## Sports

Every year since his first presidential campaign, Barack Obama has shared his NCAA men’s college basketball tournament bracket with the public, and, for the last five years, predictions for the women’s tournament, too. As he’s chosen teams, many of them from swing states, he’s resisted the urge to play politics with his picks.

Obama’s main bracket criterion: the number next to a team’s name. The lower a team’s seed, the more likely he is to overestimate its chances. The president has predicted a far smaller number of round-of-64 upsets than the tournaments have produced, and in the last seven tournaments he’s forecast just one men’s team with a seed below 9 to win more than one game.

He also has a soft spot for certain states. The president loves teams from Connecticut, a reliably blue state, and Kentucky, a reliably red one.1 He’s been pretty neutral on states where he’s lived, studied or had family ties, including Illinois, California, Kansas, Massachusetts and New York. (Hawaii hasn’t had a tournament team in the years of Obama’s brackets.)

Other public figures have picked brackets, but Obama, a big basketball fan, is unusual in having entered so many of his bracket predictions into the public record2 — enough to create a sample size bigger than some of the polls used to forecast his election and re-election, and to search for patterns in his picks.

To assess Obama’s brackets, I compiled the number of wins he predicted for each tournament team in each season,3 and compared that number to two benchmarks: How did his picks compare to what might be considered typical picks, and how did they compare to how teams actually performed?

The first benchmark was the average number of games that teams with the same seed had won for each tournament, over a period from the first year the tournament expanded to 64 teams4 through the year before he picked the bracket.5

This measure models the sort of information used by a typical tournament forecaster (or as typical as one in the White House can be): How well have teams of that seed done before?

The second benchmark for Obama’s picks was the one used to score brackets for accuracy: how many games the teams won. This is both less and more fair than the first benchmark — less fair because he couldn’t have known how the teams would do when he submitted his picks, and more fair because it credits him for insights beyond the seed numbers.

After subtracting either number — expected wins or actual wins — from Obama’s predicted win total for each team, we’re left with two possible measures for his lean toward or away from that team. When Obama picked Louisville to make the final last year as a top seed, the first measure scored that as a big pro-Louisville preference, since the average No. 1 seed from 1985 to 2012 averaged 3.375 wins. But the second measure detected a presidential slant against Louisville, since the Cardinals won the title. Conversely, Obama looks like a Washington, D.C., outsider based on his picks for the capital’s teams: He expected fewer wins for them than their seeds would have suggested. But his picks proved optimistic when the teams underperformed their seeds by a big margin.

I now had a set of over 700 teams, each one with two scores indicating whether Obama was hard or easy on each team. I then looked up each team’s home state and ran a series of linear regressions to find whether politics could be driving the president’s picks.

My first test: Was Obama backing the states that were most supportive of him, or — for his first bracket — the prior Democratic presidential candidate, John Kerry in 2004? Or, instead, was he throwing some love to the states that were most supportive of Republican candidates, hoping to sway their hoops-mad voters into his camp?

Neither, best I can tell. For each year, I took the most recent presidential election data6 and subtracted the percentage of votes going to the Republican from the percentage received by the Democrat, then normalized the results.7 I then ran two regressions against this score, one for each of my scores of Obama’s picks. And I found no relationship whatsoever.

The story repeats for other political indicators that might have steered his picks: whether states were swing states,8 and the probability that a single voter in that state — perhaps a fan of a team Obama could pick for the Final Four — could swing the presidential election.9 For each possible factor, I ran the same pair of linear regressions.10 And each time I found no clear relationship.11

There was one highly statistically significant relationship: between Obama’s picks and a team’s seed number. For each improvement in seed of one — say, from 5 to 4 — Obama was likely to give the team a bump of about 3 percent of a win relative to its seed’s expected wins, and about 4 percent of a win more than the team actually won.12

Here’s another way of saying that: President Obama backs favorites to win even more than they have historically. And he’s remained consistently risk-averse, ranging from three to six upset picks in the round of 64 in his seven men’s tournament brackets. He’s never picked a team seeded below 13 to win a game, though six have over those years.13

Obama predicted wins for 32 teams seeded 10 and below, from the round of 64 on, for the seven men’s tournaments from 2008 to 2014.14 Yet 50 percent more teams have won at least one game. He was especially downbeat about the chances of teams seeded 12 or lower, predicting just eight wins for the group. Some 29 teams seeded that low have combined to win 38 games.

Obama’s caution intensifies as he moves through the men’s bracket to later rounds. He’s picked just one team seeded 7 or worse to make the Sweet Sixteen in seven years of men’s bracket-picking.15 Yet 27 teams with seeds that low made a Sweet Sixteen since 2008 — including two teams, Connecticut and Kentucky, that qualified for the Final Four on Sunday. Conversely, though nine No. 2 seeds have lost before the Sweet Sixteen, Obama has picked every one to make it that far.16 Obama has backed no Elite Eight teams with seeds worse than 5, yet nine such teams have made it that far. And he’s picked no national semifinalist seeded worse than 4, yet seven Final Four teams have fit that category.

When Obama has predicted a men’s upset, he has guessed well. By chance alone, you’d expect that 14 of the teams he picked seeded 10 or below to pull off at least one upset of a higher-seeded team would have done so.17 Yet 18 got at least one win — including all five of the 12 seeds he backed. Obama is on the verge of displaying a statistically significant forecast skill in the men’s brackets.18

In the women’s tournament, where favorites tend to dominate, Obama’s caution has been merited. He’s picked 12 teams seeded 10 or worse to pull off first-round upsets since 2011,19 and 13 have. He’s shown no particular ability to identify upset victims, forecasting three correctly compared to an expected total of 2.4.

Obama also seems to like certain teams more than others, though with just 12 of his brackets on record, no team has a sample size large enough to draw broader conclusions. His likes, relative to expected and actual performance, include Baylor, Kentucky, Louisville, Marquette, North Carolina, Notre Dame and Ohio State. By contrast, Obama is hard on Arizona, Gonzaga, Oklahoma, San Diego State, Texas and Xavier.20 His view on some schools looks different depending on the measure: Obama seems like a Duke-backer based on the Blue Devils’ results, but his picks have been in line with their seed numbers.

Separating his preferences for certain schools’ men’s and women’s teams is especially tough because of the sample-size problem. So it’s hard to say whether he likes Cal’s women’s teams but dislikes their male counterparts, or if that’s just a statistical fluke.

Asked about the president’s picking strategy, the White House press office referred to his statements to ESPN when unveiling his brackets.21 Obama usually talks about specific teams in those broadcasts, rather than a broader strategy. The country’s chief executive has a soft spot for team chiefs, mentioning more coaches than players in recent years. (He has high praise for Michigan State’s Tom Izzo and North Carolina’s Roy Williams; in 2012, he admitted, “I’m just a sucker for the Tar Heels.”) He likes point guards and teams with momentum. Last year, he mentioned his “Big 10 bias.”22 When Obama does mention a player, he is sometimes motivated by where the player comes from rather than where his school is. For example, he said he picked Duke to reach this year’s Elite Eight partly because Blue Devils star Jabari Parker comes from Chicago. And Obama is aware of his tendency to back favorites. He said this year, “I know these are not imaginative picks, but I think they’re the right ones.”

It’s hard to argue with the president’s preference for favorites: Picking upsets incorrectly is more damaging than picking upsets correctly is valuable. And lately, it’s working for him. He picked the women’s champion correctly twice in four tries, got one of the surprising men’s Final Four teams right this year — No. 1 seed Florida — and is in the 74th percentile of ESPN’s bracket contest. However, he may regret taking the relatively daring step of backing a No. 4 seed, Michigan State, to win the title. It was his first time picking a men’s or women’s champion that wasn’t a No. 1 seed, and the Spartans’ elimination on Sunday left him without any chance of gaining further points next weekend. In presidential brackets, as in presidential politics, risk-taking sometimes backfires.

## Footnotes

1. Schools from those states get a bump of about 0.7 wins per year in his brackets, compared to the expected performance of their seed numbers. The effect is statistically significant (p<0.05). But that may just be a reflection of Obama’s preference for highly seeded teams: After controlling for seed, no state had a statistically significant effect on the president’s picks. ^
2. He usually has unveiled them on ESPN broadcasts. ^
3. There are two groups of exceptions to this list.

First, neither ESPN nor the White House was able to supply a copy of Obama’s 2010 women’s bracket, and the link from a Whitehouse.gov blog post about it points to a different ESPN.com bracket. I gathered as much information as I could from press accounts of his Final Four picks that year, plus whatever I could glean from this video clip of his ESPN interview, and excluded from all analyses the 21 women’s teams in the 2010 tournament for which I couldn’t figure out the president’s prediction.

Second, I excluded from the analysis any teams that hadn’t yet lost in this year’s tournaments, since we don’t know their final win totals.

Here are links to Obama’s 2008, 2009, 2010, 2011, 2012, 2013 and 2014 men’s brackets, and to his 2011, 2012, 2013 and 2014 women’s brackets. ^

4. 1985 for men, 1994 for women. ^
5. That calculation was based on data provided by ESPN Stats & Information. I counted only wins from the round of 64 on, since the brackets Obama entered didn’t require entrants to predict play-in games. I assigned each play-in team half its seed’s expected wins, since only half the play-in teams advance to the round of 64.

Obama, incidentally, has never picked the play-in winners to win their next game (a questionable strategy), and generally hasn’t picked the outcome of the play-in games, either, though he did write on his 2009 bracket — incorrectly, as it turned out — that Alabama State would beat Morehead State and enter the round of 64 as a No. 16 seed. ^

6. Election data from uselectionatlas.org. Since March precedes November, for Obama’s 2008 tournament picks, the most recent election was 2004; for 2012, it was 2008. ^
7. For each election year’s data, I subtracted from each state’s figure the average of every state’s figure. This data normalization put all states for each election year on the same playing field: how far they leaned Democratic or Republican relative to the average state in that year. ^
8. I defined swing states as those with a gap of less than 5 percentage points in the previous presidential election between the vote shares of the Democratic and Republican candidates. ^
9. Based on work by Columbia University statistician Andrew Gelman and FiveThirtyEight editor-in-chief Nate Silver, using Silver’s 2008 presidential-election forecasts; Gelman provided me with a table of probabilities by state. ^
10. Technically, I ran two pairs of linear regressions for the probability of one voter swinging the election: One using the raw probability, and one the logarithm of the probability, since the probabilities were minuscule and varied by orders of magnitude. ^
11. Six out of the eight coefficients were positive, which would suggest Obama was favoring Democratic states or electorally vital states, but none of the results was statistically significant. ^
12. Seed was highly significant: p<10^-6. After controlling for seed, all the political factors still produced insignificant effects and half their coefficients were negative, suggesting any hint of a lean by Obama was more a product of seed number. And for each regression, p>0.4. ^
13. It’s probably imprudent to pick too many early upsets by big underdogs since their opponents are favorites not only to win but to go on to advance far in the tournament. But many fans do pick at least one. The millions of entrants to ESPN.com’s bracket challenge this year and last averaged about one pick per three brackets of a 14, 15 or 16 seed to reach the round of 32. ^
14.  I didn’t count wins by 9 seeds over 8 seeds as upsets since those teams are so closely seeded. ^
15. North Carolina State, in 2012. Obama was right. ^
16. He evidently considered choosing Clemson to upset No. 2 Oklahoma in 2009’s round of 32 but crossed out that pick and went with the chalk — correctly, as it turned out. ^
17. That calculation is based on the actual rates of upsets for each seed number, and how many upsets he predicted. ^
18. 0.1>P>0.05. ^
19. I excluded the incomplete 2010 Obama women’s bracket from this analysis, in case his picks for the omitted teams deviated sharply from his picks that we know. ^
20. The effect for Baylor, Louisville and North Carolina is statistically significant (p<0.05) for his picks relative to the teams’ seed numbers; he typically gives those teams one more win than their seed suggests. The effect vanishes, though, for those schools when examining how they did relative to Obama’s picks. In other words, their results suggest he was mostly right when picking them to outperform their seed. The effect is also smaller and statistically insignificant after controlling for seeds, since those teams tend to have high seeds. Adding that control reveals that Obama has been hard on Washington State, to a statistically significant degree: Controlling for seed, Obama has under-picked the Cougars by an average of nearly two wins per tournament, relative to their seed. ^
21. Here are video clips of Obama’s chats with ESPN’s Andy Katz about his 2011, 2012, 2013 and 2014 brackets. ^
22. Obama does back Big 10 teams beyond their seed to a statistically significant degree, but the effect isn’t significant after controlling for seed, since Big 10 teams usually are highly seeded. Because of all the conference realignment during Obama’s bracket-picking years of 2008 to 2014, I studied only the conference he mentioned liking and counted only Big 10 teams that were in the conference throughout the period. ^

Carl Bialik is FiveThirtyEight’s lead writer for news.

All Sports

### The Republican Establishment Is Waiting On The SidelinesSep 3, 2015

Never miss the best of FiveThirtyEight.