I’ve been getting a lot of questions about the location of the models. The old site used to have that fancy graphic to guide you there. We’ll change the design off season to make this easier, but for now I’ll just periodically remind everyone. Go to the Tips+ section and look for the “2013 Models” link.
-
Peter Tiernan has been using stats to analyze March Madness for 22 years. He writes for CBSSports.com and has also contributed to ESPN.com and SI.com. His insights into the NCAA basketball tournament can help you build a better bracket.Sign-up questions? Click here.
Follow Us
Did anyone every complete the upset toss up?
It’s in there toward the back. Ryan Tressler did it for me. I think it’s the one that likes Belmont (gasp!) in the Final Four.
I hope that upset model is accurate because I’m putting my wife’s entire retriment fund into it!
Ouch. Phaedrus (great Tangerine Dream song, btw). I don’t think I’d put $1000 down on a bracket. This is the ultimate exercise in “nestled probabilities.” The fact is, all the odds are long.
My WIFE’S retirement not mine.
In general I do wonder about the impact of the Pac 12 12 seeds. The NCAA guy said they were really 11′s but havd to be moved down a line. I dont really care why. But if they are really 11s disguised as 12s does that make it more likely that they will win versus a higher seed. After watching Ok State against KSU I have already penciled in Oregon over Ok State and California isn’t an upopular choice over UNLV but I haven’t made that pick yet.
Were they any other similar moves made by the NCAA? i.e. higher seeds that got demoted due to whatever.
I forgot to add the other part of the question. If two 11′s got moved down does that mean two 12′s got moved up. If so who would those imposters be? I guess I should pull that official NCAA seed list. That would give an accurate answer right?
I don’t look so much at the raw seed as the KenPom data. That’s the closest I think you come to the true quality of a team. Might also help to look at coaching history.
Well, obviously there were a couple of 12s that were bumped up to accommodate Oregon and Cal moving down — I believe one was Bucknell and the other was St. Mary’s.
Also, Colorado got bumped to a 10 seed (true 9) and Villanova moved up (true 10.) Must have been something about the Pac-12…
I need some clarification on the Upset/Tossup Rules model. I calculated the overachievers myself and got Duke, Michigan St, Marquette, New Mexico, St. Louis, and Oklahoma St. I don’t know where Gonzaga, Georgetown, Miami, and Wisconsin are coming from. I just rechecked Gonzaga and they appear to pass.
Joel – I can’t speak for that model because a member did it…and I haven’t checked the results. I wasn’t going to provide it this year. It yields pretty unpredictable results. If you get different results and you feel strongly about them, I say run with it.
Gonzaga I can explain, according to the sos numbers Pete has always used, they do in fact pass . . . because those sos numbers were not out yet when I did the model, I used Pomeroy’s strength of schedule numbers . . . I honestly forgot to recheck that once the numbers came out that Pete has always used for it, so you are correct there . . . Georgetown and Miami get underachieving checks because of their reb+turnover margin being less than 4.5 (gtown is 2.7, Miami is 4.1) . . . Wisconsin scored fewer than 66 a game, which gives them a check for underachieving
In filling out the model, I did not use the fact Gonzaga was an underachiever to decide any matchups, so it does not change the results of the model
Ryan – thanks for the clarification.
PT,
Just about every model has St Mary’s not only winning their play-in game but also beating Memphis. Will you update the models if St Mary’s lose? Thx
I will update the Outcome Matching one, and will probably comment on the rest in a blog.
Ryan – how did you come up with Notre Dame getting past Iowa St? I calculated ND as having victim qualities with Iowa St meeting victimizer standards.
Notre Dame does not get over 78% of their scoring from their starters (they get 76.8), so they do not meet all 3 qualifications (which is the requirement to avoid a seven seed) . . . i also don’t know where you got the victimizer criteria from, as thats not in the toss up article, where you perhaps looking at the seed matchup article? (which does have Iowa State over Notre Dame)
Looking at your criteria for picking a champ in your Final Four/Champ Model….you say eliminate any teams with fewer than 7 pre-tourney wins in their last 10 games. Yet you state that your model correctly picked Uconn in 2011 (they are shaded blue). However, in 2011, Uconn won 5 games in the Big East tournament and went 1-4 in their previos 5 games before the Big East tourney to come to a combined 6-4 record in their last 10. That would have made them unqualified. Is this an outlier to your model and incorrectly shown with the blue shading to represent a correct pick of a champion in 2011 or am I looking at this wrong? I am trying to see if this model worked in previous years.
I’d have to go back in the analysis. There are a forest of conditions. Maybe the other teams tell by the wayside first.
the dis-qualifier for the championship says fewer than 7 wins in the last ten AND worse than a five game winning streak . . . if you are talking about the multi-candidate rules, UConn may have been the only team in their region to meet the criteria that year, so they would not have even had to worry about that dis-qualifier
Aren’t all of these criteria prime candidates for overfitting data?
Yes, Jim–some more than others. But I’ve tried to downplay the Seed Match-up and Upset/Toss-up models, since they overfit on such small sample sizes. Outcome Matching does consciously pick nine upsets based on proximity of competing efficiency data. And others, like KenPom, BPI and Nate Silver, are just their highest probability picks. As I’ve stressed, stats only take you so far.
Pete – on your notes for the Outcome Matching (which I love by the way), you mentioned this would be for pools with 100+ entries. I know this is year 1 for this model, but do you think brackets like this are the way to succeed in large pools given our curve? I would assume risks HAVE to be made in order to win.
Last year, this model did well by picking upsets in the early rounds–but not getting crazy in the later rounds. It also helps to peg the champion. Depending on how things are scored, that’s sometimes the overriding factor.
Can you provide details on the St. Mary’s pick over MSU, then past Creighton…is this a force?