News
New
Book: Improving your NCAA®
Bracket with Statistics
I have written a book on the bracket pool. It covers the history of the pool and the methods that have been developed to improve brackets. The book uses pictures, graphs and no complex equations. It has a chapter covering all the good bracket advice web sites that I could locate. This book is not focused on Poologic, but it covers the methods used here and much more. The publisher's page is here. Also at Amazon
Please
Help Me Measure Charitable Giving
Please mark your gifts "In Honor of" Poologic so that the V Foundation. can identify those gifts. There is no need for you to provide a snail-mail address for the honoree. I will get summary numbers from the V Foundation and they will not reveal any of your personal information.
More New
Models
Two more models have been added: KenPom and KenPomV. KenPomV uses the Vegas spreads for the first round. Both use spreads based on Kenpom ratings elsewhere.
New
and Improved Models
Two new tournament outcome models have been added: “Combo” and “LRMCLog”. And the LRMCP model has been improved to predict better according to the Likelihood Test. The LRMCLog substitutes the log of the pi factors into the LRMCP equation and this performs better.
In terms of statistical performance on the Likehood Ratio Test, Combo = Original = Sagarin = LRMCLog > LRMCP > Vegas. That is, the top four models are close enough to each other to be indistinguishable.
The metric used to evaluate each model (likelihood) is the probability that the model assigns to the joint outcome of the last 9 tournaments before 2020 (2011-2019).
Compensating
for Multiple Prizes
The ROI Calculator is tuned for a winner-take-all pool. Underestimate the number of opponents to compensate for multiple prizes. For instance, if 1st place only gets half the prize money, then cut your estimated opponents by half.
Warning:
Play-in Games Estimated in Early Data Loads
If you use Poologic results before the play-in (aka new 1st Round) games, you might find it recommending a team that loses in the new 1st Round. Probably best to wait for a later data load, after the new 1st Round is complete.
But the last play-in game will not start till 9 PM on Wednesday, so the last Poologic data load will be much later this year.
(For 2011, the Vegas spreads and totals for the Wednesday games had to be estimated in the last data load because they were not available as of 8 AM EDT on Thursday)
The
Contrarian Full Court
Clair and Letscher published a paper in 2007 that applies contrarian considerations to the entire pool entry sheet for standard scoring pools. David Letscher Homepage: NCAA Tournament has been providing optimized entry sheets based on this paper for the last few years. He has been updating the site after the play-in game. See David Letscher’s site for a link to the paper.
If you want to combine the Clair and Letscher’s strategy with the multiple-entry strategy, try this hack: Use the ROI Calculator to pick multiple champs and promote those champs to the top of one of David Letscher’s recommended entry sheets.
Futures
Model No Longer Supported
My data source for the Futures Model went belly up, so I am not supporting it this year.
New
Paper on the Contrarian Strategies
Published in Chance about a year ago:
http://stat.duke.edu/~jbn9/papers/Niem_Carl_Alex_cont_2008.pdf
New
and Improved Probability Models
I have added two Logistic Regression/Markov Chain Models (LRMCD,LRMCP) from Kvam and Sokol [1] and I have switched to using the Sagarin Predictor ratings to create the Original and Sagarin models. All four of these model should be improvements over all the 2006 models except possibly the Futures models. This claim is based on statistical tests presented in Kvam and Sokol, and other testing of the Sagarin Predictor using Kvam and Sokol's methods. There is too little data available to test the Futures model vs other models.
I cannot say that any of the LRMCD, LRMCP, improved Original, improved Sagarin, or Futures models have performed better than the others because the historical data available to me (including Kvam and Sokol's results) do not resolve them. The Kvam and Sokol provide evidence that the Vegas model (which they call "KG") has underperformed the LRMCD and LRCMP models.
New
Scoring Options
You can now specify that standard scoring are not awarded for upsets by indicating "Don't award standard points for upsets". By default, standard scoring points are awarded for upsets.
I have added "Bonus" scoring factors and a seed "Limit" for each round. Bonus points are awarded when the winning seed is greater or equal to the limit. The bonus points are awarded in addition to the standard points.
May
not Support All Models
I may configure Poologic so that some models mentioned in the web pages may not be available in the drop down menus. This will allow me to load Poologic with some models when the data for others is not available.
A
Map to the Overbet Teams
Yahoo Tournament Pick'em is providing frequency distributions for pool sheet entries in there nationwide pool. The 2005 link is here and the same link may work for later years.
Use the Pick'em distribution for the final round to estimate the entries for teams in the ROI Calculator. Take the percentage from the Yahoo distribution and multiply it by your estimate of the total opponent entries in your pool. For instance, if you estimate 150 opponents and Duke has been picked by 30% of the Pick'em entries to win, then estimate that 45 of your opponents will pick Duke.
But, you may want to try to calibrate for localized favorites since the Pick'em data is based on nationwide entries. You might need to keep track of how the Pick'em distribution compares to your local pool distribution over repeated tournaments to come up with a calibration factor.
If calibration is too much work or not possible, then just follow the practice of never betting a localized favorite as champ even if the ROI Calculator puts it among the top ROI teams.
[1] "Logistic Regression/Markov Chain Model for NCAA Basketball" by Paul Kvam and Joel S. Sokol (Naval Research Logistics Vol. 53, No 8, December 2006, pp. 788-803)