Category Archives: Uncategorized
Voter bias, football polls, and TCU
One of the topics undersold during the arguments of which four NCAA football teams deserved a spot in the college football playoff was the effect of voter bias on decision making.
Specifically, literature has found NCAA football poll voters to be biased in a few ways.
Bias #1- Associated Press (AP) poll voters are biased towards teams (i) in the voter’s home state, (ii) in the same conference as teams in the voter’s state, (iii) in BCS conferences, and (iv) teams playing in more televised games.
Bias #2- Coaching poll voters are biased in favor of both their recent opponents and their alma-maters.
Bias #3- AP voters are biased in favor of teams which were ranked higher earlier in the season.
It’s obviously too early to tell whether or not these biases will hold with the college football playoff selection committee over the long run. However, it’s particularly curious how the decision-making process…
View original post 486 more words
College Football Conference Construction
So the Big XII is pissed. (And probably rightly so). As I’ve said before, if the committee was trying to pick the best 4 teams in college football, they failed miserably. TCU and Baylor, both one loss tams, are both better than Ohio State, imho. Further, there are several multi-loss teams that are better than Ohio State or Florida State. I’d suggest Mississippi State (2 losses) AND Ole Miss (3 losses) for starters. But I’d also include basically any team from the SEC West including Georgia, Auburn, and LSU. And screw it, I’m going to include Arkansas. I think Arkansas would beat Florida State or Ohio State. The SEC is that good.
Since the simpletons on the College Football Playoff committee can apparently only see wins, is it possible to game the system? For instance, could a conference add or remove teams from their conference to maximize their potential for getting a team into the playoffs? How would a conference do that? Let’s do a simulation study.
Motivating question
How could a college football conference construct its conference to maximize their chances of getting at least one of their teams into the playoff?
Simulation Description
Let’s assume a simple model for the college football world. Let’s assume there are 5 conferences each with 10 teams. 25 of these 50 total teams are “good” and the remaining teams are “bad”. Each team plays nine games against the other teams in their conference and three “random” non-conference games for a total of 12 games. When a good team plays a good team or a bad team plays a bad team, each team has a 0.5 probability of winning the game. When a good team plays a bad team, the good teams probability of a win is expit(1)=.731. I then simulated a schedule and simulated the season. I counted the 4 teams with the most wins as the four teams that made the playoff. The tie-breakers for teams that were tied in wins was drawing lots (i.e. using runif in R). I then counted how often a team from each of the conferences made it to the playoffs as related to the number of “good” teams in the conference.
Results
Obviously, when all 5 conferences are completely balanced and have 5 good and 5 bad teams, each conference has the same chances to get a team to the playoffs. So let’s look at some unbalanced situations. If the good teams are split up so that the conferences have 2, 3, 5, 7, and 8 good teams, respectively, the conference with the 5 good teams is the most likely to get a team into the playoffs getting in about 71.3% of the time. In this setting, with 7 teams, a conference is just slightly less likely to get a team into the playoffs at 69.4% and it drops even more to 63.1% with 8 good teams. While there are more good teams in the conference to have an opportunity to get into the playoff, these good teams cannibalize each other.
This trend continues throughout all of the simple simulations that I looked at where a conference with about half good and half bad teams was the most likely conference to make the playoffs. At the extremes where there is one conference with 5 good teams and the other conferences are all or nearly all good or bad teams, the conference with 5 good teams probability is the largest.
| 2 | 3 | 5 | 7 | 8 |
|---|---|---|---|---|
| 57.7% | 66.5% | 71.3% | 69.4% | 63.2% |
| 1 | 2 | 6 | 7 | 9 |
| 53.2% | 63.4% | 74.9% | 72.6% | 62.5% |
| 0 | 1 | 5 | 9 | 10 |
|---|---|---|---|---|
| 34.9% | 59.6% | 80.6% | 71.5% | 69.9% |
| 1 | 3 | 5 | 6 | 10 |
| 51.1% | 71.3% | 75.6% | 74.0% | 51.7% |
The plot below shows graphically the results of the above table. On the x-axis are the number of “good” teams in each conference and the y-axis is the probability that a team from that conference gets into the playoffs (i.e. Is top 4 in terms of number of wins.
What does this mean?
From the point of view of a conference commissioner, if your goal is to build a conference with the purpose of putting teams in a position to win a national championship, your best bet is NOT to construct a power house conference. You want to put together a conference with about half of the teams being elite and the other half of the teams being not so good. If you construct a conference with all elite teams, the teams cannibalize each other and no team is clearly the best. No matter how hard the schedule, this committee, I believe, absolutely will not let a 2 loss team into the playoffs even if that two loss team lost to the number 1 and number 2 best teams in the country. So to all those conference commissioners looking to add a power house program to their conference, maybe they should reconsider and add a middling program to their conference and get their elite teams another win. Cause after all, that’s basically all this committee cares about.
Future Work
What I’d like to do in the future, is run this simulation on real college football teams this year to try to construct the ultimate conference for getting a team to the college football playoffs. I’d use some sort of estimated team strength (Maybe Sagarin) to simulate games during a season and then take the top 4 teams in terms of wins. That could be really interesting.
Cheers.
Just another reason why the college playoff committee is terrible at their jobs
There are 36 bowl games with lines. 5 of them have spreads of 9.5 or greater. 2 of those five are the national semi-finals. If the goal is to get the 4 best teams into a playoff, then this committee has failed miserably. But I guess that’s sort of what I expect out of the NCAA.
Cheers.
NFL Picks – Week 14
Total (weeks 1-14) – SU: 139-68-1 ATS: 105-100-1 O/U: 110-96-2
Week 1 – SU: 9-7-0 ATS: 8-8-0 O/U: 13-3-0
Week 2 – SU: 10-6-0 ATS: 10-6-0 O/U: 10-6-0
Week 3 – SU: 12-4-0 ATS: 9-6-1 O/U: 8-8-0
Week 4 – SU: 7-6-0 ATS: 5-7-1 O/U: 5-8-0
Week 5 – SU: 14-2-0 ATS: 6-9-0 O/U: 9-6-0
Week 6 – SU: 11-3-1 ATS: 8-7-0 O/U: 6-9-1
Week 7 – SU: 11-4-0 ATS: 7-8-0 O/U: 8-7-0
Week 8 – SU: 11-3-0 ATS: 8-7-0 O/U: 8-7-0
Week 9 – SU: 9-4-0 ATS: 8-5-0 O/U: 4-8-1
Week 10 – SU: 9-4-0 ATS: 4-9-0 O/U: 6-7-0
Week 11 – SU: 9-5-0 ATS: 8-6-0 O/U: 7-7-0
Week 12 – SU: 10-5-0 ATS: 7-8-0 O/U: 8-7-0
Week 13 – SU: 11-5-0 ATS: 8-8-0 O/U: 7-9-0
Week 14 – SU: 7-9-0 ATS: 9-6-1 O/U: 11-5-0
Dallas at Chicago
Prediction: Bears 24-22 (56.7%)
Pick: Bears +4
Total: Under 53
Kansas City at Arizona
Prediction: Cardinals 21-19 (54.7%)
Pick: Cardinals EVEN
Total: Under 40.5
Pittsburgh at Cincinnati
Prediction: Bengals 23-21 (57.5%)
Pick: Steelers +3.5
Total: Under 47
Indianapolis at Cleveland
Prediction: Browns 24-23 (50.1%)
Pick: Browns +4
Total: Under 50
Buffalo at Denver
Prediction: Broncos 29-20 (68.0%)
Pick: Bills +10
Total: Over 48
Tampa Bay at Detroit
Prediction: Lions 25-18 (68.0%)
Pick: Buccaneers +10
Total: Over 42
Houston at Jacksonville
Prediction: Texans 23-19 (61.9%)
Pick: Jaguars +6.5
Total: Under 42.5
Baltimore at Miami
Prediction: Dolphins 21-20 (53.0%)
Pick: Ravens +2.5
Total: Under 45.5
NY Jets at Minnesota
Prediction: Vikings 21-18 (57.6%)
Pick: Jets +6 PUSH
Total: Under 40.5
Carolina at New Orleans
Prediction: Saints 28-22 (65.9%)
Pick: Panthers +10
Total: Over 49.5
San Francisco at Oakland
Prediction: 49ers 22-16 (66.1%)
Pick: Raiders +9
Total: Under 41
Seattle at Philadelphia
Prediction: Eagles 23-22 (51.8%)
Pick: Seattle +1
Total: Under 48.5
NY Giants at Tennessee
Prediction: Titans 22-21 (50.5%)
Pick: Titans +1.5
Total: Under 46
St. Louis at Washington
Prediction: Washington Football Team 24-20 (60.9%)
Pick: Washington Football Team +3
Total: Under 44.5
New England at San Diego
Prediction: Patriots 27-25 (53.8%)
Pick: Chargers +4
Total: Under 52
Atlanta at Green Bay
Prediction: Packers 28-22 (66.3%)
Pick: Falcons +13
Total: Under 56
Building an NCAA men’s basketball prediction model
Last Spring, Loyola statistics professor Greg Matthews and I won the March Machine Learning Mania contest run by Kaggle, which involved submitting game probabilities for every possible contest in the 2014 NCAA men’s basketball tournament.
Recently, we co-wrote a paper that motivates and summarizes the prediction model that we used. In addition to describing our entry, we also simulated the tournament 10,000 times in order to help quantify how likely it was that our submission would have won the Kaggle contest.
The paper has been submitted for publication at a journal, and we are crossing our fingers that it gets accepted. The pre-published version of the paper is up on arXiv (linked here).
Quick summary: to estimate the probabilities for each game, we merged two probability models, one using point spreads (Rd. 1) and estimated point spreads (Rd. 2- Rd. 6) set by sports books, and the other using team efficiency metrics from Ken Pomeroy’s website.
According…
View original post 138 more words
Estimating causal effects with ordinal exposures
Just passing along a quick note from the world of academia; I, along with my adviser from Brown, Dr. Roee Gutman, published our first paper together.
It’s titled ‘Estimating the average treatment effects of nutritional label use using subclassification with regression adjustment,‘ and presents a case study of how to measure the causal effects of an ordinal exposure. The article is currently online in Statistical Methods in Medical Research.
The online version (paywall) of the article is linked here. You can also download a pre-published version on the arXiv by going here. Finally, here’s the abstract and keywords.
What is the main point of this paper?
Here’s one of my favorite parts, a graph showing the covariates’ bias before and after subclassifying subjects into groups. In this and many other examples, subclassifying is an important tool as it allows for more of an apples-to-apples comparison. Specifically, it only makes…
View original post 125 more words
NCAA Football Top 25 – November 30, 2014
My top 2 remain the same this week, but they’ve flipped spots. I’ve got Oregon moving up to number 1 and Alabama dropping to 2. Mississippi State and UCLA both fall out of my top 4 after losses to Ole Miss and Stanford, respectively. TCU moves up to 3, mostly because of losses ahead of them, but also because of their win over Texas. It will be interesting to see how the selection committee deals with a choice between Baylor and TCU if they both win their remaining games and end up at 11-1. My fourth ranked team is going to upset a lot of people: Georgia. They move up one spot even after their loss to Georgia Tech. First off, it was a close loss (in overtime), and Georgia Tech is actually good. So their penalty for losing to them in minimal. (Also, I’m not the only one who has these results. Sagarin has Georgia 6th.) Rounding out my top 10 is Arizona, UCLA, Miss State, Florida State, Baylor, and Auburn.
Some questions:
- If Arizona beats Oregon (again), they’d have to get in, right?
- Is Florida State in no matter what at this point? Or could Georgia Tech knock them out?
- If FSU gets into the playoff, they are going to be at least 7 point underdogs to every team in the playoff, imho.
- FSU is only a 3.5 point favorite over Georgia Tech. Georgia Tech will be the best team that FSU has played all year. (Remember when everyone thought it as Notre Dame.)
- Is Alabama in no matter what? I think the answer is yes.
- If Baylor beats Kansas State, I think you have to take Baylor, right?
- According to betting odds here are the favorites to win the College Football Playoff: Alabama, Oregon, TCU, Baylor, Florida State, Ohio State, Missouri, Wisconsin, Arizona, Georgia Tech, Kansas State, and Mississippi.
|
Cheers.
NFL picks – Week 13
Total (weeks 1-13) – SU: 132-59-1 ATS: 96-94-2 O/U: 99-91-2
Week 1 – SU: 9-7-0 ATS: 8-8-0 O/U: 13-3-0
Week 2 – SU: 10-6-0 ATS: 10-6-0 O/U: 10-6-0
Week 3 – SU: 12-4-0 ATS: 9-6-1 O/U: 8-8-0
Week 4 – SU: 7-6-0 ATS: 5-7-1 O/U: 5-8-0
Week 5 – SU: 14-2-0 ATS: 6-9-0 O/U: 9-6-0
Week 6 – SU: 11-3-1 ATS: 8-7-0 O/U: 6-9-1
Week 7 – SU: 11-4-0 ATS: 7-8-0 O/U: 8-7-0
Week 8 – SU: 11-3-0 ATS: 8-7-0 O/U: 8-7-0
Week 9 – SU: 9-4-0 ATS: 8-5-0 O/U: 4-8-1
Week 10 – SU: 9-4-0 ATS: 4-9-0 O/U: 6-7-0
Week 11 – SU: 9-5-0 ATS: 8-6-0 O/U: 7-7-0
Week 12 – SU: 10-5-0 ATS: 7-8-0 O/U: 8-7-0
Week 13 – SU: 11-5-0 ATS: 8-8-0 O/U: 7-9-0
Arizona at Atlanta
Prediction: Falcons 24-21 (58.3%)
Pick: Falcons +3
Total: Over 44.5
San Diego at Baltimore
Prediction: Ravens 23-20 (58.9%)
Pick: Chargers +6
Total: Under 46
Cleveland at Buffalo
Prediction: Bills 23-20 (58.9%)
Pick: Bills -2.5
Total: Over 41
Philadelphia at Dallas
Prediction: Cowboys 26-24 (53.9%)
Pick: Eagles +3.5
Total: Under 56
Chicago at Detroit
Prediction: Lions 24-21 (58.3%)
Pick: Bears +7
Total: Under 47.5
New England at Green Bay
Prediction: Packers 28-26 (54.0%)
Pick: Patriots +3
Total:Under 59
Tennessee at Houston
Prediction: Texans 24-18 (66.8%)
Pick: Titans +6.5
Total: Under 43.5
Washington at Indianapolis
Prediction: Colts 26-22 (61.1%)
Pick: Washington Football Team +10
Total: Under 51.5
NY Giants at Jacksonville
Prediction: Giants 23-19 (59.1%)
Pick: Giants -3
Total: Under 44.5
Denver at Kansas City
Prediction: Broncos 25-22 (58.9%)
Pick: Broncos -2.5
Total: Under 50.5
Carolina at Minnesota
Prediction: Vikings 21-20 (51.0%)
Pick: Panthers +2.5
Total: Under 43
Miami at NY Jets
Prediction: Jets 20-19 (50.6%)
Pick: Jets +6.5
Total: Under 42
New Orleans at Pittsburgh
Prediction: Saints 26-25 (51.6%)
Pick: Saints +4.5
Total: Under 54
Seattle at San Francisco
Prediction: 49ers 20-19 (52.9%)
Pick: 49ers -1
Total: Under 40
Oakland at St. Louis
Prediction: Rams 23-18 (63.3%)
Pick: Raiders +7
Total: Under 42
Cincinnati at Tampa Bay
Prediction: Bengals 21-20 (54.2%)
Pick: Buccaneers +4
Total: Under 44
NCAA Top 25 – November 23, 2014
Some questions:
Let’s say Alabama goes to the SEC championship game and loses. Does that knock out Alabama and keep Miss State in (assuming they beat Ole Miss in the Egg Bowl)?
If Missouri somehow wins the SEC, they have to get in to the playoff right? Who would get knocked out then?
I feel confident that Florida State will lose one of their next two games. Does a 1 loss Florida State get into the playoff over a TCU or a Baylor?
Is Ohio State losing to Virginia Tech the weirdest loss of this season?
What is the playoff committee going to do with a team like Arkansas? Are they top 25?
Seriously, why is Condoleezza Rice involved in this?
Do you think when Ole Miss was 7-0 that fans really thought they had a chance to win the national championship?
Is there anyone in the world happier than me that Notre Dame has lost 4 out of their last 5?
Cheers.
|
NHL game outcomes using R and Hockey Reference
I’m always impressed with the contest and accessibility of the Baseball with R website (here), which features a great cast of statisticians writing about everything from Hall of Fame entry to umpire bias.
In a similar vein, I highly recommend Sam and AC’s nhlscrapr package in R. I’ve used it extensively to analyze play-by-play data from past seasons (for example, this post on momentum in hockey).
However, I have a soft spot for overtime outcomes in the NHL, and while the nhlscrapr package has game-by-game results, there isn’t a straight-forward mechanism for identifying whether or not a given game went to overtime. Further, data in the nhlscrapr package only goes back about a decade or so.
Thankfully, Hockey Reference has easily accessible (and scrapable) tables for us to use. Given that I am doing some updated analyses over NHL overtime rates, and that I wanted an easier method than copying and…
View original post 71 more words
Stats in the Wild 