Great example, ku. I did not know that.

jskenner, don't forget as well that RPI gives equal weight to OOWP as the contribution from your record. That is, how many wins and losses your opponents' opponents have is equally important as your record.

Agree, hannibal. This (relative weighting) seems silly as well. Winning percentage of opponents' opponents is just as important as a team's actual winning percentage? Silly.

And I bet this opp-opp win % is very similar team to team, since it 2 layers down. After all, a team can choose what teams it plays, and can schedule good teams, but it has no control over what teams those opponents play, and there are so many (roughly 26 opponents x 24-25 games each), that it couldn't be that much different, even among teams who play a tough schedule vs those who play an easy schedule.

If HD is going to use such a formula, it might better be weighted as: win % = 50%, opp win % = 40%, opp-opp win % = 10%.

the funny thing is, in real life as well as here, rpi is kind of used as a "standard" to measure other things against. for instance, some folks will point to to situations where a high rpi team is not even in the top25 and say "see, the top25 is broken".

in real life (as well as here) the teams that fall on the wrong side of the bigdance-bubble that are thought to have the most legitimate gripe are usually those with the lowest rpi ("i had a 53rpi and got out in favor of a team with a 62rpi? thats bullcrap")

I even have to admit that i am among the folks (a vast majority, i think) who treat rpi as if it was a mathmatical "TRUTH" like E="mc2." It would be interesting to compare different models and discuss which types of teams (which type of "schedules" would probly be more accurate) each formula favors. And, by the way, if rpi is not a great measure of team success... then that suggests that one type of team (schedule) is consistently favored by it. Id be curious as to who you guys think is benefiting from this flawed formula and who is consistently underrated.



discuss.

i had never really given much thought to the actual formula, but now that i look more closely at it, Im kind of amazed that this formula works at all. It almost seems like you could go 0-26 and have a real good rpi if you could schedule tough enough opponents. and conversely, it would seem like teams could go 26-0 and have horrible rpis, but I rarely see that happen (unless it is warranted)

i was not aware of the mizzSt example either. ill have to look that up.



in any event... some interesting stuff. very interestting.

Quote: Originally Posted By oldave on 9/21/2009
i was not aware of the mizzSt example either. ill have to look that up.

It happened in 2006

Interesting perspective, oldave. As a realistic possibility (if someone could manage the schedule correctly), let's say Team A plays 10 non-con games against teams who eventually average .900 win percentage (after allowing for home/away adjustments). And let's say Team A is in an average conference where conference opponents average 5-5 in non-con games (they'll pretty much average .500 within games amongst themselves). Plus, let's assume opp opp win % averages .500. Then let's assume Team A goes 0-26.

RPI = A) 1/4 win % + B) 1/2 opp win % + C) 1/4 opp opp win %

Part A = 1/4 * 0 = 0

Part B = 1/2 * [(.900 * 10) + (.500 * 16)] / 26 = 1/2 * (9 + 8) / 26 = 1/2 * .654 = .327

Part C = 1/4 * .500 = .125

A + B + C = 0 + .327 + .125 = .452

That would be good for an RPI around 200 (for a team that did not win 1 game). Put that team in a great conference (with avg non-con win % as .800), and the RPI becomes .487 (good for about 150 RPI).

Reverse the exercise (Team A goes 26-0 with non-con opps winning 10% of games, etc.), and the RPI becomes .548 (in avg confenence, good for about 77 RPI) and .513 (in poor conference, good for about 120 RPI).

The break-even RPI (for 26-0 and 0-26 teams) would occur when the 26-0 team plays a schedule with an opp win % that is .500 less than the 0-26 opp win %. For example, 26-0 team's opponents win 25% and 0-26 teams' opponents win 75%. Yes, 25% is a lot easier schedule than 75%, but remember, this produces equal RPIs for a team that won all its games and a team that lost all its games.

Quote: Originally posted by oldave on 9/21/2009i had never really given much thought to the actual formula, but now that i look more closely at it,  Im kind of amazed that this formula works at all. It almost seems like you could go 0-26 and have a real good rpi if you could schedule tough enough opponents.  and conversely, it would seem like teams could go 26-0 and have horrible rpis, but I rarely see that happen  (unless it is warranted)

The 25% that your own record contributes is enough to prevent those extreme examples.

A more meaningful example might be to compare 2 bubble teams each with the same records in conference (against the same conference schedules), wtih differing non-con difficulties, where Team A wins a lot at home against a poor non-con schedule, and Team B loses a lot on the road against a good non-con schedule. I'll see what that might be.

maybe the reason for the emphasis on opponents record (in the rpi formula) is that in most realistic scenarios, you dont normally see the extreme differences (probably .200 is about as extreme as you would normally get? .600 vs .400? maybe a bit more)

whereas, you are much more likely to find extreme differences in winning %. (.500 is not uncommon at all)



there fore, the opponents winning percentage is not REALLY twice as important as your winning percentage, even though that would seem to be the case at first glance.

Quote: Originally posted by oldave on 9/21/2009maybe the reason for the emphasis on opponents record (in the rpi formula) is that in most realistic scenarios, you dont normally see the extreme differences (probably .200 is about as extreme as you would normally get?  .600 vs .400?  maybe a bit more)whereas, you are much more likely to find extreme differences in winning %.  (.500 is not uncommon at all) there fore,  the opponents winning percentage is not REALLY twice as important as your winning percentage, even though that would seem to be the case at first glance.

OK, I see where you are coming from. However, you overstate the case a little bit. Checking Naismith, .6000 vs .4000 for SOS seems about right. Since that is 75% of RPI, that gives effectively a range of about .1500. Then your own winning %, which can range 0-1.000 is 25% which gives an effective range of .2500. Using these numbers the effect of SOS is about 37.5% of RPI.

I need to digest it a little more to see if I totally agree, but I see the argument.

You make a solid point, oldave. However, for me it comes down to a basic fact. In the comparison between A) beat a team with a poor record and B) lose to a team with a great record, RPI values B more. This is well-meaning (reward those who schedule tough), but ultimately and absolutely wrong. ANY team can lose to a good team (all it takes is scheduling the game), but it actually means something to WIN, no matter how bad the opponent. Any mathematical system used (almost) as a litmus test as a NT selection/seeding tool, needs to be more in line with the basic truth that it means more to beat a bad team than lose to a good one.

can't it equal the same thing? why is it necessarily better to beat a bad team than lose to a good one? how does either one indicate anything good about a team?

i also think you overstate the case that nobody plays good teams at home anymore-- i'm in several top-notch d2 and d3 conferences, and have lots of home-and-home series going with other good teams. so do lots of other coaches in these leagues...

there are definitely those who only schedule road games, but as someone pointed out, it's a lot more fun to play good teams and coaches wherever than to play bad teams on the road. build a good enough team, win enough games to get into the NT.

For sh*ts and giggles lets do an example with a different formula.

Say 50%/25%/25%



Team A - goes 26-0. Plays OOC competition that ends up 5-21 at season's end. And all his conference mates go 0-10 in OOC. Basically he has the easiest schedule ever.

"Assuming" a super-low OOWP of .250 Team A's end of regular season RPI is .6239
Win% (26-0) = 1.000/2 = .5000
OWP (126-510) = 126/676 = .2456/4 = .0614
OOWP (25%) = .250/4 = .0625
Thats an RPI of .6239

Team B - goes 16-10 (8-8 in conf). Plays nothing but NT caliber teams. All their OOC opponents ends up 17-9. and their conference mates all went 6-4 in OOC. a super-tough schedule for Team B.

Again, "assuming" a tougher OOWP of .600 Team B's end of regular season RPI is .6034
Win% (16-10) = .6154/2 = .3077
OWP (394-282) = 394/676 = .5828/4 = .1457
OOWP (60%) = .6000/4 = .1500
Thats an RPI of .6034

***This doesnt take into account home and away***

But as you can see. Team A is an NT lock and they played nothing but the worst 100 teams in the league. Team B is a bubble team and they went 16-10 against all top 100 teams.

Basically, it SEEMS like OWP being 50% is skewing the final RPI number. But you guys dont factor in the fact that conference play levels the OWP playing field. ALL conferences (as a whole) go .5000 during conference play. thats means everybody's OWP AND OOWP is closing in on .500 the more games you play. Doesnt matter if you're well above .500 or well below .500 in SOS. As the conference season wears on, you will get closer and closer to .500

The only real issue, as someone mentioned before, is about the .6 and 1.4 weighting at DIII. it should be changed. I like it, but its not really fair. Other than that the system is fine and any adjustment to it would created lots of options to "game" the system