FiveThirtyEight's Tournament Predictions

[Jimbo]

Not surprisingly, Nate Silver & co. like UConn's chances. They currently give UConn a 70% probability to win the tournament. Notre Dame (9%) is next, followed by South Carolina (8%), Baylor (7%), and Maryland (3%). Everybody else in the field is at less than 1%. For some perspective, their "favorite" on the men's side this year is Kansas (19%).

Somewhat interesting: they were actually more bullish on UConn at this time a year ago. Heading into the 2015 tournament, they gave UConn a 74% chance of winning the whole thing, followed by South Carolina (10%), Notre Dame (9%), and Maryland/Baylor/Tennessee (2% each).

Anyway, for whatever it's worth (hovering over various parts of the bracket is fun, at least): The bracket
They'll keep updating the win probabilities for the remaining teams as the tournament progresses, so if you like this kind of thing, it's worth checking out every so often.

Methodology

Last year's bracket

 

[UcMiami]

Interesting, and if anything, I think it might understate the difference year to year.
Uconn itself is trading a senior KML for a freshman Lou which is on its face a negative trade, but the offsets are senior Breanna, Moriah, and Morgan, and sophomore Kia all I think better. And then trading senior Kiah for sophomore Gabby who are very different players but equally important. Add in a slightly better bench this year that may not be a factor, but in an emergency (fouls) could come in handy. In all about a wash.

Competition wise - I think Baylor is clearly better, SC is a little better and certainly more experience in the tournament, and ND might be just a bit weaker with Jewell and Reimer gone (they are a better perimeter team though.)
Outside the one seeds - Maryland may actually be better, Ohio State is definitely better, Louisville is better, and the rest of the 2,3,4 seeds I have no idea about comparatively as they are all changed.

 

[MilfordHusky]

Last year, our odds were noticeably better than those of the Kentucky men's team.

Though I prefer 74%, I'll take 70%. We miss Kiah and Kaleena, but the seniors are incredible and Lou is doing her Stewie impersonation.

 

[bags27]

If you play around with this:

http://projects.fivethirtyeight.com/2016-march-madness-predictions/womens/

the toughest game up into the final four is considered ND-MD. It's by far the toughest test for any of the four 1 seeds before getting to Indianapolis. Not surprising.

 

[Jimbo]

Interesting, and if anything, I think it might understate the difference year to year.
Uconn itself is trading a senior KML for a freshman Lou which is on its face a negative trade, but the offsets are senior Breanna, Moriah, and Morgan, and sophomore Kia all I think better. And then trading senior Kiah for sophomore Gabby who are very different players but equally important. Add in a slightly better bench this year that may not be a factor, but in an emergency (fouls) could come in handy. In all about a wash.

Competition wise - I think Baylor is clearly better, SC is a little better and certainly more experience in the tournament, and ND might be just a bit weaker with Jewell and Reimer gone (they are a better perimeter team though.)
Outside the one seeds - Maryland may actually be better, Ohio State is definitely better, Louisville is better, and the rest of the 2,3,4 seeds I have no idea about comparatively as they are all changed.

Solid analysis as usual, UcM. Qualitatively, I think that all makes a lot of sense. What surprised me a little bit was that 538's numbers-based model seemed to like last year's one-loss UConn team more than this year's undefeated UConn team. But upon further review, almost the entire difference between the 74% and the 70% can be attributed to the model's assessment of our potential competition in the national semifinal game.

2015
Probability of UConn reaching the Final Four: 96%
Independent probability of UConn beating the winner of Maryland's region: 92%
Independent probability of UConn beating the winner of Notre Dame/South Carolina's half of the bracket: 84%

2016
Probability of UConn reaching the Final Four: 95%
Independent probability of UConn beating the winner of Baylor's region: 88%
Independent probability of UConn beating the winner of Notre Dame/South Carolina's half of the bracket: 83%

This is admittedly a pretty crude analysis, but without digging deep into the numbers that the 538 folks are crunching, it seems that their model likes Baylor's chances in a potential matchup against us slightly better than Maryland/Tennessee's chances last year (which makes sense, IMO), but otherwise its assessment of our path hasn't changed much at all.

 

[bags27]

If you go to this link:
http://fivethirtyeight.com/features/how-fivethirtyeight-is-forecasting-the-2016-ncaa-tournament/

they say that they've slightly modified their methodologies from last year (on a men's bb page; but they say they applied it to women's too). In order words, comparisons of this year's "chances" to last year's are not a precise reflection of relative mathematically perceived strength.

 

[UcMiami]

If you go to this link:
http://fivethirtyeight.com/features/how-fivethirtyeight-is-forecasting-the-2016-ncaa-tournament/

they say that they've slightly modified their methodologies from last year (on a men's bb page; but they say they applied it to women's too). In order words, comparisons of this year's "chances" to last year's are not a precise reflection of relative mathematically perceived strength.

If you read through the methodology which is thoroughly explained on the link from last year there are a couple of interesting things that are different for the women:

1. They mention the large disparity in available data for Women (See the Sue Bird Thread and the quote below) - only three of their five power rating database sources are available for women (they add in Massey for the women to bring their DBs to four), no advanced metrics to handle any player injuries, and the NCAA doesn't provide the actual S curve with the exception of the first 4 teams. (The men's committee does publish this.)

2. There is a large disparity in strength that requires they modify some of their calculations - specifically the 'league of there own' at the top of the ratings for which they have to go to Europe to find similar situations - citing Scottish premier league that has split the last 29 champions between two clubs Celtic and Rangers.

3. The home court advantage for the first two rounds.

"However, there are several challenges unique to the women’s tournament. One is the comparative lack of data — a persistent problem when it comes to women’s sports:"

The 2015 link:
http://fivethirtyeight.com/features/how-we-made-our-forecasts-for-the-womens-ncaa-tournament/

And the other story on 538 re lack of data that is cited:
http://fivethirtyeight.com/datalab/hey-nate-there-is-no-rich-data-in-womens-sports/

 

[bags27]

If you read through the methodology which is thoroughly explained on the link from last year there are a couple of interesting things that are different for the women:/

So the statistical comparisons, which are deeply problematic under ideal conditions (something attributed to Mark Twain about lies and statistics), are here, where n = 2 and not even that because n is not constant, not all that helpful in comparing UConn 2014/15 to UConn 2015/16. Certainly, the comparisons are well within the "margin of error" so that nothing meaningful can be derived.

The only statistic that will be meaningful is if UConn scores one point more than its opponents in every one of the next 6 games.

 

[Jimbo]

This 538 podcast that I just found discusses UConn and 538's probability forecasts quite a bit, including specifically last year's 74% and this year's 70%. I don't necessarily agree with all of their on-the-court analysis, but the statistical talk is pretty interesting.

They talk about the men's tournament first; discussion of UConn and the women's tournament lasts from about the 24:00 mark to 41:30 or so, in case anyone wants to skip straight to that part.

http://espn.go.com/espnradio/play?id=14982169

 

[bags27]

They talk about the men's tournament first; discussion of UConn and the women's tournament lasts from about the 24:00 mark to 41:30 or so, in case anyone wants to skip straight to that part. http://espn.go.com/espnradio/play?id=14982169

Thanks for this, Jimbo. But it's amazing how UNscientific the discussion is. It's pretty much barroom conversation with a few (questionable) statistics thrown in. The only "scientific" point is that UConn's chancess are slightly lower this year because there are this year four rather than two teams with a realistic statistical possibility of beating UConn. But UConn doesn't have to play all four of those teams, because only two of them stand in their direct path. I dunno. I have some respect for statistical sciences (not much, but some); but what 538 has morphed into -- a for-profit BS machine that applies some probability theory to issues with far too many variables (how do you factor the possibility of injury for example) -- makes me sad.

 

[UcMiami]

Thanks for this, Jimbo. But it's amazing how UNscientific the discussion is. It's pretty much barroom conversation with a few (questionable) statistics thrown in. The only "scientific" point is that UConn's chancess are slightly lower this year because there are this year four rather than two teams with a realistic statistical possibility of beating UConn. But UConn doesn't have to play all four of those teams, because only two of them stand in their direct path. I dunno. I have some respect for statistical sciences (not much, but some); but what 538 has morphed into -- a for-profit BS machine that applies some probability theory to issues with far too many variables (how do you factor the possibility of injury for example) -- makes me sad.

But last year only one of them stood in their path.
As for the injury issue - they can do that on the men's side because they have the database support to analyze individual players on teams in meaningful ways beyond just points scored and rebounds. Those statistics do not exist for the WCBB scene. And the injury data is only useful when you actually have an injury occur - it is too random and too infrequent to build into an overview of 64 games - you can only add it when it occurs.
Their podcasts are generally as you say, a group of geeks sitting around a table with or without the beverage of their choice. They are sort of fun additions to the drier reports they produce, but are certainly little more than say Around the Hoop podcasts.

 

[bags27]

But last year only one of them stood in their path.

Right, so on a purely statistical basis that fact doubles the number of realistically possible upsets. But it doesn't account for intangibles, like a coach who progressively builds his team's intensity by going from one tough challenge to the next. I just think there are far too many variables in a situation where there are only 5 participants at a time on each side with a ball that will bounce where it will, where freshmen will play under the bright lights for the first time, and with referees who will call what they see (or don't see). And, as you say, injuries are "too random and too infrequent to build into an overview of 64 games - you can only add it when it occurs."

Billy Beane's cybermetrics was going to be the lock-sure way of assessing Major League player worth. Sure it works to a certain extent, but not nearly what we thought it would, otherwise, everyone would be using it all the time, and the richest teams (Yankees, Red Sox, and Dodgers) would be snapping up that talent and be in the world series every year.

 

[cbraarj]

Stats are overrated IMO because I remember that 2010 NC game was the worst I have ever seen by both teams to play in that game since 1994.

 

[UcMiami]

bags27 - I agree in general about the advanced metrics data - it is interesting but of very limited usefulness in basketball. I wrote a long post in another thread a while ago the gist of which was that baseball, which is the sport that really started the cybermetrics revolution is a game played almost completely in 'one on one' match-ups or 'one on none' defensive plays. And the play is all 'single events' with no continuos action. Because of that it is very easy to evaluate each part of a 'play' for the player involved in that play. As you move to more complicated team games - football which again is discrete single plays, but where on any given play the success/failure of the play is dependent on 10-22 other players on the field - it becomes harder to evaluate a single player on each play. And when you get to basketball the clock doesn't stop very often, and all ten players on the court are involved in each offensive and defensive play, that evaluation becomes even harder.
That said - the more stats and types of stats available the more ways you have of looking at teams and players comparatively and the better chance you have of making predictions accurate.
And when you get to win vs loss which is what 538 is trying to do - it really is pretty accurate - the discussion talks about the bell curve of performance and how you look at how two teams curves overlap - the further apart the top of the curves are, the greater the percentage chance for a team to win. Massey does a pretty good job on their predictions - and if you look at their 'series' predictors you see a 60% chance results to a 5/2 win rate on a 7 game series - if you run it a few times there will be some that end up 6/1 and others that end up 4/3 and if you ran it enough you might get one that ended up 3/4 - it is all probabilities,

 

[bags27]

UcMiami, Good points. But I hope you'd agree that the difference between last year's 74% probability of a NC vs. this year's 70% probability is essentially no difference at all.

 

[UcMiami]

UcMiami, Good points. But I hope you'd agree that the difference between last year's 74% probability of a NC vs. this year's 70% probability is essentially no difference at all.

Oh yes - absolutely - it is based on a series of six unique events with multiple variations on 5 of them - so the margin is far greater than 4% - the margin of the predictions is probably 10+/-% so the overlap in prediction is more than 50% of that margin.

 

[Waquoit]

So what they are saying is the field has a 30% chance of winning. If you bet the field in Vegas you get 6-1. That's a pretty big premium. If these 537 guys really believe in what they are putting out, why aren't they putting down some cash on the field?

 

[bags27]

So what they are saying is the field has a 30% chance of winning. If you bet the field in Vegas you get 6-1. That's a pretty big premium. If these 537 guys really believe in what they are putting out, why aren't they putting down some cash on the field?

Like when someone's trying to sell you a stock: if it's such a good investment, why are you trying to sell it to me; why aren't you buying up all the shares yourself?

 

[UcMiami]

So what they are saying is the field has a 30% chance of winning. If you bet the field in Vegas you get 6-1. That's a pretty big premium. If these 537 guys really believe in what they are putting out, why aren't they putting down some cash on the field?

Because it doesn't matter where Vegas places the payout - if you bet against Uconn you are still much more likely to lose. Its like the lottery - doesn't matter what the jackpot is, your chance of winning it still remains infinitesimal - you hopefully aren't taking a mortgage on you house to buy lottery tickets or to bet on the field.