Two years ago, I filled out one of my NCAA brackets using nothing but team pythags. The thing actually placed second in the Rocky Top Talk Bracket Challenge, so I figured I'd put the concept to the test again to see what happens this year.
As a refresher, here's the explanation I offered back in 2008:
After staring at A Sea of Blue's post setting forth the odds for each of the NCAA Tournament teams to win it all for a half an hour, I concluded: whoa, numbers. But once I recovered, I thought it would be interesting to see how each game of the bracket plays out using Ken Pomeroy's data. Below is the NCAA Tournament bracket, with each region on top of each other (as opposed to the four regions facing each other in groups of two) because who among us has a 34" computer monitor? The play-in game is on top and the final four is on the bottom. The four digit number immediately next to each team is that team's pythag, and the percentages next to the pythags are those teams' respective "chances of winning" that game. For instance, based on their respective pythags, Mt. St. Mary's should beat Coppin State 88.85% of the time and so we'll consider them the winner and move them forward to play North Carolina. Yes, you can get the same result by looking only at the teams' respective pythags, but I think the percentages give you a better feel for how those numbers might play out on the court.
A couple of pre-post observations from Hooper:Pythag uses "pace-neutral" weighting, which means it compares the ratios of point differentials but not the actual values of point differentials. For example, a team that averages 80 pts. for and 70 pts. against looks identical to a team that averages 64 pts. for and 56 pts. against. KenPom does this so he doesn't have to worry about how many possessions a team normally has in a game. But that does make a big difference in the play of a game. So you lose valuable data at the start. This is normal for every numerical method, but it's good to know what is being lost. What this means is that Kansas has had the best pts. for / pts. against performance per possession in the league. But we don't know how many possessions is ideal for Kansas, or even if it makes a difference.
We also don't know the uncertainty. For example, if a team has a "70% chance of winning", does that mean 70% +/- 15%, or 70% + / 5%? If it's the first, then it's not unreasonable to see an upset. If it's the second, then an upset would be a tremendous shocker. We don't have a feel for the significance of a point spread, in other words.
If anyone wants to update the table to account for any of that, have at it. Start here, then read this, then read this. In the meantime, though, have a look at the table below, all dressed up in pretty Easter pastels for your enjoyment.
Oh, and one more thing: I've entered this bracket into the RTT ESPN Tournament Challenge so we can keep track of how well it does. If anyone calls Ken Pom an idiot because this entry doesn't finish in the top spot, I will personally come over to his house, pull out his toenails with a pair of pliers one by one every hour on the hour, write "no, you're the idiot" on each one in pink fingernail polish, and feed them to him. And Jackson the Mule will be right behind me to finish you off.
Okay, so that was two years ago. Here's this year's pythag bracket picks, subject as always to over-tired operator error:
Yeah, a couple of eyebrow-raisers, huh? No real surprise in the Midwest, but 7-seed BYU making the Elite Eight in the West? So says the pythag. How about 2-seed West Virginia and 4-seed Wisconsin in the East? And Duke winning it all? Is this thing crazy? Have a look at the data from Ken Pomeroy's site and decide for yourself.
Any pythag skepticism aside, there are always upsets, right? Right. So I have also created a modified pythag bracket to account for teams CBS's Peter Tiernan has identified as fitting the upset profile. That post goes live this afternoon. I think Tennessee fans will be pleased.