Michigan Basketball's RPI Problem
Following Michigan's two big wins the last 5 days, I thought I would take look at our RPI as it stands today and how it will play out for the rest of the season. Fair or not**, the RPI is a big factor in determining seed lines. Brian and many others have been critical of Beilein's inability (or unwillingness) to game the RPI by playing #125-175 teams as opposed to #250+ teams.
RPI as of today: #21. Terrific, right?
In the next three weeks, Michigan will play #260 Detroit, #327 Jacksonville, and #347 Alabama A&M. Live-rpi.com has a neat little tool that allows you to simulate results to give you an estimate of RPI in the future. Let's assume that Michigan gets the three wins as anticipated.
RPI as of January 1: #52. And this is after three wins.
This demonstrates how much of a negative impact poor scheduling can have on your RPI. Most would agree that non-conference play (LSU game aside) has been a relative success this season. Michigan will likely be sitting at 12-3 (1-1) with multiple quality wins but will have the RPI profile of a bubble team. When Selection Sunday rolls around and Michigan comes in 2 seed lines lower than expected, this is probably your reason why.
**The RPI is just an incredibly stupid tool to measure the quality of teams, but the system is the system.
December 13th, 2017 at 11:07 PM ^
December 13th, 2017 at 11:11 PM ^
Yes but the selection committee still uses it, so wouldn't it be smart to base your scheduling around improving RPI?
December 14th, 2017 at 8:22 AM ^
or selection and there are plenty of examples that back that up. They know about cases like this too. They have used it for evaluating the quality of wins, but have been replacing it with KenPom and other better ratings.
Their newest formula for evaluation of wins treats beating teams ranked below #160 at home the same. It doesn't really matter whether it's #170 or #340. That RPI makes a big distinction is completely illogical as the expected win % for tournament teams would be very close.
December 14th, 2017 at 8:51 AM ^
...if instead of evaluating teams, the purpose of the RPI all along was to discourage the scheduling of noncompetitive games.
There had never been much evidence that they used it for seeding and selection once they got in the room, even before they acknowledged that they didn't.
December 14th, 2017 at 9:23 AM ^
was to seed and select teams. The tournament rules had recently changed to include seeding and multiple teams from a conference. The formula was simple by design, given that it was a first attempt. I think one of the main purposes was to favor major conference teams over independents with better win-lost records. Those independents had enjoyed a tremendous advantage under the old rules.
December 14th, 2017 at 11:08 AM ^
has become a regular exercise here. I'm surprised that more people don't assume that they try as hard as they can to maximize their attractiveness to the committee, scheduling sometimes years out as they do. Someone should at least sound Beilein out about how he goes at it; he's gotta have some bright people on the case, no?
What I am more concerned with--and maybe they take this into account, too--the fact that Beilein's squads really do seem to take a number of games to gel. How do you schedule around that?
December 14th, 2017 at 11:57 AM ^
Basketball schedules are typically set no more than one season in advance, particualrly for non-marquee matchups. In fact, Michigan will usually finalize its basketball schedule during the summer leading into the season.
KenPom even offers a service to interested coaches to suggest non-conference opponents that will help the RPI for the coming season, and it doesn't come out until his offseason updates are made.
So, yes, I think it's fair to say that Coach Beilein either does not understand the way to schedule in order to maximize RPI or doesn't care, presumably because some other factor is more important. If it's "find teams that play schemes we'll see during Big Ten play," then fine. If it's "this team wants $200K and this one wants $175K," well, that's just sad.
December 14th, 2017 at 12:36 PM ^
decided not to argue
December 14th, 2017 at 12:01 PM ^
Perhaps Beilein schedules a certain level of competetion during a certain time of season because he knows he can implement different aspects of the offense and defense without fear of an upset. And these additions are one reason why his teams gets better over the long haul.
December 13th, 2017 at 11:17 PM ^
December 14th, 2017 at 12:48 AM ^
"In its current formulation, the index comprises a team's winning percentage (25%), its opponents' winning percentage (50%), and the winning percentage of those opponents' opponents (25%)."
RPI is very noisy this early in the season. Its problem is that it doesn't really measure a team's ability. It's really easy to game the RPI, just beat teams who have high winning percentages. The big ten beats up on one another every year, which lowers the 25%, 50%, and 25%. Thus, in the non-conference portion, it's crucial to play teams who do well in their conference. That is, the teams that go 14-2 in their shitty conference and become 14+ seeds in March. Even better if those teams regularly schedule games against good teams (thus inflating that last 25%).
Beilein* is completely off his rocker when it comes to this. Great coach... but this isn't hard to understand, and he is doing his team a disservice.
*I always see him blamed for this, but I'm not completely sure that he gets to decide who we play, maybe it's the AD as well.
December 14th, 2017 at 6:32 AM ^
show this to Warde.
December 14th, 2017 at 9:41 AM ^
Yeah, I thought the AD's office have as much to say about scheduling as the head coaching staff. Or am I wrong?
December 14th, 2017 at 12:38 PM ^
Regardless of who makes the final decision, I think Beilein has to have the majority opinion. He should be able to stress the need to play RPI beneficial competition. The problem is that because we are such a slow starting team most seasons, I'm not confident our record would still be as good. The 2014-15 team was 7-5 going into conference play and that was with a healthy Caris Levert. We played a pretty tough non-conference schedule that year. Maybe Beilein knows his team so well and that's why he'd prefer the softer start to the season.
December 14th, 2017 at 8:45 AM ^
are based on misunderstanding how RPI gets used, not how it's calculated. It's helpful to play teams ranked around #125-160, but those are hard to predict and harder to schedule just at home.
Playing more road games is the easiest way to improve your tournament resume. That's not what Michigan wants to do. Their home and homes are against big name schools. Tournaments give more opportunities for quality wins. The rest of the home schedule is mostly going to be weaker teams and it really doesn't matter if those teams are in the bottom half or the bottom tenth.
December 15th, 2017 at 1:54 AM ^
than you might think. Don't schedule any SWAC or MEAC teams which are almost all always in the 300s. Schedule more MAC teams which are typically in the 130-250 range. We used to to do this. We used to play a lot of MAC teams. My guess is MAC teams command a bigger paycheck and that we're content to give five games to the lowest bidders.
December 13th, 2017 at 11:27 PM ^
December 13th, 2017 at 11:46 PM ^
December 14th, 2017 at 12:26 AM ^
December 14th, 2017 at 9:44 AM ^
When they schedule these games, does anyone really know that Nicholls State is going to ranked 150 spots ahead of U of D?
December 14th, 2017 at 10:28 AM ^
I don't know, to be honest. I would think you could somewhat accurately know which teams were expected to be good in each conference. You could also look at the end of season KenPom conference ratings and see that you should stay away from anyone in the MEAC or SWAC.
1 | Big 12 Conference | 19.81 |
2 | Atlantic Coast Conference | 16.02 |
3 | Big East Conference | 14.54 |
4 | Big Ten Conference | 13.66 |
5 | Southeastern Conference | 12.91 |
6 | Pac 12 Conference | 10.12 |
7 | American Athletic Conference | 5.9 |
8 | Atlantic 10 Conference | 3.95 |
9 | Mountain West Conference | 3.41 |
10 | Missouri Valley Conference | 2.08 |
11 | West Coast Conference | 0.81 |
12 | Colonial Athletic Association | 0.45 |
13 | Mid American Conference | -0.41 |
14 | Southern Conference | -0.61 |
15 | Sun Belt Conference | -1.68 |
16 | Summit League | -2.14 |
17 | Western Athletic Conference | -2.89 |
18 | Ivy League | -2.99 |
19 | Metro Atlantic Athletic Conference | -3.08 |
20 | Horizon League | -3.22 |
21 | Conference USA | -3.79 |
22 | Ohio Valley Conference | -4.64 |
23 | Patriot League | -4.79 |
24 | America East Conference | -6.63 |
25 | Atlantic Sun Conference | -6.78 |
26 | Big Sky Conference | -7.59 |
27 | Southland Conference | -8.34 |
28 | Big South Conference | -8.55 |
29 | Big West Conference | -9.04 |
30 | Northeast Conference | -10.49 |
31 | Mid-Eastern Athletic Conference | -17.47 |
32 | Southwestern Athletic Conference | -17.66 |
December 14th, 2017 at 12:04 PM ^
Ken Pomeroy does, or has an idea, anyway. He sells this information to interested coaches ahead of the season.
December 13th, 2017 at 11:28 PM ^
December 13th, 2017 at 11:38 PM ^
It obviously wasn't ideal to lose that game, but I think LSU is going to surprise some people. Their two losses are to the #18 and #52 kenpom teams immediately after losing their (probably) 2nd best player in Sampson (who's still out).
They just beat #44 Houston and should get Sampson back soon. Obviously it would have been ideal to win that game especially since they could be a bubble team, but I'd be surprised if that qualifies as a "bad loss" (especially a loss against RPI 100+) when its all said and done. Tremont Waters is the truth.
December 14th, 2017 at 12:07 AM ^
in the SEC and make a run at the tourney themselves. But that f'ing RPI again. They're 167th! Despite beating us and being 85th in kenpom. So they'll have to do some damage to get up into the RPI top 100 that the committee seems to use as an (arbitrary) demarcation between "good" and "bad" wins/losses.
And yeah, the win against Houston is nice, but it was by three at home so it didn't move the needle for kenpom when they have a 40 point loss to ND and ten point loss to a meh Marquette team. We'll see what they do. Sampson back and playing effective would help. But their defense is just...woof.
December 13th, 2017 at 11:41 PM ^
What's interesting is if you drop the next three games from the schedule and substitute -
- Albany (128)
- Navy (140)
- Toledo (150)
And you let the simulator pick the remainder of games based on odds of winning/losing, Michigan would end up with an RPI of 27 with a SOS at 30. That's a solid resume for a 5 or 6 seed.
December 13th, 2017 at 11:45 PM ^
Yep, exactly my point here. And the difference between the games on our schedule and the hypotheticals you gave is pretty small in terms of winning percentage. You're still going to be something like 90-95% favorites in those games compared to the 98-99% we'll have in our three body bag games coming up.
December 14th, 2017 at 12:22 AM ^
what this tool uses to project "expected" outcomes in future games? If it's using RPI, we know that to be not the best predictor of future results.
Also, I wonder if it projects individual game outcomes vs. expected outcomes over multiple games. If it's using individual game probabilities to project individual game outcomes, it will skew pretty heavily in favor of good teams, and that's why the projection will show a lot of teams passing us that close behind us.
For example, let's assume the 30th ranked team is favored in its next three games and is given a 70% chance to win each game. If the system is predicting they go 3-0, it's doing a poor job of projecting. In reality the expected number of wins is 0.7+0.7+0.7=2.1 wins. They're projected to win a full game more than realistically expected.
i.e. some of the those teams that have "gamed" the system by schedule some decent teams will lose to those teams even though the projections assume they'll win them.
December 14th, 2017 at 1:18 AM ^
You are assuming an inconsistent team (so far) will win all three of those games. What happens when you put an L next to Albany, for example?
December 14th, 2017 at 1:18 AM ^
You are assuming an inconsistent team (so far) will win all three of those games. What happens when you put an L next to Albany, for example?
December 14th, 2017 at 1:18 AM ^
You are assuming an inconsistent team (so far) will win all three of those games. What happens when you put an L next to Albany, for example?
December 14th, 2017 at 1:42 AM ^
going 2-1 (with loss against Albany) puts them at an RPI of 44.
The only thing dumber than the RPI is our stubborn refusal to game it.
December 14th, 2017 at 4:46 AM ^
December 13th, 2017 at 11:43 PM ^
December 13th, 2017 at 11:49 PM ^
It's not a bad thing to be sitting at #52 going into conference play, it's just something that could have easily been avoided. It's hurts our margin for error (if we need it) to get into the tournament, and hurts our seed line if were safely in.
December 14th, 2017 at 3:37 AM ^
See my example above - the result in swapping those three games is -
December 13th, 2017 at 11:45 PM ^
ever been asked for an explanation by the media?
December 14th, 2017 at 12:09 AM ^
December 14th, 2017 at 7:55 AM ^
December 14th, 2017 at 9:45 AM ^
Maybe. But a little stiffer competition still allows for that and it has the added advantage of elevating both individual and team play. And it would show who is ready and willing to be leaders on this team. I'm not sure you get that with cupcake scheduling.
December 15th, 2017 at 1:59 AM ^
Beilein says, schedule five likely wins per year, and Warde just takes the teams willing to come play for the least payout (SWAC teams, ugh).
December 14th, 2017 at 1:03 AM ^
Suppose Beilein is scared there is a real risk of losing to a mid-100s RPI team?
Our RPI drops to 52 after 3 wins against RPI 300 teams, whereas it would stay in the upper 20s if we beat 3 mid 100s RPI teams (as Megan pointed out).
But if we played 3 mid 100s RPI teams and lost one game (using same forecaster as Megan)...our RPI is still 44!!!
While this highlights how stupid the RPI ranking system is, it also shows that these scheduling decisions are terrible.
December 14th, 2017 at 12:53 AM ^
December 14th, 2017 at 3:50 AM ^
December 14th, 2017 at 8:18 AM ^
December 14th, 2017 at 3:55 AM ^
December 14th, 2017 at 6:54 AM ^
December 14th, 2017 at 11:17 AM ^
Our RPI does not matter. Gaming the system to raise it is pointless. The committee uses their own calculation methods to rank the teams.
The RPI gets talked about in the media a lot before the selections are announced. I guess that's a big reason people think it's so important. As you point out, there is always somebody with a high RPI that gets left out. There are also teams with high RPIs that get seeded low and vice versa.
December 14th, 2017 at 12:37 PM ^
To add into the discussion, here is what UM's KenPom rank was heading into the NCAA tournament going back to 2011 (first year he has current ranking on each game)
Year | KP Rank | KP Seed | Actual Seed |
2011 | 39 | 10 | 8 |
2012 | 25 | 6 | 4 |
2013 | 11 | 3 | 4 |
2014 | 16 | 4 | 2 |
2016 | 55 | 14 | 11 |
2017 | 20 | 5 | 7 |
If you add up the deltas you get a net of +6 which means on average, Michigan has gotten one seed better than KenPom would have predicted over the last six NCAA tournaments. Of course all caveats about this not being what KenPom is supposed to be apply, but it's another data point that maybe this doesn't matter as much as I thought it did.