BIG EAST 2016

Chris...what's the difference major difference between your simulation and the BR


Sent from my iPhone using Tapatalk

 

Unadjusted RPI = { (own W/L) + 2(opponent's W/L, excluding you) + (opponents' opponents' W/L) } /4.
Adjusted RPI adds bonus/penalty points, for top-XX wins or bottom-YY losses.
You can find this data in excruciating detail at AWK's 2016 College Women's Soccer Schedule.

(0.500 + 2(0.679) +0.540)/4 = 0.59950 +0.0029 = 0.6026 #44 Marquette
(0.688 + 2(0.576) +0.557)/4 = 0.59925 +0.0022 = 0.6013 #45 Washington State ... oh those micropoints
(0.375 +2(0.634) +0.472)/4 = 0.5287 +0.0000 = 0.5287 #118 DePaul
(0.722 +2(0.416) +0.538)/4 = 0.5233 +0.0004 = 0.5237 #124 St. John's (NY)

This early in the season, a team's W/L (1st term) is mostly getting dominated by opponent's W/L (because of the *2 coefficient for the 2nd term). Marquette's own W/L is only -0.222 behind St. John's, but their SOS is way higher, at +2(0.283) = +0.566, which completely trumps it. Then Marquette earned more bonus points (else my Cougs woulda jumped them ). Similarly, DePaul's relatively ghastly -0.347 in own W/L vs. St. John's is trumped by their +2(0.218) = +0.436 for stronger opponents.

RPI SOS (2nd and 3rd terms) ensures that it is better to schedule good teams, even if you lose to them.
RPI bonus points ensure that it is better to visit good teams and beat them at their place

 

Early in the season, yes, Elements 2 and 3 are overweighted. But they converge by the end of the season, to roughly these weights (see cpthomas's RPI: Formula page):

0.50 Element 1: your W/L
0.40 Element 2: opponents' W/L (excluding you)
0.10 Element 3: grand-opponents' W/L

It's because most teams play most of their matches in-conference, and each conference is zero-sum. The net result is that Element 2's range of variation is narrower than Element 1, and Element 3 is even narrower. So their ability to swing the weighted average decays over time. By the end, each team's own performance is, indeed, the greatest factor in its own RPI.

Conversely, during non-conference pre-season, all teams independently hunt for opponents, so they can have disjoint sets of opponents (and even opponents' opponents). So Elements 2 and 3 are still all over the place in Week 3.

ARPI isn't perfect, but it's a reasonably good input to human (selection committee) judgment. It does do pretty well at isolating the ~20 or so teams in "the bubble". Thereafter, we come here to vent

 

I suggest you take a look here: Bennett Ranking Methodology

The system I'm using is tied to the RPI, since that's what the NCAA uses and since my bracket formation system is RPI based since that's what the Women's Soccer Committee uses. This creates a number of key differences, a key one of which is that the RPI's purpose only is to assign a value to teams' results over the course of the year, whereas a main purpose of BR is to assign values based on which one can predict the outcomes of next week's games. In a sense, the RPI is not intended to tell you who the best teams are but only which teams have performed the best. BR is intended to tell you, at each stage of the season, who the best teams are at that time.

Here are some significant differences:

Bennett assigns a high value to game scores/margins of victory. The RPI, in conformity with NCAA policy, does not consider game scores/margins of victory at all. Rather, the RPI considers only whether the game was a win, loss, or tie.

The simulation system I'm using, for games not yet played, uses ratings that are fixed at the beginning of the season and do not change. BR does not simulate the results of future games, although it predicts the results of next week's games, but to the extent one could consider it as simulating the results of next week's games, the basis for its simulation changes from week to week as game results come in. Since my system substitutes actual results for simulated results as games are played, over the course of the season the effect of my using ratings fixed at the beginning of the season will wash out.

BR has home field advantage built into its system as part of its main body. The RPI itself doesn't consider RPI as part of its main body but rather only as a minor factor in its good result/poor result bonus/penalty system. My simulation system, however, does include HFA as part of its main body though in a different way than BR.

BR, being a predictive system, during roughly the last 2/3 of the season, weights recent results more heavily than earlier results. The RPI, and my system, treats each game as having the same weight regardless of when it is played.
​

A main purpose of my system is to allow a coach to look at simulated results for the entire season and see, if they achieve the simulated results for their team, where they are likely to end up in the ratings/rankings and what their NCAA Tournament prospects are. If their prospects don't meet their objectives, then they can look to see where they need to achieve better results than what the simulation says. And, if they do more poorly than what the simulation says for a game, they can see that they will need to do better than the simulation says in another game. It's not perfect as a tool -- and indeed, there is no perfect tool to achieve this purpose -- but it may give a reasonable picture.

Since this is the first year doing this, we'll have to see how it worked once the season is over. One thing I'm seeing personally is that it helps one keep a focus on the entire season, counterbalancing a tendency to treat the results already achieved as defining the entire season. This works in both directions -- results already achieved can cause one to think the season has been great or awful. Looking at the updated simulation results, however, may say that if the rest of the season's results fit the simulated results the season won't be that great or that awful. In a sense, what this means is that the simulation treats actual results that are contrary to the initially simulated results as upsets. This, of course, isn't necessarily the case, they may be indications that the initially simulated ratings are wrong, which certainly is likely to be true in some cases. One of the purposes of the table I've published showing the changes in weekly simulated rankings is to allow an identification of those teams whose initially simulated ratings probably were wrong -- Arkansas, for example?

Hope this helps.

​

 

I hate to say it, but my simulation has only Georgetown (with a #2 seed) and Marquette in the NCAA Tournament. Go here and to the post at that blog that precedes the linked post. Of course, things can change over the next week, to some extent.

 

I have to agree with cpthomas. Two teams in - three (maybe) if DePaul gets the automatic bid. Butler, Providence or St. Johns have to win the conference to get in, which I don't think will be happening this year. I personally am pulling for DePaul; I like the way they pass, move and attack relentlessly. A fun group to watch play!

 

The bonus and penalty adjustments apply only to non-conference games. St. Johns has no bonuses from non-conference games, which is why its RPI and ARPI are the same.

I don't know what you're looking at in relation to the St. Johns v Seton Hall game, but the NCAA has the correct result in its RPI data base.

Sorry, but it looks like Providence and St. Johns both will be too far down in the ranks to get at large selections ... assuming St. Johns does not survive its next game.

 

I'll only make a slightly educated guess: DePaul, probably, even a tie (losing on PKs) might suffice since it counts as a tie and they'd amass a lot of credit due to Georgetown's high ranking; St. Johns, more likely they'd have to win their next game and at least tie in the finals.

 

Not sure how a team with an RPI of 60 makes the tourney...

Big east gets 2 and that's it IMHO.

 

DePaul's final ARPI rank is 54. But it's a long shot to get into the tournament. You're probably right with just two: Georgetown and Marquette.