Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Does anyone remember a season where the various computer rankings differed this much this late? It's a good thing that we use the system that we do; it would be a lot harder for a committee to send a team sitting in the third spot to visit No. 1 to save a flight.

Women’s Division I PairWise Rankings

Rank	Team	PWR	W-L-T	Win %	Win % Rank	RPI	RPI Rank	vs. TUC	TUC %
1	Minnesota	11	36-0-0	1.000	1	.7344*	1	10-0-0	1.000
2	Cornell	10	25-5-1	.8226	2	.6186*	2	9-5-0	.6429
3	Boston College	9	26-5-3	.8088	3	.6138*	3	10-3-1	.7500
4t	Harvard	7	23-5-3	.7903	6	.6044*	4	6-5-1	.5417
4t	Boston University	7	24-5-3	.7969	5	.6027*	5	6-4-1	.5909
4t	Clarkson	7	28-8-0	.7778	7	.5985*	6	10-6-0	.6250
7	Mercyhurst	5	26-6-1	.8030	4	.5842*	7	3-3-1	.5000
8t	North Dakota	3	25-10-1	.7083	9	.5803*	8	2-8-0	.2000
8t	Wisconsin	3	23-9-2	.7059	10	.5770*	9	3-5-0	.3750
8t	Northeastern	3	22-10-2	.6765	11	.5569*	10	3-9-0	.2500
11	Quinnipiac	1	20-11-4	.6286	12	.5435	11	3-8-2	.3077
12	St. Lawrence	0	18-13-5	.5694	15t	.5371	12	3-12-2	.2353

Women’s Division I Ratings Percentage Index

Rank	Team	RPI	W-L-T	Win %	Win % Rank	SOS	SOS Rank
1	Minnesota	.7344*	36-0-0	1.000	1	.5275	4
2	St. Anselm	.6999*	6-1-1	.8125	3	.5023	22
3	Holy Cross	.6420*	4-1-1	.7500	9	.5316	2
4	Cornell	.6186*	25-5-1	.8226	2	.5226	8
5	Boston College	.6138*	26-5-3	.8088	4	.5131	14
6	Harvard	.6044*	23-5-3	.7903	7	.5180	12
7	Boston University	.6027*	24-5-3	.7969	6	.5048	21
8	Clarkson	.5985*	28-8-0	.7778	8	.5190	11
9	Mercyhurst	.5842*	26-6-1	.8030	5	.4839	31
10	North Dakota	.5803*	25-10-1	.7083	10	.5250	6
11	Wisconsin	.5770*	23-9-2	.7059	11	.5217	9
12	Northeastern	.5569*	22-10-2	.6765	12	.5055	20
13	Quinnipiac	.5435	20-11-4	.6286	13	.5070	19
14	St. Lawrence	.5371	18-13-5	.5694	16t	.5232	7
15	Dartmouth	.5354	16-10-5	.5968	14	.5092	18
16	Ohio State	.5350	19-14-3	.5694	16t	.5202	10

KRACH: Division I and II Women

Rank	Team	Rating	RRWP	Win % Rk	W-L-T	Win %	Win Ratio	SOS Rk	SOS
1	Minnesota	1.834e+04	.9917	1	36-0-0	1.0000	**	4	251.2
2	Cornell	835.9	.8590	2	25-5-1	0.8226	4.636	9	192.9
3	Boston College	781.0	.8517	4	26-5-3	0.8088	4.231	8	195.2
4	Boston University	682.5	.8365	6	24-5-3	0.7969	3.923	11	183.7
5	North Dakota	666.5	.8337	10	25-10-1	0.7083	2.429	2	282.0
6	Harvard	659.9	.8325	7	23-5-3	0.7903	3.769	10	184.8
7	Wisconsin	627.4	.8264	11	23-9-2	0.7059	2.400	3	268.9
8	Clarkson	598.6	.8205	8	28-8-0	0.7778	3.500	12	178.5
9	Mercyhurst	407.8	.7672	5	26-6-1	0.8030	4.077	19	105.7
10	Northeastern	325.3	.7311	12	22-10-2	0.6765	2.091	15	159.2
11	Ohio State	287.9	.7102	16t	19-14-3	0.5694	1.323	5	219.3
12	Minnesota-Duluth	272.7	.7007	21t	14-16-4	0.4706	0.889	1	305.7
13	St. Lawrence	222.3	.6632	16t	18-13-5	0.5694	1.323	14	169.4
14	Quinnipiac	213.9	.6558	13	20-11-4	0.6286	1.692	18	128.3

Current Rutter Computer Rankings
For games played through 03/02/2013

 	Team	Rating	RPI.Rank	RPI
1	Minnesota	 3.2898	 1	0.6687
2	Cornell	 1.3267	 2	0.6111
3	North Dakota	 1.2917	 7	0.5779
4	Wisconsin	 1.2346	 8	0.5761
5	Boston College	 1.1889	 3	0.6010
6	Boston University	 1.1715	 6	0.5919
7	Harvard	 1.1020	 4	0.5990
8	Clarkson	 1.0253	 5	0.5947
9	Mercyhurst	 0.7795	 9	0.5698
10	Minnesota Duluth	 0.7261	15	0.5206
11	Ohio State	 0.6843	14	0.5328
12	Northeastern	 0.6438	10	0.5558
13	Dartmouth	 0.3603	13	0.5346
14	St. Lawrence	 0.3465	12	0.5350
15	Quinnipiac	 0.3255	11	0.5412

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Yeah, never seen such a huge ranking disparity. And remember, earlier in the year I was posting about how consistent the PWR/KRACH/RPI were. That changed fast.

As I said before, probably one of the main factors is how to assess the losses to Minnesota, as Minnesota's strength is difficult to identify. RPI/PWR imposes a huge penalty, KRACH imposes a small penalty, and Rutter imposes a tiny penalty. The strength they each assign to Minnesota is fairly arbitrary.

Another factor, though this may be no worse than usual, is the lack of games between the best Eastern and Western teams. It looks to me like the only series between East-West NCAA teams will be UND-Clarkson, and maybe there won't be any if UND doesn't make the cut.

So when KRACH and Rutter identify UND & Wisconsin as being as good as their are, a lot of that identification is coming from relative performance against UNH & UMD --- BC/BU split UMD, while Wisconsin also split UMD and UND went 3-0-1 vs. UMD.... and then meanwhile, Minnesota and Wisconsin swept UNH, while BC&BU each lost once to UNH. And then you look at how the ECAC teams did against BC/BU/UNH. So that's really what's driving a lot of the East-West comparison in these statistical rankings like KRACH/RPI.

Now I think it's kind of weak to make strong conclusions about the WCHA's greatness this season based on those games, but those are the games we have. Now one outcome I would hope would result from moving to a better a system is for the top WCHA teams, esp. Wisconsin, to play tougher schedules. I can't blame them for not doing it now when they're not actually rewarded for it.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Now I think it's kind of weak to make strong conclusions about the WCHA's greatness this season based on those games, but those are the games we have.

That's true; the strength of the WCHA is always guesswork because of so few games against Eastern teams, and the majority of those come in the first half. PWR makes the problem worse by ignoring many of those games outside of RPI, which can't even figure out that Holy Cross and St. Anselm really aren't two of the top three teams in the country. It's a difficult problem to solve with the best of tools, and all we have is a rusty old Crescent wrench that is missing the gear used to adjust the jaw.

 

[OnMAA]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Looking at this without doing the math, looks to me the top 6 are all still playing, and are fairly safe and in the tournament. Just a matter of positioning and home ice. Only slot really locked in is the top spot.

Mercyhurst is fairly safe, unless they lose the CHA semi or if there are two surprise autobids. Surprise Autobids outside of the top 8 can still come from NE,Providence,Ohio, SLU or Quinnipiac.

Barring a surprise Autobid, The race for eight is most intriguing...At first glance the winner of UND/Wisco is most likely to take that spot. NE winning their semi but not getting the autobid might not be enough, but not sure on that.

IMHO, best chance of Autobid would be out of HE in the form of NE. They have proven they can beat the top 2. Hard pressed to see the ECAC going to SLU or Quinny, and even more hard pressed to see Minny lose to OHIO.

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

And just to clarify what I meant about WCHA teams being "rewarded" for playing eastern teams: right now those games are low-upside, high-downside for all the teams that play them in both leagues. What you want is reward when you win the games and appropriate punishment when you lose the games. You're always going to have a little more variance in the assessment -- maybe you still end up with a situation where you're a WCHA team that ends up ranked behind Cornell, because Cornell beats UMD once in 1 try and you only beat UMD 4 of 5 times (totally hypothetical). But ideally you have some head-to-head and quality common opponents to make these decisions.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Mercyhurst is fairly safe, unless they lose the CHA semi or if there are two surprise autobids. Surprise Autobids outside of the top 8 can still come from NE,Providence,Ohio, SLU or Quinnipiac.

FYP.

Also, for the purpose of Mercyhurst, UND or Wisconsin is just as big a problem when it comes to taking an autobid as is Ohio State. The Lakers are currently ahead of those teams, but that won't help them if one of those WCHA squads holds an auto bid.

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Barring a surprise Autobid, The race for eight is most intriguing...At first glance the winner of UND/Wisco is most likely to take that spot. NE winning their semi but not getting the autobid might not be enough, but not sure on that.

If Northeastern loses in the Hockey East final, then Northeastern is not passing either UND or Wisconsin in the PWR. I think No. 8 PWR is simple: whoever wins the WCHA semifinal gets it. (I ignore disaster scenarios for Mercyhurst.) Only question is whether there's a surprise autobid winner.

The NCAA QF home ice scenarios are what's a lot murkier. I think Clarkson is pretty safe to get QF home ice if it beats Harvard AND its performance matches BU this weekend. If Harvard beats Clarkson, and its performance matches BU this weekend, then that's a really tight comparison between Harvard and BU -- but probably doesn't matter much because they'd probably be playing each other anyway. (But my rough prediction is that if both Harvard & BU lost in the conference finals to BC/Cornell, BU would get home ice --- if both Harvard & BU won their leagues, then probably the committee would find a way to give them both home ice, and then BC would get the last spot and send Cornell on the road (assuming BC hasn't lost in the HE semis.))

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Or to clarify the 2-6 situation, I think the way the committee sees it right now is BC > Cornell > Clarkson > BU > Harvard, and whoever loses Harvard-Clarkson is clearly 6 if BU beats Providence. If BU loses to Providence, probably a decent chance they're No. 6 regardless. The teams are close enough that winning a conference title clearly gets you home ice. The last home ice slot would probably go to the best conference final loser based on the ranking above (assuming that conference final loss is to one of the current 2-6 ranked teams.)

I say BC > Cornell now because BC beat Cornell H2H and has the much better record against teams under consideration, while Cornelll has tiny edges in RPI and common opponents -- and unlike the men, the committee doesn't have the obligation to use any tiny edge in RPI as the tiebreaker when the criteria are evenly split.

 

[LakersFan]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

One of my favorite features of my rankings (not that I am biased or anything) is the "estimates of uncertainty" page (http://math.bd.psu.edu/faculty/rutter/CurrentD1Unc.html). Here are the top 12 teams. Since I use a Bayesian approach, I can look at random samples of the distribution of the ratings. From there, I can rank the teams and keep keep track of this for 1000s of samples. Taken together, the 25th percentile and 75th percentile represent the middle 50% of the distribution of rankings, and the "Percent Top 8" shows the percentage of time the team is ranked in the top 8.

  	Team 	25th Perc. 75th Perc. Percent Top 8
1 	Minnesota 	1 	1 	100.0
2 	Cornell 	2 	6 	94.3
3 	North Dakota 	3 	6 	93.6
4 	Wisconsin 	3 	7 	90.2
5 	Boston College 	3 	7 	89.6
6 	Boston Univ.	4 	7 	86.6
7 	Harvard 	4 	8 	79.1
8 	Clarkson 	5 	9 	71.5
9 	Mercyhurst 	8 	11 	32.4
10 	UMD 		9 	12 	24.0
11 	Ohio State 	9 	12 	20.3
12 	Northeastern 	9 	13 	14.6

I just wanted to demonstrate how difficult the east/west comparision is this year. As you can see from the table, Cornell is ranked anywhere from 2 to 6 while North Dakota is anywhere from 3 to 6. More games between the top east and west teams would help eliminate some of this uncertanity as to how to place the teams relative to each other. This is also a function of how closely rated teams 2-8 are, but the lack of good contrast between the regions is evident. I don't recall seeing this much "noise" in the rankings this late in the season ever.

My point is that ranking the teams relative to each other is very difficult this year. Pick a different set of criteria and a different set of rankings will result.

For reference, here is the same table at the end of last year. Last year it was who is 7 and 8. The top 5 have less variablity as to where they should be ranked.

  	Team 	25th Perc. 	75th Perc. 	Percent Top 8
1 	Wisconsin 	1 	2 	100.0
2 	Minnesota 	1 	2 	100.0
3 	Cornell 	3 	4 	99.2
4 	North Dakota 	4 	5 	96.1
5 	UMD 	        4 	6 	93.1
6 	Boston College 	6 	10 	55.2
7 	Harvard 	7 	12 	45.3
8 	Bemidji State 	7 	12 	38.8
9 	Boston Univ. 	7 	12 	38.7
10 	Mercyhurst 	8 	12 	34.2
11 	Northeastern 	8 	12 	33.1
12 	St. Lawrence 	8 	13 	28.3
13 	Ohio State 	8 	13 	27.0

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Thanks for reminding me of the uncertainty estimates.

My point is that ranking the teams relative to each other is very difficult this year. Pick a different set of criteria and a different set of rankings will result.

I don't think that's quite the right conclusion to draw here. If the rankings were based on just one game starting with a symmetric prior, we'd still say the rankings were very noisy, but every ranking system would say that the team that won the game was better.

So even though the rankings are uncertain, I don't think you'd easily find a sound model with Wisconsin and N. Dakota out of the top 8, no matter how flimsy the East-West comparisons might be.

(Maybe one possible model is if you built in that 10% probability a team randomly played 2 rating units below its usual mean potential, so then Clarkson is not beaten up so badly for losing to Princeton and Colgate.)

 

[Eeyore]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Something I found interesting looking at the full table for Rutter is that UND and Wisconsin both have higher raw RPI ratings than Mercyhurst does. It's only after subtracting out the wins that hurt RPI that Mercyhurst passes them.

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Right. Mercyhurst plays in the weakest league so that is inevitable.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

Something I found interesting looking at the full table for Rutter is that UND and Wisconsin both have higher raw RPI ratings than Mercyhurst does. It's only after subtracting out the wins that hurt RPI that Mercyhurst passes them.

An oddity in the RPI for Minnesota is that other than its wins against North Dakota, all of the other wins hurt and have to be subtracted.

 

[OnMAA]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

If SLU repeats as ECAC champ, the odds are that there will be four ECAC teams in the tourney and only one WCHA team. (is that correct ?).

Who would have thunk that at the start of the season.

Begs the question (too lazy to look it up)...When was the last time we had four teams from the same conference in the tournament ?.

P.S: Not in the last four years 2009-2012. Most was three WCHA teams a few times.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

If SLU repeats as ECAC champ, the odds are that there will be four ECAC teams in the tourney and only one WCHA team. (is that correct ?).

Correct on only one WCHA. If Northeastern or PC wins Hockey East and SLU takes ECAC, I think that would mean 4 ECAC, 3 HEA, 1 WCHA, and no Mercyhurst, but I'm not 100% on the Lakers being out.

Begs the question (too lazy to look it up)...When was the last time we had four teams from the same conference in the tournament ?.

Never. It's only been a eight-team field since 2005.

 

[dave1381]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

If Northeastern or PC wins Hockey East and SLU takes ECAC, I think that would mean 4 ECAC, 3 HEA, 1 WCHA, and no Mercyhurst, but I'm not 100% on the Lakers being out.

In that scenario, I see no case for Mercyhurst getting in over BU or Clarkson. Their best shot would be if Harvard lost to Clarkson. Then I'm guessing Harvard would have a half-game edge in RPI but trail in common opponents, while being even in record vs. teams under consideration, so some chance Mercyhurst gets selected under such conditions.

 

[NUWH DogHouse]

Correct on only one WCHA. If Northeastern or PC wins Hockey East and SLU takes ECAC, I think that would mean 4 ECAC, 3 HEA, 1 WCHA, and no Mercyhurst, but I'm not 100% on the Lakers being out.

Never. It's only been a eight-team field since 2005.

And yet the one team is most likely going to win it.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

And yet the one team is most likely going to win it.

One never knows. When I look at the record book, I see the 2002-03 Harvard team all over the place, but the Crimson didn't end up winning either the ECAC Tournament or the NCAA crown. That team has to be one of the 10 best teams of all time, and the UMD team they lost to in double OT is as well, but it is still odd that they didn't win either tourney.

 

[OnMAA]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

One never knows. When I look at the record book, I see the 2002-03 Harvard team all over the place, but the Crimson didn't end up winning either the ECAC Tournament or the NCAA crown. That team has to be one of the 10 best teams of all time, and the UMD team they lost to in double OT is as well, but it is still odd that they didn't win either tourney.

There is that Patriot's curse that may come into play !.

 

[ARM]

Re: Preliminary & Ongoing Pairwise Watch and 2013 NCAA Tournament Musing Thread

There is that Patriot's curse that may come into play !.

That's an East thing. In Boston, fans blame dead guys and crippled-up 1st basemen when they don't win. Out here, we either win or lose.