#1 seeds being bumped off by a #16

[davekay1971]

Time to start playing the "no respect" card.

ESPN.com's front page currently with a picture of Sheed looking glum and K looking ticked with the headline "Cinderella's Revenge. Could this be the year a No. 16 seed takes out a No. 1? We're looking at you, Duke." The article in an inside access one, which I don't pay for, but the headline should be a banner to the Duke players: you just got called overrated.

I'm not sure what in the world would make ESPN think Duke, with Ryan back, is particularly vulnerable to a staggering upset compared to the other current projected 1 and 2 seeds. But there we are. Print it out and wallpaper the locker room it, Coach.

[Merlindevildog91]

Time to start playing the "no respect" card.

ESPN.com's front page currently with a picture of Sheed looking glum and K looking ticked with the headline "Cinderella's Revenge. Could this be the year a No. 16 seed takes out a No. 1? We're looking at you, Duke." The article in an inside access one, which I don't pay for, but the headline should be a banner to the Duke players: you just got called overrated.

I'm not sure what in the world would make ESPN think Duke, with Ryan back, is particularly vulnerable to a staggering upset compared to the other current projected 1 and 2 seeds. But there we are. Print it out and wallpaper the locker room it, Coach.

Having paid absolutely no attention to ESPN's bracketology other than what I saw on Twitter, my first thought in seeing this was, "When did they finally decide we were deserving of a number 1 seed?"

[BlueDevilBrowns]

Time to start playing the "no respect" card.

ESPN.com's front page currently with a picture of Sheed looking glum and K looking ticked with the headline "Cinderella's Revenge. Could this be the year a No. 16 seed takes out a No. 1? We're looking at you, Duke." The article in an inside access one, which I don't pay for, but the headline should be a banner to the Duke players: you just got called overrated.

I'm not sure what in the world would make ESPN think Duke, with Ryan back, is particularly vulnerable to a staggering upset compared to the other current projected 1 and 2 seeds. But there we are. Print it out and wallpaper the locker room it, Coach.

Well, I just read the article you mention and to briefly summarize, it uses statistical models to "prove" that Duke is the most likely of potential #1 seeds to be defeated this year by a #16 seed. They give Duke an 85% chance of beating a #1 seed, significantly lower than all other current top teams except one other.


What's interesting to me is in the comments section, a reader calls out ESPN for not including Duke's metrics With-Kelly and Without-Kelly. An ESPN representative responded by saying "we're trying to figure out whether our model can do a with-Kelly/without-Kelly thing at some point."


How lame.

I'm not a stat-head or a computer expert but I'm pretty sure it's not very hard to simply delete the 13 games Kelly didn't play from the data and compare the results.

[toooskies]

A telling quote from the article on the ESPN home page's linked article:

Our dream scenario: Canisius storms out of the MAAC to take on Duke, where the Golden Griffins would have a 41.5 percent chance of pulling an all-time shocker.

Any system that proposes that an 18-12 team from the MAAC conference has a 40% chance of beating a 26-4 team from the ACC is worthless. Except not literally worthless; it's designed to sell ads.

Do not be mistaken, writers don't always write to tell the truth; some write to get a reaction. This is mostly the second. (It may have a nugget of truth in it; but that truth is simply, Duke is a "Giant" nearly every year, and we have similar statistical profiles every year. If we lose early, it significantly affects the data.)

[Kedsy]

I'm not a stat-head or a computer expert but I'm pretty sure it's not very hard to simply delete the 13 games Kelly didn't play from the data and compare the results.

Actually, it's got to be WAY more complex than that. First of all, the rest of the team's performance (without Ryan) shouldn't be 100% discounted, second of all, most good ratings systems connect every game to every other game in myriad ways. How can you adjust the ratings for competition if you simply eradicate 13 games from the system? And even if you have an answer for that, my guess is most computer models won't have an easy way to remove games like that anyway. Finally, if you take away those 13 games, you take away almost all of Duke's conference schedule, and you'd be comparing apples-to-oranges with Duke's mostly non-conference schedule against most teams full schedule including conference games, and the overall reliability of the ratings would go down.

As far as what chance Duke has to be upset by an yet-to-be-determined #16 compared to other #1 seeds, it would seem to hinge on their computer ranking compared to the other #1s. Duke is currently #1 in RPI and #2 in BPI (behind Louisville, who most people aren't talking about as a #1 seed anyway). In Pomeroy and in Sagarin, Duke is #6, behind Indiana, Gonzaga, and Kansas, so if that was the system used than you'd think the odds would be highest on Duke losing to a yet-to-be-named opponent. Seems weird that ESPN would use Pomeroy rather than its own rating system, but that must be what they're doing, right? Even then, without seeing the analysis, 85% sounds off.

[hurleyfor3]

Without Kelly and as a #1, I would in fact agree. Not that we'd lose to a 16, but that we'd have the highest probability of doing so. Although in the real world we would likely drift down to #3 or 4 territory.

[JasonEvans]

One thing worth pointing out is that as the NCAA tournament continues to expand and make the #16s play in "play-in games," it raises the chances of a #1 seed being upset because they teams who end up occupying the #16 seeds get better and better.

Before the advent of the play-in game in 2001, the #16s were the 4 worst teams in the tournament. If you were a #1 seed then, at best, you were playing one of the bottom 4. Then from 2001-2010, they added one play in game and the #16s became the 5 worst teams in the tournament. At that point one of the #1 seeds was forced to play a team that would have been a #15 seed in the past. Since 2011, we have had even more play-in games, two of which involve the bottom 4 teams in the tourney. This means that two of the #1 seeds are playing teams that would have been #15 seeds in the past. The other two #1 seeds are playing teams that were good enough to beat a #16 seed, an indication that they may be better than anticipated.

I think the dual play-in games has created a scenario where an upset of a #1 seed is inevitable and will happen some time soon. We have raised the quality of the teams playing the #1 seeds, perhaps significantly. As a result, the odds of an upset creep higher, even if it is only by a tiny bit.

For 27 years we have had #16 seeds playing #1 seeds. The #1 seeds are 108-0 in those games. While I am quite familiar with the rules of statistics that say past results do not affect the odds of a future event happening (flip a coin 10 times and even if it has hit heads 9 out of ten, the odds of a heads on the next toss are still 50-50), I tend to think that the fact that we have gone 108 games without the ultimate upset perhaps says we are "due" for it to happen. We stack the odds against the #1 seeds more and more each year. It will happen someday.

-Jason "that said, 85% chance that Duke will lose to an unknown #16 seed... that seems a tad crazy" Evans

[hurleyfor3]

For 27 years we have had #16 seeds playing #1 seeds. The #1 seeds are 108-0 in those games. While I am quite familiar with the rules of statistics that say past results do not affect the odds of a future event happening (flip a coin 10 times and even if it has hit heads 9 out of ten, the odds of a heads on the next toss are still 50-50), I tend to think that the fact that we have gone 108 games without the ultimate upset perhaps says we are "due" for it to happen. We stack the odds against the #1 seeds more and more each year. It will happen someday.

Actually 112 such games in 28 tournaments.

For there to be a 50% chance of no 16-over-1 upsets in 112 games, the chance of the #16 winning an individual game must be less than or equal to 1 minus the 112th root of one-half. This is .00617, meaning the (geometric) average 16 seed has less than a .617 of 1% chance of winning. This sounds about right to me.

[/math geek]

[SilkyJ]

Time to start playing the "no respect" card.

...The article in an inside access one, which I don't pay for

Maybe I'm so numb to all the Duke hate at this point that I'm seeing things upside down, but I don't think this is a fair example of the typical duke hate/no respect card. I'm an insider and read the article, and as others pointed out they ran a quant/statistics model and the computer said we are the most vulnerable "giant" i.e. potential 1 seed. Each year after the brackets come out, ESPN runs this model (or at least something very similar) and lists out their most likely upsets. This is the same thing, except run a few weeks early. No biggie.

And the computer clearly does not have a high enough WRC (white raven coefficient). We are undefeated with Ryan Kelly. I don't expect that to change any time soon--as in this year

[tommy]

One thing worth pointing out is that as the NCAA tournament continues to expand and make the #16s play in "play-in games," it raises the chances of a #1 seed being upset because they teams who end up occupying the #16 seeds get better and better.

Before the advent of the play-in game in 2001, the #16s were the 4 worst teams in the tournament. If you were a #1 seed then, at best, you were playing one of the bottom 4. Then from 2001-2010, they added one play in game and the #16s became the 5 worst teams in the tournament. At that point one of the #1 seeds was forced to play a team that would have been a #15 seed in the past. Since 2011, we have had even more play-in games, two of which involve the bottom 4 teams in the tourney. This means that two of the #1 seeds are playing teams that would have been #15 seeds in the past. The other two #1 seeds are playing teams that were good enough to beat a #16 seed, an indication that they may be better than anticipated.

I think the dual play-in games has created a scenario where an upset of a #1 seed is inevitable and will happen some time soon. We have raised the quality of the teams playing the #1 seeds, perhaps significantly. As a result, the odds of an upset creep higher, even if it is only by a tiny bit.

For 27 years we have had #16 seeds playing #1 seeds. The #1 seeds are 108-0 in those games. While I am quite familiar with the rules of statistics that say past results do not affect the odds of a future event happening (flip a coin 10 times and even if it has hit heads 9 out of ten, the odds of a heads on the next toss are still 50-50), I tend to think that the fact that we have gone 108 games without the ultimate upset perhaps says we are "due" for it to happen. We stack the odds against the #1 seeds more and more each year. It will happen someday.

-Jason "that said, 85% chance that Duke will lose to an unknown #16 seed... that seems a tad crazy" Evans

Jason, either I'm misunderstanding what you're saying, or else I think you may be misunderstanding, or more likely inadvertently misstating the way this works.

The expansion to a 68 team tournament means this: Teams 65 through 68 play each other, and the two winners advance to face #1 seeds. So those two #1 seeds play teams that wouldn't have made the top 64, meaning they're the equivalent of #17 seeds. Yes, because those teams would've received automatic bids anyway, they would've acutually been #16 seeds in the old days, but nevertheless they are teams not considered by the committee to be in the Top 64 (i.e. the "real" top 16 seeds) in this tournament.

The other two #1 seeds play teams designated as #16 seeds as well, so they're teams 63 and 64. Basically, there are six #16 seeds, and the bottom four of them play off to get into the field. But no #1 seeds have to play a #15 seed, or any team rated by the committee as being overall #s 59 through 62, which correlates to a 15 seed. Note: normally, the #15 seeds would be overall #s 57 through 60, but everyone gets bumped down two spots because of the peculiar way this tournament is structured, with six #12 seeds instead of four.

For the same reason, when the tournament had a 65 team field, none of the #1's played a team that would've been a #15. They all played a 16, or perhaps the lone true #17 seed (the #65 overall team).

All that being said, I do agree that the chances of a #1 losing its opener are rising, but I think it's because the small conference teams who are those #16s are becoming better and better, while the elite teams (the #1's) are not as good as they were in years past. Or even close.

[-jk]

Actually 112 such games in 28 tournaments.

For there to be a 50% chance of no 16-over-1 upsets in 112 games, the chance of the #16 winning an individual game must be less than or equal to 1 minus the 112th root of one-half. This is .00617, meaning the (geometric) average 16 seed has less than a .617 of 1% chance of winning. This sounds about right to me.

[/math geek]

I was there for Mississippi Valley State. Whew!

It almost happened to us. It will happen to someone someday. Just hope it isn't us - but if we continue get more 1 seeds than most (a nice problem to have!), odds are...

-jk

[toooskies]

Jason, either I'm misunderstanding what you're saying, or else I think you may be misunderstanding, or more likely inadvertently misstating the way this works.

The expansion to a 68 team tournament means this: Teams 65 through 68 play each other, and the two winners advance to face #1 seeds. So those two #1 seeds play teams that wouldn't have made the top 64, meaning they're the equivalent of #17 seeds. Yes, because those teams would've received automatic bids anyway, they would've acutually been #16 seeds in the old days, but nevertheless they are teams not considered by the committee to be in the Top 64 (i.e. the "real" top 16 seeds) in this tournament.

The other two #1 seeds play teams designated as #16 seeds as well, so they're teams 63 and 64. Basically, there are six #16 seeds, and the bottom four of them play off to get into the field. But no #1 seeds have to play a #15 seed, or any team rated by the committee as being overall #s 59 through 62, which correlates to a 15 seed. Note: normally, the #15 seeds would be overall #s 57 through 60, but everyone gets bumped down two spots because of the peculiar way this tournament is structured, with six #12 seeds instead of four.

For the same reason, when the tournament had a 65 team field, none of the #1's played a team that would've been a #15. They all played a 16, or perhaps the lone true #17 seed (the #65 overall team).

All that being said, I do agree that the chances of a #1 losing its opener are rising, but I think it's because the small conference teams who are those #16s are becoming better and better, while the elite teams (the #1's) are not as good as they were in years past. Or even close.

I think the confusion lies in referencing them as 16 seeds. Instead, think of the four worst automatic-bid conference winners. The quality of the automatic conference winners doesn't change based on how many at-large teams are invited to the tournament to fill up the bracket. In the days of a 64-team bracket, those four teams would all play against 1 seeds. In a 68-team bracket, two of those four teams play the 1 seeds (and they're coming off a win), and the 5th and 6th worst conference winners are playing 1 seeds instead of the teams which were eliminated.

[A-Tex Devil]

Wasn't there a statistic last year that Nicholls St beating Mizzou was as big as any 16 over 1 seed upset would have been? Or maybe it was less statistically probable than any of the 16s beating any of the 1s last year. Don't remember.

But that was essentially as close to a 16-1 (statistically) as we've seen.

[sagegrouse]

I think the confusion lies in referencing them as 16 seeds. Instead, think of the four worst automatic-bid conference winners. The quality of the automatic conference winners doesn't change based on how many at-large teams are invited to the tournament to fill up the bracket. In the days of a 64-team bracket, those four teams would all play against 1 seeds. In a 68-team bracket, two of those four teams play the 1 seeds (and they're coming off a win), and the 5th and 6th worst conference winners are playing 1 seeds instead of the teams which were eliminated.

The automatic qualifiers, by definition, are the bottom of the field. It doesn't matter how many at-large bids are given, and the number of automatic qualifiers has been fixed at 31 (I believe) for the past few years.

With the scheme at work now, the four #1 seeds will play two #15 seeds and two #16 seeds, based on a 64-team field and assuming the better teams advance from the first round. Here's why:

The outcome of the 68-team field is that the four #1 seeds play four of the worst six of the automatic qualifiers. Of these six, two get a bye and the other four play a first-round game to advance to the second round, reducing these six teams to four. If the two better teams win every time (not likely), then the #1 seeds are, in effect, playing teams that in a 64-team field would have been two #15's and two #16's.

Everybody got it? End of discussion?

It will be a final exam question.

sagegrouse

[theschwartz]

My biggest takeaway from the article: ESPN is getting desperate for page hits and trying to get readers to sign up for Insider. There's definitely a mismatch between the alarmist headline/picture and the sobering, data-heavy content of the article. I'm guessing the writer of the article and the writer of the headline were not the same person. Probably the headline writer read the article, saw that the model gives Duke the highest chance of falling to a 16-seed, got excited, and plastered a negative Duke headline to give all the Duke-lovers and Duke-haters--pretty much the entirety of college basketball fandom--a reason to click and/or subscribe to Insider. Honestly, it's savvy marketing. Would as many people have clicked the link if it was Gonzaga or Georgeotown on the cover? Definitely not. What about Kansas or Indiana, two great but also flawed teams? Probably not.

I would also love to see the with/without-Kelly numbers. I'm sure you'll be looking at two completely different teams and their prospects in the Tourney based on that analysis - the dichotomy is significant enough that it's worth the effort for one of ESPN's myriad analysts. And honestly, if you're determined to say that Duke--or any team really--is the weakest potential 1-seed, it's a cinch to rig your model to push that argument and say that there's a good chance that a 1-seed will lose in the 1st round. I'm very curious what their "secret sauce" is. Probably a metric to see what draws the most traffic. As an ex-Excel jock, I can definitely say that if I wanted to drive to a certain conclusion, I could easily build a model with the right variables that gets me there.

The article states that there's less than a 15% chance that Duke gets upset in the 1st round (and that's including the without-Kelly results). That's still pretty low. The bottom line is, A LOT of things would have to go wrong for Duke for that to happen. For Duke to have a 14.3% chance of losing to a 16-seed and a 41% chance for Canisius to beat Duke, the independent probability of Canisius beating a 1-seed would have to be really high, even when you factor in secret sauces and style of play and all those other variables that the writers probably used. Again, you can get the numbers to tell you anything if you try hard enough and don't have to show your model to anyone.

[JasonEvans]

I think the confusion lies in referencing them as 16 seeds. Instead, think of the four worst automatic-bid conference winners. The quality of the automatic conference winners doesn't change based on how many at-large teams are invited to the tournament to fill up the bracket. In the days of a 64-team bracket, those four teams would all play against 1 seeds. In a 68-team bracket, two of those four teams play the 1 seeds (and they're coming off a win), and the 5th and 6th worst conference winners are playing 1 seeds instead of the teams which were eliminated.

Exactly.

Tommy, you are looking at it from top down and thinking that rather than inviting the top 64 teams, we now invite the top 68. As tooksies points out, my way of looking at it is to go from the bottom up... because no matter how large you expand the at-large field, the bottom is still the bottom. That bottom is made up of automatic qualifiers who will always be there and the quality of those teams cannot be changed by the number of automatic qualifiers. So, rather than playing the 4 worst automatic qualifiers, the #1 seeds are now playing 4 of the 6 worst automatic qualifiers. What's more, they are assured of playing the 2 best of those 6 who are not in the play-in games... and the other 2 they play are teams that won a play-in game thereby indicating they may be stronger than anticipated. I think a strong case can be made that the quality of the teams facing the #1 seeds is significantly higher than 4 or 6 or especially 10 or 20 years ago.

Yes, as you point out there are now 68 teams in the tournament and, in theory, these teams at the bottom are therefore seeded lower (some of them are de-facto #17 seeds) than the #16 seeds of the past. But, that is because we have grown the middle of the tournament and does not have a practical impact on the quality of the bottom 12 or so teams in the Dance. All expansion has done is force the #1s to play better and better automatic qualifiers.

Does that make more sense? Sorry if it is confusing.

-Jason "I am sure someone is going to point out some massive fatal flaw in my argument because I consider Tommy a very smart poster and if I confused him, odds are excellent my entire theory is bad" Evans

[hurleyfor3]

Nah, I'm pretty sure their "model" consisted of asking Doug Gottlieb.

[Olympic Fan]

I'm not going tio debate statistical models, but I would make one obnservation:

Of all the potential No. 1 seeds, Duke is among the LESS likely to lose to a marginal team. This season -- even counting the 13 games without Kelly -- Duke has not lost to anybody in the RPI range that the 16s will be. Duke's worst loss was to No. 86 Maryland.

On the other hand, we have a potentia No. 1 in Kansas that lost to No. 235 TCU.

We have potential No. 1 Georgetown losing to No. 127 Souh Florida

We have potential No. 1 Miami (although that's a lot less likely now) losing to No. 163 Wake Forest.

[tommy]

Exactly.

Tommy, you are looking at it from top down and thinking that rather than inviting the top 64 teams, we now invite the top 68. As tooksies points out, my way of looking at it is to go from the bottom up... because no matter how large you expand the at-large field, the bottom is still the bottom. That bottom is made up of automatic qualifiers who will always be there and the quality of those teams cannot be changed by the number of automatic qualifiers. So, rather than playing the 4 worst automatic qualifiers, the #1 seeds are now playing 4 of the 6 worst automatic qualifiers. What's more, they are assured of playing the 2 best of those 6 who are not in the play-in games... and the other 2 they play are teams that won a play-in game thereby indicating they may be stronger than anticipated. I think a strong case can be made that the quality of the teams facing the #1 seeds is significantly higher than 4 or 6 or especially 10 or 20 years ago.

Yes, as you point out there are now 68 teams in the tournament and, in theory, these teams at the bottom are therefore seeded lower (some of them are de-facto #17 seeds) than the #16 seeds of the past. But, that is because we have grown the middle of the tournament and does not have a practical impact on the quality of the bottom 12 or so teams in the Dance. All expansion has done is force the #1s to play better and better automatic qualifiers.

Does that make more sense? Sorry if it is confusing.

-Jason "I am sure someone is going to point out some massive fatal flaw in my argument because I consider Tommy a very smart poster and if I confused him, odds are excellent my entire theory is bad" Evans

Well first of all, thanks for the compliment. Appreciated.

But maybe not deserved. Because your post made me think more about this, and you're right! The bottom six auto qualifiers, had the tourney only included 64 teams, would've been four 16's and two 15's. The two which would've been 15's but are now 16's will theoretically be the two 16's that don't have to play-in, and instead are matched up directly with a #1, while the four that would've been 16's anyway play-in, with the two winners advancing to play #1's also.

So in that sense yes, two #1 seeds in today's tournament have to play teams in the first round that would've been 15's back in the day. Good call.

Only other thing I would say is that, despite that reality, I'm not sure how much greater it makes the chances of a #1 losing its opener than what the chances would've been under the old sytem. Two #1's are still playing teams that would've been 16's under the old system -- matchups in which the #1 has never lost -- while two are playing teams that would've been 15's. So as a whole, the #1's play two 15's and two 16's instead of four 16's. Of course it makes it a little more likely than if all four were playing 16's, but keeping in mind that there have only been six times that a 15 has beaten a 2, including the two from last year, and given the fact that theoretically the 1's are going to be some degree stronger than the 2's, it seems like the do-the-math statistical increase in the likelihood (which I'll leave to someone smarter than I) wouldn't be all that much of a bump.

It seems that we agree that the real driver of the increased chances of a 16 beating a 1 is the improvement in the quality of players and the quality of play that the 16's are bringing to the table, and I wouldn't be surprised if you also agree that the frequency of the creation of true powerhouse #1 seeds, those more impervious to huge upsets, has diminished over the years too.

[striker219]

Nah, I'm pretty sure their "model" consisted of asking Doug Gottlieb.

And if Doug isn't around they can just ask Digger, assuming they can get him to stop sniffing his highlighters long enough to form a coherent sentence.