FiveThirtyEight’s NCAA Tournament Predictions | Hacker News Reader

FiveThirtyEight’s NCAA Tournament Predictions

fivethirtyeight.com

54 pointsgmu318 comments

What's often quoted in Warren Buffet's Billion dollar NCAA contest are the odds of randomly picking the perfect bracket of teams, but I have to think that Buffet's insurance actuaries actually calculated the odds of randomly selecting the right bracket with a millions of random guesses around a locus of likely guesses informed by analysis like the posted page (and from other stats analysis professionals.. vegas oddsmakers for example...). Otherwise the risk of taking that bet is actually much higher than a fully random pick, yes?

Edit: As it turns out, Warren Buffet, as usual, seems to have done his homework and even adds the interesting angle of "buying out" a possible winner partway through the contest:

http://www.slate.com/articles/sports/sports_nut/2014/03/bill...

Similarly, this NPR article discusses that a bit: http://www.npr.org/2014/03/15/290283283/why-you-wont-win-war...

I think I read somewhere that the fine-print states one entry per household, so no brute forcing :)

Silver's round-by-round projection percentages are included in the HTML source. If you like building rule-specific points projections, here's the data, extracted:

https://gist.github.com/ajstiles/06e2ac52927d3f292255

I find the discussion of their model much more interesting than the results: http://fivethirtyeight.com/features/nate-silvers-ncaa-basket...

A #16 seed has never beaten a #1 seed in the history of the tournament, yet the percentages of moving on for #1 seeds against #16 seeds vary from 96%-99%.

I wonder if there would be any of this variance (4 percentage points from 100%) if computers, without human input, were used to place the likelihood of moving on.

There was a big variance in the regular season performances of the #1 seeds this year, which is what the computers are picking up on. If anything, I think humans have the opposite effect of what you suspect, as they do things like overrating Wichita St for emotional reasons (undefeated record) while the most famous computer ranking http://www.usatoday.com/sports/ncaab/sagarin/ puts Wichita St as only the 13th best team.

Also, historically #1 seeds were stronger than they are now because top players didn't go to the NBA so quickly.

I would say the exact opposite, really. Players can't jump straight from high school anymore. So you have a larger base of extremely talented freshmen.

Also, I would consider the Pomeroy rankings to be better (many more variables, Sagarin uses point differentials which are silly).

There were only like 5-6 seasons when it was popular to go straight to the NBA from high school. In the last millennium, it was very uncommon. Instead, it was typical for top players to stay for 3-4 years. Now, they usually stay for just 1. Compare:

Mostly freshman/sophomores: http://en.wikipedia.org/wiki/2013_NBA_Draft#Draft_selections

Nearly all juiniors/seniors: http://en.wikipedia.org/wiki/1991_NBA_Draft#Draft

(EDIT: Christ, look at UNLV in that '91 draft. Seniors from that team were picked #1, #9, #12, #29. That's completely unfathomable today.)

Along the lines of UNLV, check out the 2005 draft with UNC. #2, #5, #13, & #14. I think that's the most lottery picks from one school in a year, and three were in the same class (#2 was a freshman)!

> Also, historically #1 seeds were stronger than they are now because top players didn't go to the NBA so quickly.

Would love to see some data on that. My perception that it's been the same since the early 2000s.

I'd like to see data too, but I don't have it :)

My perception agrees with yours, I was thinking of the 80s and 90s when it wasn't uncommon to see superstars hone their skills for 4 years in college.

Just because it's never happened doesn't mean that it won't happen, just as the percentage's suggest it's very unlikely.

I think TheBiv's point is that, even though you know it can happen, historical data suggests that 1-4% is too high a probability. It's never happened in, say, 20 years of a 64-team tournament? That's 80 #1 seeds? Then 1% might be a reasonable probability, but 4% does not seem reasonable.

You said it better than myself!

Is this the first year Nate and his team have modeled the NCAA Tournament? I'd be curious to know if they applied a similar statistical approach last year and, if so, how that model faired against the actual tournament results.

Florida has a higher percentage of winning every round than Louisville, but Louisville has a higher overall chance of winning the tourney.

What am I missing?

They don't have as high a percentage of winning the final round - 15/24 vs 14/26

Reply on news.ycombinator.com