI until the end of the season... same goes for the Pairwise... this is why, in a lot of ways, early season Pairwise rankings are somewhat garbage because it doesn't take into account TUC situations yet to occur and other games yet to be played.

I've, in the past, proposed some sort of simulator... but I judged that it was way too much work... the professional sports are easier because of the lack of tournaments with complex rules and because you only have a single set of tie-breakers to construct.

That being said, if you want to anticipate the SOS calculation for the entire season it may be reasonable to substitute in the schedule... the number you get won't really be that much better... and certainly can't account for various conference tournament opponents... but it'd give you a better feel for what may be going on.

Short answer... you can do anything you want... anticipating the rest of the schedule can help to give a better idea... but there is no good solid answer.

 

[FlagDUDE08]

Re: PWR modification for mid-season only

Could the stats gurus and programmers help me with this:

Right now, when you look at the "Current pairwise" on this site (or any other that keeps that), the only thing that is considered is games already played. It is as if the remainder of the schedule did not exist. That's OK. But, i have been wondering if it is possible to do some thing like this:

RPI = 25% own win % + 54% opp win % + 21% opp's opp's win%. (Is that right?)

What if the last two pieces were calculated, including results from games already played, for teams remaining on a schedule?

For example, I am a Gopher fan. We have Notre Dame on our schedule next month. Win or lose, that game is going to help our strength of schedule. Can we add NoDame's present results (and all other games yet to be played as well)?

To give a better idea of the concept, right now RPI can be thought of as win% + Strength of Schedule. In the middle of the season, that only pertains to games already played.
What I want is: win% + Strength of entire schedule for the year. I think that would help remove some of the crazy fluctuations that seem to occur between now and the end of January.

Can anyone else comment or help? Thanks.

Nice to know we can be thought of that way.

I believe it's 25% win + 21% opp + 54% opp-opp. Unless they changed it AGAIN...

 

[Craig P.]

Re: PWR modification for mid-season only

Could the stats gurus and programmers help me with this:

Right now, when you look at the "Current pairwise" on this site (or any other that keeps that), the only thing that is considered is games already played. It is as if the remainder of the schedule did not exist. That's OK. But, i have been wondering if it is possible to do some thing like this:

RPI = 25% own win % + 54% opp win % + 21% opp's opp's win%. (Is that right?)

What if the last two pieces were calculated, including results from games already played, for teams remaining on a schedule?

For example, I am a Gopher fan. We have Notre Dame on our schedule next month. Win or lose, that game is going to help our strength of schedule. Can we add NoDame's present results (and all other games yet to be played as well)?

To give a better idea of the concept, right now RPI can be thought of as win% + Strength of Schedule. In the middle of the season, that only pertains to games already played.
What I want is: win% + Strength of entire schedule for the year. I think that would help remove some of the crazy fluctuations that seem to occur between now and the end of January.

Can anyone else comment or help? Thanks.

I think your RPI modification isn't entirely unreasonable on the face of it, but I don't think it will really help much. There's still going to be a big "own record" influence, and the schedule strength part of it isn't going to fluctuate that much later in the season.

The main issues with PWR fluctuation have less to do with RPI and more to do with the TUCliff (tm).

 

[LynahFan]

Re: PWR modification for mid-season only

I think your RPI modification isn't entirely unreasonable on the face of it, but I don't think it will really help much.

I concur with this opinion. Two additional points:

You don't know who all of your remaining opponents prior to NCAA tournament selection will be. If you have a couple regular season tournaments plus a best 2 of 3 conference quarterfinal, then conference final 4, that's up to 7 games with an unknown opponent - that's ~18% of the games (7/39, including 5 potential conference playoff games) where you don't know the opponent yet. Secondly, the technique would also implicitly assume that each team will continue to maintain its current win % for the remainder of the season, which clearly never happens - there are always teams which go on fire after January or completely flame out. Given these two factors, I have a hard time believing that this modification would lead to a mid-season prediction that is any more accurate for the final standings than the current method provides.

 

[Patman]

Re: PWR modification for mid-season only

I concur with this opinion. Two additional points:

You don't know who all of your remaining opponents prior to NCAA tournament selection will be. If you have a couple regular season tournaments plus a best 2 of 3 conference quarterfinal, then conference final 4, that's up to 7 games with an unknown opponent - that's ~18% of the games (7/39, including 5 potential conference playoff games) where you don't know the opponent yet. Secondly, the technique would also implicitly assume that each team will continue to maintain its current win % for the remainder of the season, which clearly never happens - there are always teams which go on fire after January or completely flame out. Given these two factors, I have a hard time believing that this modification would lead to a mid-season prediction that is any more accurate for the final standings than the current method provides.

I think trying to predict the final RPI itself is rather foolhardy... really its not worth the time and effort expended into it. I mean, yes, you can... but no, you shouldn't. The most volitale component is winning percentage... this means that further refinements may not say much. It also doesn't take into account the removal of games and other assorted garbage.

From utility standpoint, however, its not a bad refinement. Conversely, there just isn't a particularly good one.

 

[FlagDUDE08]

Re: PWR modification for mid-season only

I think trying to predict the final RPI itself is rather foolhardy... really its not worth the time and effort expended into it. I mean, yes, you can... but no, you shouldn't. The most volitale component is winning percentage... this means that further refinements may not say much. It also doesn't take into account the removal of games and other assorted garbage.

From utility standpoint, however, its not a bad refinement. Conversely, there just isn't a particularly good one.

Absolutely correct. Even 2 weeks before the tournament started last year, there were a couple million permutations of what could happen with who makes it. Where's that news article on RHamilton's program to calculate that...