Updated through the
2017 Tournament
Who should you pick? Well, the tables below might be of some
assistance. Feel free to consult them for guidance if you wish. Or
just for enjoyment. Or look at them and wonder when Walt is going
to get a life. One caveat though: using the stats below is
no guarantee that your pool performance will improve; in fact, it
could get worse! (Another caveat: these stats are from my personal
records; I do not vouch for the absolute veracity of the numbers).
In addition to the summary stats below, you can now track the
records of every single team that has made the 64team field since
1985. You can see yearbyyear seedings for each team and
yearbyyear wins. Plus, teams are ranked by various criteria such
as total wins, sweet sixteens, final fours, high seeds, etc.
Check it out at the Tournament Teams Page.
The stats have been compiled since 1985, the year the tournament
went to 64 teams: 33 years total (through 2017). For each of the
pairings in the first four rounds there are 132 results (4
regions/year x 33 years = 132). For the final game there are
64 results (2 teams x 32 years); of course, there have been 32
champions (as of the end of the 2016 tournament).


(actual) 
(and predicted record) 
































Listed in the first table above is the frequency of wins by favorites (higher seeds) over the 132 firstround games between each pairing over the last 33 years. The hardest games to pick are the first round upsets. How hard? For reference, in the right column is a predicted winning % and record over the 132 games by the higher seed based on a very crude model. The model works as follows: Notice that all seed matchups add up to 17 (1 vs. 16, 8 vs. 9). Assume that the seed is an indication of how many times out of 17 that seed would be expected to lose. So, a #8 seed would be expected to lose 8 times (and win 9 times) every 17 games.
Remarkably, this simple model works amazingly well, outside of a couple exceptions. First, a #1 seed has never lost, though the model predicts 8 such losses since 1985; the number of #2 seed losses is also overestimated (16, double the actual 8). Until recently, the largest deviation from expected has occurred for the #8 vs. #9 matchup, where the underdog #9 seed actually won more total games through most of the history, until 2015; since then, the #8 seeds have faired better, winning 7 out of 8 in 2016 and 2017. However, actual results for seeds #3 through #7 are quite close to the statistically predicted results  within ~23% (2 or 3 games over 132 games for each seed pairing). Recently, there has been a skew in the 5 vs. 12 games, especially after 2013 and 2014 when three #12 seeds won each. A 40 run by #5s in 2015 brought things more in line, though #5 seeds still underperform. However, overall the model only errs substantially for the most mismatched teams! One question might be, is this a statistical fluke that just comes out of a few years? Perhaps, but I've updated this for many years now and the percentages haven't changed much from year to year.
Why does the model break down for the 116 and 215 matchups? In
terms of the mismatches, the worst 68 teams (#15, #16 seeds) are
all automatic bids from very weak conferences; these teams are
usually not anywhere near the top 5764 teams (often their rated
in the 100s); so there is much more of a mismatch than the seeds
indicate. Why have the 89 and 512 matchups been off? I have no
idea! (It could be that committee simply misseeds the teams).
So, now you have the necessary information to decide which seeds are most likely to get upset. Of course, knowing that it's likely that at least one #5 seed will lose to a #12 seed this year doesn't help one determine which one of the #5 seeds will be the one to get upset!
Now, what about the second round? See
below:

(of 132 possible) 

(max. 4 times per year) 
































































Looking at the 2nd round winners (those making the Sweet 16), more intriguing things show up. While a #9 seed beats a #8 seed more often than not in the first round, don't pick a #9 to make the Sweet 16. #8 seeds are far more successful in the 2nd round (in absolute numbers even though they get to the 2nd round less frequently than #9 seeds). In fact, getting a #9 seed, while giving a team a chance to win one game, is a virtual guarantee that the team will be gone by the end of the first weekend. You're better off picking a #11, #12, or even a #13 seed to make the Sweet 16.
How does one explain this? A #9 seed,
having just won a very tough first round game, faces a nearly
impossible task of playing a #1 seed with only one day to
practice for them. Lower seeds face an easier 2nd round opponent
and have momentum and perhaps a psychological advantage (the
Cinderella effect) on their side. Often, there may be multiple
upsets, so that a #12 seed actually becomes a favorite against
#13. In fact, once the #12 seed wins their first round game,
close to 50% of the time (20 out of 46 through 2016) they've won
their next game and gone to the Sweet 16. Picking a #15 or #16
is generally throwing money away, but in 2013 FloridaGulf Coast
became the first #15 to make the Sweet Sixteen. And now, the
third round and beyond:

(of 132 possible) 

(max. 4 times per year) 

















































(132 total) 
(66 total) 
(33 total) 











































