RPI & Strength of Schedule...Huh?
I'm sure this makes sense to someone, but after just playing 4 top-20 teams (3 top 10) within 10 days, our strength of schedule is #25 in the country. Who has played a "tougher" schedule so far (with SOS rank in parentheses):
-New Mexico (#4): played 0 (currently) ranked teams
-Miami (#1): played 4 ranked teams
-Oklahoma (#7); played 5 ranked teams
-Arizona (#10): played 2 ranked teams
-Florida (#16): played 5 ranked teams
In short, WTF? In a world where NMU is considered to have the 4th toughest schedule and #3 RPI, I wonder what the point is of any of these numbers. Of course, they make for conversation, but clearly something is massively out of whack. Maybe this is a reason that the B1G fairs not as well as expected come tourney time. Thoughts? Am I just looking at these ranking incorrectly?
February 15th, 2013 at 4:53 PM ^
What was our SOS before the last 4 games? I believe 25th is the SOS of the whole season.
February 15th, 2013 at 4:57 PM ^
February 15th, 2013 at 4:59 PM ^
despite only getting 1 highly-ranked team at home so far.
February 16th, 2013 at 11:05 AM ^
Strength of "schedule" would seem to deal with the entire schedule, not just wins.
February 15th, 2013 at 4:58 PM ^
You're right on point. The RPI is unbelievably simple and stupid.
It often leads to unusual rankings, and I've never seen a good argument for why this kind of ranking is reasonable. There are a lot of good articles on this, and here's one of them:
http://sportsillustrated.cnn.com/2012/writers/luke_winn/09/27/schedule-…
February 15th, 2013 at 4:58 PM ^
news at 11. Kenpom has our SOS at 8.
February 15th, 2013 at 6:46 PM ^
I would be interested, since I am not familiar with the method.
February 15th, 2013 at 6:50 PM ^
"Strength of schedule - The components here are similar to the team components. Because we’re measuring a team’s offensive and defensive ability, we can also assess its schedule in similar terms. The Pyth is derived from these components in the same way that is for a team rating. (This is also I compute conference ratings.)"
http://kenpom.com/blog/index.php/weblog/entry/ratings_glossary
February 15th, 2013 at 6:59 PM ^
thanks.
February 15th, 2013 at 4:58 PM ^
but I'd guess IUPUI, Binghamton, and Cleveland State are weighing down our SOS like an anchor made of anchors
February 15th, 2013 at 6:18 PM ^
The bad teams are really weighing down our RPI. And some really good teams are undervalued by RPI (Indiana, Pittsburgh).
February 15th, 2013 at 5:27 PM ^
Sure UM just got done with a killer stretch but they've also played some of the worst teams in D1. Binghampton has an rpi of 336 (out of 347 teams). IUPUI has an rpi of 308. CMU is 249 and EMU is 235. Like someone mentioned those 4 teams are weighing down UM like a bunch of anchors.
Brian's mentioned this the past few years - sure you want a few cupcake games but you need to make sure those cupcakes are going to be somewhat decent. You throw a couple teams with rpi's above 300 on your schedule and it's going to hurt your SOS.
February 15th, 2013 at 5:28 PM ^
RPI calculates Stregnth of schedule as simply your opponents Winning % weighted at 2/3 plus your oppenents opponents winning % weighted at 1/3. It doesn't claim to be anything more than that and it's not a bad approximation as long as you understand it for what it is.
New Mexico might be surprising up there, but they've played only 5 of their 25 games against teams sub 150 of RPI. They play in a league that's pretty good this year Moutain West, and scheduled Uconn, Cincinnati, USC, Indiana St., Valporaiso, and Davidson among their non-conference games, all pretty good teams.
Michigan had only 6 sub 150 teams, but 2 of those are among the worst teams in college baseketball and play in AWFUL conferences.
It's really nothing to get worked up about.
February 15th, 2013 at 5:49 PM ^
I believe it's the team's own winning percentage at 25%
Opponents' winning percentage at 50%
and opponents' opponents' winning percentage at 25%
Still an unbelievably simple and unreliable metric.
February 15th, 2013 at 6:15 PM ^
Why would a team's own winning percentage be a factor in calculating SoS? That would punish teams for winning their games.
February 15th, 2013 at 6:55 PM ^
then, as you both correctly note, the method would make even less sense. But the SI article linked in a post below says the RPI uses 2/3 of opponents' WL and 1/3 of opponents' opponents WL.
February 15th, 2013 at 7:00 PM ^
The confusion is that you are talking about 2 different things.
The RPI rating is WP x 25%, OWP x 50%, OOWP x 25%. The RPI does not rate strength of schedule, the RPI is the overall rating.
The "strength of schedule" component of the RPI is what people call everything except the team's winning percentage in the RPI formula: OWP x 2/3 and OOWP x 1/3.
(Note: WP = Winning Percentage, OWP = Opponents' Winning Percentage, OOWP = Opponents' Opponents' Winning Percentage).
February 15th, 2013 at 7:48 PM ^
I had just realized this source of confusion. As I note above, this formula essentially ranks you according to a measure related to your WL record (win pct or wp *.25) and a measure related to your SOS (.5*OWP + >25*OOWP). So it is indeed like buying ONE pound of balogna at $5 per pound and paying 1+5=$6 for it rather than 1*5=$5.
To me it seems far more reasonable to rank teams by multiplying the SOS by a team's win pct. In fact, you might weigh each "win" individually according to its quality (perhaps related to the WL pct of an opponent * times its own SOS (eg the win pct of its own opponents). You could then add up these "quality-weighted" wins.
(you may also want to consider other factors as well like whether the win was at home or on the road and the win margin)
February 15th, 2013 at 7:08 PM ^
It could drop UM's seeding.
Although the selection committee supposedly discounts the RPI "when appropriate", I doubt that the committee knows very well when or why the RPI is providing inappropriate guidance.
Wasn't the head of that committee very recently none other than Gene Smith, the Ohio AD?
His past judgments in his post as AD do not inspire much confidence.
.
February 15th, 2013 at 5:37 PM ^
Ranking teams according to the RPI is like buying baloney at a very bizarre store. You tell them you want 1lb at $5 per pound, and the clerk tells you that you should pay 1+5= $6 rather than 1 times 5 = $5.
Similarly, to determine the rank of each team you play (and therefore your own SOS), the RPI adds rather than multiples a teams WL record by its SOS.* Thus, if you play four teams ranked 1-4 and then one ranked 290, your SOS will be an average of those--which is a whopping #60. So, if you go 5-0, your rank will be the same as if you played teams ranked 58-62 on five successive nights.
A more sensible approach taken by computer polls is to multiply the ranks--just like you would multiply the number of pounds by the price per pound of bolgna. (which in the case of multiple games would be like computing their geometric mean). That would give you a much more reasonable average opponent rank of around #6. That's more like the method of other computer polls, which rank UM's SOS much more highly .
*as noted below, it's some kind of weighted average---if I read the post below correctly, it weights very good or bad teams differently but then still adds the ranks. The weighting will not adequately correct the error if you play a few really bad teams. .
February 15th, 2013 at 5:40 PM ^
RPI doesn't do anything like that at all. It doesn't average ranks. If someone like ESPN is figuring SOS that way, that's their own prerogative, but that's not how RPI does it. It doesn't at all take into account a team's ranking because then it would be recursive, which isn't how RPI works. SOS is just your opponents' W/L records (figured RPI-style with the .6-1-1.4 formula), nothing more.
February 15th, 2013 at 6:40 PM ^
What you describe is not an iterative method, like many other polls use. But I think my main point still applies (perhaps with less extreme results).
if you combine the WL records of multiple teams, you are essentially adding or averaging a measure of their ranks---even if it is not a very good measure. Although the RPI makes an additional correction*, let me first illustrate the problem without this correction--considering only the opponents' WL records.
Suppose that you played four teams with 20-0 records, then a team that was 0-20, your opponents would be 80-20. But would that be the same as playing vs five teams with the same records of 16-4 (totalling 80-20 also)? The latter would be like playing five decent B1G teams. By contrast, the former would be like playing the first four games vs "final four" opponents in the tourney--in fact, if the teams were from the four best conferences, it would be a killer schedule---like playing the four best teams ever to play in the tourney.
So, I still think the RPI is not a very good method. It's full of baloney--even if it tries to disguise the taste (as noted below).
*Even if the RPI also uses some kind of ad hoc weights not just for the opponents but also for the opponents' opponents WL records, you are still computing a weighted average. So, you may think you are "correcting" the errors but you may often over or under-correct. Playing vs a 20-0 team in a marginal conference, whose opponents also usually play marginal teams, give nearly the credit as playing vs a 20-0 team in the B1G.
February 15th, 2013 at 8:36 PM ^
Well, I think, for one, that the difference between a 16-4 team and a 20-0 team is not as great as you portray - truth is, the B1G teams with four losses are the cream of the B1G crop, not merely decent. So if I get you right you're saying that four 20-0 teams is a killer schedule while five 16-4 teams is not, but I don't agree with the assessment.....so I would argue from that, that the RPI isn't doing as bad a job as you say.
I also disagree that playing 20-0 teams in bad conferences is similar to playing 20-0 teams in good ones. Let's say I played and beat Team A and you played and beat Team B. Both have exactly the same records. For the sake of coming up with numbers relatable to real-world numbers, let's say Team A and Team B are 10-10. It doesn't change the effect. Team A is in a good conference and their opponents are .667. Team B is in a lousy one and their opponents are .333. My one-game RPI is .6667 - yours is .5833. That's the difference between 4th and 54th. So in other words, if we each play 20 games against teams similar to A and B and have the same record, I will end up with a far better RPI than you will, as it should be.
As Alton said below, RPI is not perfect and it spits out a ballpark, not a perfect ranking. That's true of everything from KenPom to Sagarin to the AP poll. It's a tool that the committee uses to start and lead the discussion, not make a concrete seeding out of. (And fortunately so.) Yes, it can be gamed a little bit, as can any mathematical system. RPI's gameability is a function of its simplicity to understand, and I'm not sure coming up with something that only a mathematicion can figure out is the solution.
February 15th, 2013 at 5:46 PM ^
I wanted to say not that the RPI uses exactly a sum of ranks but that the method is very similar in principle (by adding the ranks of teams after crudely weighting them by a factor of 2/3 or 1/3,)
February 15th, 2013 at 5:54 PM ^
The RPI does not look at ranks; it looks at winning percentages of the opponents (and of their opponents). If Michigan plays a .500 team, the RPI doesn't care if that team is ranked #100 or #200. It only cares about that team's .500 winning percentage, the team's opponents' winning percentage, and the result of the game.
The RPI should never be used as an exact ranking, but it's not terrible for ballparking teams. There is no one ranking that is "correct," so it's kind of a useless argument whether the RPI does better or worse than anything else out there, as long as the selection committee isn't taking the ranking too literally (as they do in hockey and lacrosse and, recently, softball). The basketball committee does a reasonable job of ignoring the RPI when it needs to be ignored. They make some strange decisions, but it isn't the RPI forcing them into those decisions.
February 15th, 2013 at 6:28 PM ^
if you prefer
February 15th, 2013 at 5:38 PM ^
I can't figure it out.
I put Michigan and New Mexico's schedule's side by side and don't see why New Mexico is higher rated. as far as SOS: It must be those really bad teams on Michigan's schedule dragging them down.
I agree. Something amiss.
Michigan | New Mexico | |||||||
Wins (RPI 1-50) | Wins (RPI 1-50) | |||||||
RPI | DATE | OPPONENT | SCORE | RPI | DATE | OPPONENT | SCORE | |
28 | Nov. 21 | Pittsburgh* | 67-62 | 21 | Nov. 19 | Connecticut* | 66-60 | |
22 | Nov. 23 | Kansas State* | 71-57 | 32 | Dec. 27 | @ Cincinnati | 55-54 | |
18 | Nov. 27 | North Carolina State | 79-72 | 25 | Jan. 9 | UNLV | 65-60 | |
14 | Jan. 17 | @ Minnesota | 83-75 | 43 | Jan. 16 | @ Boise State | 79-74 | |
27 | Jan. 27 | @ Illinois | 74-60 | 13 | Jan. 23 | Colorado State | 66-61 | |
23 | Feb. 5 | Ohio State | 76-74 | |||||
Losses (RPI 1-50) | Losses (RPI 1-50) | |||||||
RPI | DATE | OPPONENT | SCORE | RPI | DATE | OPPONENT | SCORE | |
23 | Jan. 13 | @ Ohio State | 53-56 | 31 | Jan. 26 | @ San Diego State | 34-55 | |
12 | Feb. 2 | @ Indiana | 73-81 | 25 | Feb. 9 | @ UNLV | 55-64 | |
35 | Feb. 9 | @ Wisconsin | 62-65 | |||||
5 | Feb. 12 | @ Michigan State | 52-75 | |||||
Wins (RPI 51-100) | Wins (RPI 51-100) | |||||||
RPI | DATE | OPPONENT | SCORE | RPI | DATE | OPPONENT | SCORE | |
88 | Dec. 8 | Arkansas | 80-67 | 59 | Dec. 1 | @ Indiana State | 77-68 | |
89 | Dec. 15 | West Virginia* | 81-66 | 93 | Dec. 5 | USC | 75-67 | |
87 | Jan. 6 | Iowa | 95-67 | 73 | Dec. 8 | Valparaiso | 65-52 | |
91 | Jan. 9 | Nebraska | 62-47 | 84 | Dec. 15 | New Mexico State | 73-58 | |
84 | Dec. 19 | @ New Mexico State | 68-63 | |||||
60 | Jan. 30 | @ Wyoming | 63-59 | |||||
62 | Feb. 6 | Air Force | 81-58 | |||||
Losses (RPI 51-100) | Losses (RPI 51-100) | |||||||
No current losses in this range. | RPI | DATE | OPPONENT | SCORE | ||||
78 | Dec. 22 | South Dakota State | 65-70 | |||||
55 | Dec. 31 | @ Saint Louis | 46-60 | |||||
Wins (RPI 101-150) | Wins (RPI 101-150) | |||||||
RPI | DATE | OPPONENT | SCORE | RPI | DATE | OPPONENT | SCORE | |
116 | Dec. 4 | Western Michigan | 73-41 | 102 | Nov. 13 | Davidson | 86-81 | |
103 | Jan. 3 | @ Northwestern | 94-66 | 109 | Nov. 16 | Illinois-Chicago* | 66-59 | |
125 | Jan. 24 | Purdue | 68-53 | 131 | Nov. 18 | George Mason* | 70-69 | |
103 | Jan. 30 | Northwestern | 68-46 | 127 | Feb. 2 | Nevada | 75-62 | |
Wins (RPI 150+) | Wins (RPI 150+) | |||||||
RPI | DATE | OPPONENT | SCORE | RPI | DATE | OPPONENT | SCORE | |
308 | Nov. 12 | IUPUI | 91-54 | 190 | Nov. 23 | Idaho | 73-58 | |
184 | Nov. 13 | Cleveland State | 77-47 | 222 | Nov. 25 | Portland | 69-54 | |
175 | Dec. 1 | @ Bradley | 74-66 | 154 | Nov. 28 | Mercer | 76-58 | |
336 | Dec. 11 | Binghamton | 67-39 | 156 | Jan. 12 | Fresno State | 72-45 | |
235 | Dec. 20 | Eastern Michigan | 93-54 | 156 | Feb. 13 | @ Fresno State | 54-48 | |
249 | Dec. 29 | Central Michigan | 88-73 |
February 15th, 2013 at 6:42 PM ^
RPI is old news. KenPom is where it's at. He has our SOS at #8. Besides, whether our SOS is #1 or 8 or 25, it really doesn't matter.
February 15th, 2013 at 7:27 PM ^
Here is the basic form of the RPI. It supports my reasoning below about the RPI being a bunch of baloney--in part because it adds rather than multiples a measure of SOS with a measure related to the WL record (win pct or WP). The RPI formula: that determines your rank is:
RPI = (WP * 0.25) + SOS
where SOS=(OWP * 0.50) + (OOWP * 0.25)
and OWP=opponents win pct and OOWP = opponents' opponents win pct
Note that there is some correction made for factors like home vs away games etc in computing percentages.
February 15th, 2013 at 11:05 PM ^
RPI is, well, it's pretty asinine. I'm not making any strong statements here either. Most people who know anything are now leaning toward KenPom as far as evaluating College Basketball and it's teams as a whole. RPI is very flawed, not that KenPom is perfect.
February 16th, 2013 at 9:44 AM ^
Massey has M's SOS at #4, behind Minnesota, Duke and Wisconsin. At Sagarin it's #3 behind Minnesota and Wisconsin.
RPI puts a lot more weight on the extreme mismatch games than any of the decent computer-ranking systems. As a predictive measure it's terrible, partly for that reason. But it makes sense for the NCAA to use it, or to be more precise to announce publicly that it's what they use (behind closed doors they can look at anything they want) if their purpose is to influence scheduling. Teams can't do much about their conference schedule but they can do something about the OOC cupcakes; the NCAA's choosing a measure that punishes teams for scheduling them.
---
One of the reasons I like Massey is that he rates all divisions, not just D1, and it lets you see just how really terrible some of these teams were.
Binghamton is #678 in the country. (For comparison, Slippery Rock, which was only considered an exhibition, is #480.) In their last 19 games they've lost 18 and have a one-point win over Maine. Their other two wins are over #328 St. Peter's and #1806 Marywood, who's 6-19 playing in one of the worst D3 conferences in the country.
IUPUI is #525 and hasn't beaten anyone outside the state of Indiana since November (they've got wins over Indiana-East, Indiana-Northwest, Ball State and IPFW.