Schedules, Resutls and Probabilities

I added Schedules, Results and Probabilities to the individual team pages for each team. These include the date, opponent, opponent's RPI Forecast, Location, scores, probabilities of winning and the predicted spread. These probabilities (and spreads) are all derived from Jeff Sagarin's PREDICTOR and are used to generate the random wins and losses used in the 10,000 simulations each day. Basically, you can come up with probabilities of winning if you have a prediction of the spread and some sort of a standard deviation. Click on any team's name to see this new feature. Here is what Florida's looks like:

http://www.rpiforecast.com/teams/Florida.html

As with everything else (except for the tournament projections), these are updated daily.

One more thing: The expected record can be easily calculated from the projections. If you make the assumption that the probabilities are independent across games, then you simply add up the probabilities which will give you the expected wins. The expected losses will be the total number of games minus the expected wins.

Thus, even if a team is favored win in each individual game, like UCLA, for example, when you look at them all as a whole, on average, UCLA will lose 5 out of the 17 remaining. Naturally, this all depends on the probabilities being correct and again, this is an expectation. UCLA could do better or worse in any given simulation (and in reality). But, what I am presenting is the best estimate. If you think your team will do better (or worse? What kind of a fan are you?) then simply look at the RPI Forecast broken down by end-of-season record and pick your favorite.

Blog Archive

If you are new to the site and want to learn a little more about what all of this means, then the blog archive is here:

http://www.rpiforecast.com/blog.html

RPI Forecast Tournament Projections

I have started making the tournament field projections for next March. I know, I know, it is very early in the season to be able to say much about March, but as long as others are doing so, I figured I may as well too. Because my projections are purely objective, they should do quite a bit better than others early in the season, and maybe not quite as well as some very late when there is more likely to be some inside information about which teams will make the cut.


Here is the basic procedure:

1. Make end of season RPI Forecasts which entails:
   a. take completed games as given
   b. Use Jeff Sagarin's updated PREDICTOR to generate probabilities over future games
   c. Simulate 10,000 seasons based on random draws using the probabilities
   d. For each simulated season, calculate RPIs by conference, team, vs. 1-25, etc.
   e. Take averages to get an estimate of Expectations of the above



2. Using the "Dance Card Methodology" and the end-of-season projections from step 1, project the field (including automatic bids).

To be included in the Bracket Project Matrix, you need to take some sort of stance on seeding which is really quite subjective. Basically, I am not spending too much time on that aspect.

Last season, I kept track of the number of correct picks (for the field, not for the seeds) for many of the top sites including:

Basketball Predictions
Beat the Experts
Bracket Express
Bracket Project
Bracket Racket
Bracket WAG
Bracket Watch
Bracketography
Bracketology 101
Bracketology 3
Breaking Down the Bracket
Bryce's Bracket Predictions
Build a Bracket
CBS Sportsline
College Hoops Net
College RPI
Colton Index (JCI)
Crashing the Dance
DhankLily
ESPN Bracketology
FOX Sports
JCI
Jerry Palm
MAG
March Madness 07
Mr. Bracket
MRI Sports
NCAA Bracket Predictions
NCAA Hoops Digest
NetWire
PHSports

I calculated the number of teams that were correctly picked from the field over time, and compared it to the RPIForecast projections. Here are the results:

As you can see, the RPIForecast tournament projections did reasonably well compared to others. One thing to note is that I didn't start tracking others until early January, so there is really no telling what happend before then with the others.

New Season of Forecasts

NCAA Basketball season is here again. Forecasts are now up and running at rpiforecast.com. Because the probabilities used in the simulations are based on Jeff Sagarin's PREDICTOR, for the first few weeks of the season, those are tied to the starting ratings. Keep that in mind. Eventually, everything will be based only on games played this season.

One thing that I am implementing from the get-go is the RPI forecasts broken down by different end-of-season records. That way, if you disagree with what the projected W-L reccord is for your team, you can insert what you may think to be a better prediction and see what the resulting RPI forecast is. Another way of looking at it is: How many games does the team have to win to have an RPI above a certain number.

Click here for an archive of blog entries from last season