The pairwise and CBB: Why/not?

Submitted by CarrIsMyHomeboy on February 27th, 2011 at 12:59 PM

Not much in the way of interesting information provision here. I'm just posing a scenario and question:

The pairwise, though imperfect, seems to work relatively well for selecting the NCAA hockey tournament field. It is an objective approach (or at least it's as objective as an algorithm-like substance created by man can be), which avoids the need for committee free-wheeling-n-dealing, which is nice. And it seems to yield rankings less prone to the "Wha!?" effect than RPI. Tell me though: Why couldn't it work for CBB? Is that field of teams too large with too few interactions to fairly yield good comparisons? In which ways would Pairwise be better than the current system?

Clearly, I've only given this a rudimentary amount of thought so far. Help me take the analysis to the next step. Please and thank you.

Comments

Mr Miggle

February 27th, 2011 at 4:00 PM ^

RPI doesn't seem to be given that much weight as it is.

The great disparity in schedule strength in CBB would make me reluctant to follow some strict algorithm. Those formulas tend to be much better at comparing apples to apples. I can't help but think of some of the crazy computer polls in CFB.

justingoblue

February 27th, 2011 at 1:06 PM ^

I don't have any statistical knowledge about this whatsoever, but I would bet that well over 50% of Division 1 AD's would flip shit about drastically changing the basketball formula after about thirty years.

CarrIsMyHomeboy

February 27th, 2011 at 1:23 PM ^

When I posed the question, I meant for it to be an academic one. I'm not trying to start a grassroots campaign for change (and, to be frank, I'm relatively disinterested in basketball). I'm simply curious as to whether or not a pairwise-based tourney field might provide CBB enthusiasts fewer headscratchers than an RPI-based tourney field. Right now, I'm leaning toward "no."

South Bend Wolverine

February 27th, 2011 at 6:50 PM ^

How would using the pairwise to determine the tourney field change March Madness?  If anything, it would improve it, since the bubble/bracketology conversations would involve a lot less "I'm kinda vibing these guys" and a lot more "these metrics indicate that ..."  The tourney itself would proceed apace.

joeyb

February 27th, 2011 at 1:14 PM ^

I think there are too many teams for the pairwise to be effective. The pairwise works because many of the top teams have played each other. I mean, how many teams on the top half of the RPI has Michigan played? 15, if that? Now, try to compare to a team like Maryland. 2 teams in common out of 30 on each teams schedule and 150 or so teams in contention? There just isn't enough to compare teams on.

Plus, I don't see anything wrong with adding a little subjectivity to the selection process. Subjectivity is Michigan's only chance for the Kansas, OSU, or Wisconsin losses to count for anything.

Seth9

February 27th, 2011 at 2:08 PM ^

The Pairwise is a an absolutely terrible system. It's RPI with three extra factors, two of which should never have been added to the system, namely record vs. Teams above a certain RPI and record vs. common opponents. The first has little to do with which team is the better team. The second has absolutely nothing to do with which team is the better team.

This leads to really stupid situations. Case-in-point: Michigan won a comparison with Merrimack to make us 4th in the Pairwise outright because Alaska-Anchorage (a team we did not play this season) beat Alaska-Fairbanks (a team we played 4 times this season). Things like this that lead to incredibly non-sensical rankings happen all the time in the Pairwise. So no, we don't want it in College Basketball under any circumstances. Actually, we'd be better off getting rid of it for hockey too.

Michigan Arrogance

February 27th, 2011 at 2:54 PM ^

well, to be fair... the pairwise isn't intended to be looked at on an hourly, daily or even weekly basis. i doesn't matter if team X beats team y and causes team z to jump ahead for a day. the math only gets applied at the end of the season.

it's like the BCS formula. people flip their shit in Oct when one team wins and jumps another, but it doesn't matter a lick.

Seth9

February 27th, 2011 at 3:14 PM ^

The problem is that this kind of thing happens at the end of the season too for the same stupid reasons. Basically, the Pairwise decides between which teams go to the tournament and which teams don't on a set of criteria that is rather arbitrary and nonsensical. It attempts to take the things that a tournament committee would take into account and quantify it, so as to remove subjective input into the process. The problem is that it is far from a perfect system, and that we need people to actually look at weird results and determine whether a ranking makes sense or not. Alternatively, they could devise a superior statistical system. Or they could do something similar to the BCS and average a variety of computer rankings to determine seeding and the like. Anything would be better than factoring in one point of comparison where a win over North Dakota is worth the same as a win over Alaska-Anchorage and another where it's to your benefit to play Michigan Tech 4 times in a year.

josejose50

February 27th, 2011 at 2:14 PM ^

Dont forget the minor sh*t storm that happened when the Hockey committee decided to change the rules on what a TUC is midway through the season. I think it would be better at figuring out the bubble teams that should be in or out, but I think joeyb is right that it might be too difficult to correlate results. 

As to subjectivity, I think that there can be some subjectivity in the selection process, but there's been a lot of Wha?! picks the last few years and adding an objective data point like an Pairwise would be a good way to take out some of the uncertainty. 

Alton

February 27th, 2011 at 2:30 PM ^

To be fair, the rules didn't change in the middle of the season.  They were changed in September, but nobody noticed the change until the middle of the season.

Hockey, I think, is the only sport with 100 percent transparent rules for selecting the participants in the tournament.  For the most part, NCAA sports committees don't want their process to be transparent--they want to give themselves some room to make spontaneous decisions...often for political reasons. 

mfan_in_ohio

February 27th, 2011 at 2:54 PM ^

I like the pairwise for hockey because it does a really good job of simulating what the committee takes into account.  I don't know how the basketball committee works, though.  The biggest problem is that hockey has 58 teams, compared to almost 350 for basketball.  That means, first, that most teams don't have common opponents, as opposed to hockey where most teams have several.  So that comparison can't be used for the pairwise.  Also, even though there are fewer than 100 teams to look at for the tournament selection, you have to have results for each game for every team to do pairwise the way that hockey does. 

To set up a pairwise comparison, I would do it like this: Compare each pair of teams in the RPI top 100 in the following categories:

1.  RPI

2.  Record against the RPI top 100

3.  Marquee wins minus bad losses, adjusted for home/road.

4.  Head-to-head

A team gets one point for winning each of the first two categories, and a point for any head-to-head victory.  Categories 1, 2, and 4 are basically the same as those used by the hockey pairwise.  Rather than use common opponents (since that is largely irrelevant), I thought I would include a category where each team gets a point for any "marquee" wins and loses a point for any loss against a team outside the RPI top 100.  Since "key wins" and "key losses" are often cited as a reason to put a team in or out of the field, it seems like a reasonable stat.  It also makes sense that a win over a team like Purdue should be canceled out by a loss to Indiana. As far as "adjusting" for home and road, I think a loss to the #110 team on the road is roughly the same as losing to the #70 team at home, give or take 5 spots.  So while our loss at Indiana (#185) is bad, it's nowhere close to Penn State's loss at home to Maine (#198).  The same is true for marquee wins.  I would include top 40 on the road and top 20 at home (top 30 at a neutral site) for "marquee" wins, and a bad loss would be outside the top 80 at home, top 100 neutral, and top 120 on the road.

Of course, you'd have to enter all of these values for the top 100 teams, understanding that who the top 100 is will change from day to day.  I'm sure that you can write a computer program to calculate this (I think using Excel might be too difficult), but it's more of an exercise in curiosity since we don't really know exactly what the committee takes into account.  If anyone without a wife, kid, and job (i.e. not me) wants to do this, I'd love to see it.

GCS

February 27th, 2011 at 3:10 PM ^

I don't think Pairwise is a good idea for CBB because the basketball committee hasn't proven itself incapable of leaving out brand name teams having subpar seasons. My hockey knowledge isn't deep enough to provide specific examples (is anybody from HSR around?),  but there were several instances in the 90's of name teams underachieving and still getting in the tournament ahead of teams that were clearly more deserving. Since the basketball committee has shown that they can won't make the same type of decision (for example, North Carolina last year), there is no need to move their sport to a flimsy mathematical model for tournament selection.

MaizeAndBlueWahoo

February 27th, 2011 at 4:33 PM ^

I actually went and did this last year, pairwised-out the CBB field.  I was thinking of making it a diary - what March Madness would look like if seeded via pairwise.  I took the top 100 teams.  Result?  Borrrrrriiiiiinnnnggg.  Kiboshed the whole idea.  Very few changes in RPI order ended up being made.  Why?  Because teams typically had maybe one or two common opponents at most, unless they were conference opponents.  And those common opponents were usually shitcans, so that particular part of the comparison was always tied at 1-0.  So throw that out for most comparisons.  Most teams didn't play each other, either, unless, again, they were conference opponents.  And if they did it was only once, unlike in hockey where you have multiple chances to earn a Pairwise point in a game, and there are ties to boot. 

This all means that "record against TUC" and "RPI ranking" were the only possible points of comparison in probably 75-80% of the matchups, which automatically threw the comparison to the team with the RPI because when all else fails, RPI is the tiebreaker.

The end result is that the pairwise order almost - not quite, but almost - mirrors the RPI.  So it's basically a worthless exercise. Occasionally a team makes a major leap or takes a major drop, but the only result there would be a lot of squawking as to why we're doing this just to give some team - which would probably otherwise look awfully undeserving - a seat at the table and bumping someone else.

Tater

February 28th, 2011 at 11:23 AM ^

They could use the RPI to take all at-large teams except for four "committee choices."  The committee could use the RPI for the next four most years, but also use the four choices to correct any really deserving team being left out. 

Basketball can be a capricious game in which the "best" team doesn't always win, so I can see going outside of a statistical model for a few teams.  Really, though, the selection process is, in many cases, as subjective as the game is capricious. 

It all works out, though.  Usually, a team seeded third or higher wins the tournament, so any eligible team with a true chance to win will almost always make it in.  From "the math" and history, only the top twelve teams have a real chance to win, anyway.  So, while a bubble team that isn't picked can have a legitimate gripe about not getting in, they really weren't going to win the tournament anyway.