I realize that criticizing the RPI is like beating a dead horse. Well, not exactly since the RPI is still very alive. But the problems with the RPI are pretty widely acknowledged and clearly documented.
That doesn't mean that we can't be redundant and vent a little bit, too.
Lately Kansas has been #1 in the RPI. That in itself isn't unusual; it seems that Kansas is always #1 in the RPI. The Jayhawks are usually one of the best teams and, more importantly, they tend to play the toughest schedule of any of the top teams. Hence, due to the RPI formula which rewards schedule strength above all else, we often find Kansas at #1 in the RPI. Often it's very justifiable.
This year though, the juxtaposition of Kansas at #1 and Kentucky behind them at #2 is too ludicrous to ignore. There should never be a system that has Kansas ahead of Kentucky at this point in the season, especially not one that's so important come Selection Sunday.
Kansas is 20-4, and that's great. Kentucky is 24-0, and that doesn't automatically mean they should be ahead of a 20-4 Kansas. Schedule strength, and more specifically, which teams you've beaten and which you've lost to, will obviously affect a rating system. Last year's Wichita State team was also 24-0 after 24 games, but you could argue for Kansas ranking ahead of the Shockers based on schedule strength. But not Kentucky. Let's take a look (RPI through games of 2/10/15):
RPI Div I SOS vs. 51- 101- Record
Rnk Team record RPI rnk t50 100 200 Road Neut Home
1. Kansas 20-4 .6965 1 8-4 4-0 8-0 5-3 3-1 12-0
2. Kentucky 24-0 .6960 10 9-0 5-0 8-0 7-0 2-0 15-0
You'll notice Kansas is #1 in SOS, but that Kentucky is #10. Should that override the fact that Kentucky is 24-0 and Kansas is 20-4? In a system where SOS is 3/4 of the formula, it does. Here are the SOS breakdowns:
RPI Div I [.25 .50 .25 ]
rnk Team record RPI= [AWP + OPW + OOWP] SOS [rnk]
1. Kansas 20-4 .6965 .8600 .6785 .5691 .6420[ 1]
2. Kentucky 24-0 .6960 1.000 .6078 .5685 .5947[ 10]
Kansas has an OPW of .6785 to Kentucky's .6078—that's the Opponents' Win Percentage. The OOPW's (Opponents' opponents win percentage) are about even. The OPW is 50% of the RPI formula, while the AWP (the team's own Adjusted Win Percentage) counts for just 25%.
Bottom line, Kansas' OWP overwhelms Kentucky's AWP and ranks the Jayhawks in front, .6965 to .6960.
In any case, the balance of the formula, while one of its problems, is not the main issue. Adjusting the weights might make the RPI a bit better, but its real shortcoming is that the formula loses all knowledge of which teams a given team has defeated or lost to.
This is why various pundits and bracketologists—and the selection committee—are forced to rely on "top 50 wins" and "quality road wins" to see if a team is actually worthy of a bid. If the RPI was good enough on its own, we wouldn't have to do this. But we know that the RPI can't see the forest for the trees, for example, let's go back to Kansas and Kentucky:
RPI vs. 51- 101- Record
Rnk Team t50 100 200 Road Neut Home
1. Kansas 8-4 4-0 8-0 5-3 3-1 12-0
2. Kentucky 9-0 5-0 8-0 7-0 2-0 15-0
Look at top 50 wins. Kentucky is 9-0 and Kansas is 8-4. Not only does Kentucky have a better percentage (100% to 66.7%) against the top 50, they actually have more top 50 wins. It's not like Kentucky is 2-0 and Kansas 8-4. Kentucky is 9-0. They also have more wins against the second 50, and have a better road record (7-0 to 5-3). The RPI includes a bonus for road wins (that's the "adjusted" part of the APW) but it doesn't know which teams you beat on the road; the bonus is the same.
We're also choosing to ignore the real elephant in the room—that is, Kentucky's 72-40 neutral court win over the Jayhawks. Mainly because it's too easy. Any 5-year-old would rank Kentucky ahead of Kansas based on that. And it's a bit unfair, too, since the RPI is explicitly designed to ignore victory margin. For all the RPI knows, the game went into triple overtime so there really isn't much difference between the teams.
However, the RPI is supposed boost a team that plays and beats good opponents. The problem is that it quickly forgets that Kentucky beat Kansas. That information is lost once Kentucky's AWP adds 1.0 to the win total. The OWP adds Kansas' record, and the OOPW adds Kansas' opponents' records into the mix. For Kansas, the RPI knows they lost 4 games, but not that one of those losses was to Kentucky. It knows they played the Wildcats and lost to 4 teams total.
Kansas shows up as one of the 9 top 50 wins Kentucky has, but only if you do some post-processing and tease this information out. It doesn't affect Kentucky's rating, and any bad teams Kentucky plays dilute their schedule strength so that playing Kansas might not matter. In fact, just playing Kansas is almost as good as beating Kansas—and for Kansas, that's true of playing Kentucky, too.
So what's the upshot? When listing a team's NCAA résumé, many sportswriters cite the RPI then also list other information that is needed to account for the RPI's shortcomings. Sometimes it's top 50 wins, or road record, but often they also list the SOS. That, I think, is a mistake. It implies that the RPI doesn't tell us enough about the team's SOS, when in fact it tells us too much if anything. It's adding redundant information, and suggests that a good SOS somehow "validates" a team's RPI, when in reality the SOS is guiding the RPI from the start. An example is ESPN's Bubble Watch, which typically lists teams résumés like this:
SMU [19-5 (10-2), RPI: 22, SOS: 57]
Here we get some relevant information: overall record, conference record, RPI, and SOS. Our point being that the last bit of information is not informative: the RPI alone should tell you that SMU has a decent enough schedule, after looking at their overall record. An SOS of 57, being worse than their RPI, looks like a flaw, but remember that all sorts of teams can have a good SOS, whether they're good or not. So most top teams teams are not going to have an SOS that matches or exceeds their RPI.
In fact, of the top 25 in the RPI, only 5 teams have an SOS better than their RPI. 18 have an SOS ranking worse than their RPI, and two teams match. This is normal and expected, since the RPI is a mixture of winning percentage and SOS. Therefore, listing a team's SOS after the RPI is usually going to look like a criticism. SOS is such a huge part of RPI, it would be better to just list a team's record and then their SOS, since those are completely unrelated stats that come together to form the RPI.
A better idea would be to list their record vs. Top 50 RPI teams, since that is something that the RPI doesn't really reflect. Look how it changes Notre Dame's line:
Notre Dame [21-4 (9-3), RPI: 28, SOS: 113]
Notre Dame [21-4 (9-3), RPI: 28, Top 50: 4-3]
What do we learn from seeing the Irish have a 113 SOS? Nothing that their RPI—low at 28 given their 21-4 record—didn't already tell us. But seeing that they have a winning record against the top teams should tell us that they're a good team that has played its share of other good teams.
SMU's listing would be:
SMU [19-5 (10-2), RPI: 22, Top 50: 2-5]
and that actually tells us something new about the Mustangs. It tells us how many of the best teams they've played as well as how many they beat. It gives us information about the team's schedule strength, but in a different way than the RPI's SOS does, by including only the top teams played (who really cares if the cupcakes they beat were ranked 175 on average rather than 250?). And it immediately tells us that all of their losses were to top 50 teams—they didn't lose to any cupcakes or even below-the-bubble level competition.
Just like Kansas. Hey, maybe SMU should be the RPI #1 ?
Comments