A Composite of Ranks and Ratings, Using Massey, LRMC, Sonny Moore and RPI

[Fightin Choke]

So I wanted to create a composite ranking of teams using computer ranking systems, as 538 does when creating their NCAA Tournament model. They used Sagarin, Massey, LRMC (Bayesian version), and Sonny Moore last season, but as we stats geeks know, Sagarin ratings are no longer available for WCBB. As a less than excellent fill-in, I decided to use RPI. Why? Because I'm not really trying to determine winners (which RPI is not designed for anyway), but rather trying to determine an S-curve for tournament seeding (and so this is my nod to the NCAA Tournament Committee's reliance on RPI). As an aside, if I were trying to determine winners, I would certainly have included Massey's Power Rating, which his model uses to predict games. Instead, I used the regular Massey Rating.)

I didn't want the huge task of collecting all the data for all 300+ teams, or even the top 50 teams, so I only included teams that were top 25 in at least 1 of the 4 systems. If I team were top 25 in at least one system, then I collected their data for the other systems as well. For example, Michigan St. was 25th in LRMC, but was not top 25 in any other systems. Thus I looked up their data is the other three systems anyway, so their data weren't incomplete. Here are the composite rankings, where I averaged the ranks for each team and then ordered them from lowest to highest. By the way, just because Michigan State is the 32nd team on my list does NOT mean that they are the 33rd best team, as there are probably others that weren't top 25 in any system that have composite rankings that are higher than the Spartans.

One last caveat: I didn't look at good wins or bad losses, but the committee will surely do that. Of course you can use this composite to see the average rank of teams to help gauge good wins and bad losses.

So on to the data!

But I wasn't satisfied because Notre Dame still wasn't high enough for me, so I tried another variation. But since I cannot attach more than 1 file in a post, I will cover that in my subsequent post.

 

[Fightin Choke]

I joke about Notre Dame not being high enough, but seriously, ranks are not designed to be averaged, but ratings are, so I decided to average the ratings for each team. If you aren't following me with this argument, think about the Massey rankings. Notre Dame is 4th but has the same rating (2.65) as the team ranked 3rd (Oregon) and is only .01 below the team ranked second (UConn at 2.66). Also, Baylor (2.75) has a decent lead over UConn, so I thought that information should be included in a revised and I believe improved composite.

The next problem I encountered was that every rating system has a different numerical value, so I normalized each rating where the top team had a rating of 100 and the worst team had a rating of 0. LRMC did not have a rating, so instead of not using the system, I converted ranks into numbers from 0 - 100 so all 4 systems were equivalent. Here are the results:

So does this say anything about which teams should be on the top seed line? Was my math and reasoning correct? Any feedback would be great!

 

[CajunHusky]

So I wanted to create a composite ranking of teams using computer ranking systems, as 538 does when creating their NCAA Tournament model. They used Sagarin, Massey, LRMC (Bayesian version), and Sonny Moore last season, but as we stats geeks know, Sagarin ratings are no longer available for WCBB. As a less than excellent fill-in, I decided to use RPI. Why? Because I'm not really trying to determine winners (which RPI is not designed for anyway), but rather trying to determine an S-curve for tournament seeding (and so this is my nod to the NCAA Tournament Committee's reliance on RPI). As an aside, if I were trying to determine winners, I would certainly have included Massey's Power Rating, which his model uses to predict games. Instead, I used the regular Massey Rating.)

I didn't want the huge task of collecting all the data for all 300+ teams, or even the top 50 teams, so I only included teams that were top 25 in at least 1 of the 4 systems. If I team were top 25 in at least one system, then I collected their data for the other systems as well. For example, Michigan St. was 25th in LRMC, but was not top 25 in any other systems. Thus I looked up their data is the other three systems anyway, so their data weren't incomplete. Here are the composite rankings, where I averaged the ranks for each team and then ordered them from lowest to highest. By the way, just because Michigan State is the 32nd team on my list does NOT mean that they are the 33rd best team, as there are probably others that weren't top 25 in any system that have composite rankings that are higher than the Spartans.

One last caveat: I didn't look at good wins or bad losses, but the committee will surely do that. Of course you can use this composite to see the average rank of teams to help gauge good wins and bad losses.

So on to the data!

View attachment 40021

But I wasn't satisfied because Notre Dame still wasn't high enough for me, so I tried another variation. But since I cannot attach more than 1 file in a post, I will cover that in my subsequent post.

I was thinking maybe the mode might give us a more realistic rank. The rank using mode would be
1. ND
2. Baylor, UConn
4. Oregon

I think that looks reasonable. But I also think the top six or seven are very close and on any given night any of them can beat any other.

 

[MainefanSC]

The parity that we have this season, among the top 6 teams, is very exciting. As stated by CajunHusky, any of those could take a loss on any given night. The most important ingredient is health. As we have seen with Oregon, any loss of a player can "blow up" all stats. I hope and pray that all teams stay healthy throughout the season. I cringe when any player goes down, during any game I watch.

I enjoy perusing the data provided by Fightin Choke. Thanks for pulling it all together. As we know, the selection committee will surprise us all, with a few of their selections and placements. Here's to continued health of all of our women and let "clean play" be the rule. The way some coaches coach, leads me to believe they are in the wrong sport.

 

[diggerfoot]

Given that Sagarin is no longer available (I wondered why no one referred to it this year, thanks for filling me in), do you think 538 would also substitute the RPI in to make the best calculation they can? Or do you think they would just use one less data point as a more valid and/or reliable alternative? Whatever you think the answer to that is should be what you use, whether it strengthens Notre Dame's case or not. My hunch, including both Massey and Massey Power would satisfy 538's need for validity better than RPI while also helping out your objective as well.

 

[msf22b]

What the stats show (above) is pretty self-evident,

Without all the stats...I'm an impressions guy

I've written about fearing ND and clearly Baylor is a bad matchup for us.

It would be wonderful if they knock each other off in the semis with perhaps ND prevailing.

That would give us he best chance of Geno's magic (and the play of our young ladies) finding the right formula to fend off ND's starting-5 advantage.

But there is a huge amount of BB yet to be played before we get to that point with many dangers lurking.

As farmFan noted with regard the Pac-10

Tonight demonstrated just how deep and competitive the PAC-12 is. On any night, 9 of the 12 teams can threaten any other team

You can expand that out to the entirety of March's NCAA's.

Funny, I don't particularly fancy our chances.

 

[Fightin Choke]

As farmFan noted with regard the Pac-10

Tonight demonstrated just how deep and competitive the PAC-12 is. On any night, 9 of the 12 teams can threaten any other team

You can expand that out to the entirety of March's NCAA's.

Funny, I don't particularly fancy our chances.

I think UConn has an excellent chance, but in comparison to the past 5 years, there is definitely a different vibe. UConn is not the overwhelming favorite to win the NC (as they have been heading into the Big Dance the past 5 seasons). But they have earned a 1-seed and will be a very tough out.

 

[Vowelguy]

The AP poll is more correlated with seeding than RPI, so I would add AP and downweight RPI.

 

[azfan]

What the stats show (above) is pretty self-evident,

Without all the stats...I'm an impressions guy

I've written about fearing ND and clearly Baylor is a bad matchup for us.

It would be wonderful if they knock each other off in the semis with perhaps ND prevailing.

That would give us he best chance of Geno's magic (and the play of our young ladies) finding the right formula to fend off ND's starting-5 advantage.

But there is a huge amount of BB yet to be played before we get to that point with many dangers lurking.

As farmFan noted with regard the Pac-10

Tonight demonstrated just how deep and competitive the PAC-12 is. On any night, 9 of the 12 teams can threaten any other team

You can expand that out to the entirety of March's NCAA's.

Funny, I don't particularly fancy our chances.

The Pac-12 comment is dead on. Yesterday Cal at home beat ASU. Looking at these rankings and where ASU is placed I think this is a tad high. Looking at the Pac-12 I can see why the committee also needs to use the eye test and look at recent games. UCLA is the hottest team in the Pac-12 right now and I think a legitimate shot to be a dark horse in the tournament. While I hope ASU ends up in the top 16 I fear that won't be the case as our last three games are Stanford, Oregon, Oregon State. That said ASU will be playing on the first day of the tournament in Las Vegas. And assuming we win that game will get to play UCLA. I would urge you all to look for UCLA to make a run this year in the tournament.

 

[Fightin Choke]

So I made some changes based on comments I received. As @diggerfoot suggested, I ditched RPI and added Massey Power. I did not, however, add the AP poll as @vowelguy suggested because I was trying to see what the computers thought. But I added the AP poll ranks to the table so you can see which teams seem to be undervalued/overvalued by the pollsters. Each of the 4 ratings I used were normalized so that their scores ranged from 0-100. LRMC does not provide scores, so converted the ranks into scores. Teams were included if they were in the top 25 in any of the 4 ratings or the AP poll.

 

[IWearShoes]

So I made some changes based on comments I received. As @diggerfoot suggested, I ditched RPI and added Massey Power. I did not, however, add the AP poll as @vowelguy suggested because I was trying to see what the computers thought. But I added the AP poll ranks to the table so you can see which teams seem to be undervalued/overvalued by the pollsters. Each of the 4 ratings I used were normalized so that their scores ranged from 0-100. LRMC does not provide scores, so converted the ranks into scores. Teams were included if they were in the top 25 in any of the 4 ratings or the AP poll.

View attachment 40179

This Top 6 exactly matches my impression of team strength based on eye test only.