I've been asked to post this in the wagering section, so I know I'm reposting... and I must stress... that this is for entertainment purposes only...we have never won anything with this system, it's just fun and can give a good guess for what cames may be close but anyway here it is...

My office mate from grad. school and I use multivariate statistics to try (stress...try) to predict the winner of all the games. We use a multivariate technique called profile analysis which basically means we create a profile (algebraic equation: take a team, multiply their ppg by the ppg constant, fg% by the fg constant...etc... add them all up...that's their score) for the 'winners' and 'losers' of past tournaments (we go back to 99). There are different equations for each round..and different rounds have better predictability (the earlier rounds are better because of a larger n size). We have two different desicion rules so we have different brackets. One is absolute value of the winner formula, and the other is a difference score between winner score and loser score. Granted, we have never won my pool with it, but it is still fun to do. We have had moderate success: It predicted George Mason into the final four. However, there has been some setbacks, it did predict Albany to beat UCONN in the first round (which almost happened, but almost doesn't cut it...).
Anyway, we will be putting the bracket together later today, i'll post it here for you all to snicker at.

And thus, proving I'm a statistics nerd.

_________________

post it here?
and then have everyone that has a bracket in the CSFMB tourney using it as a tool?


please wait until after the games start tomorrow....once all of our brackets are in so we can see how we "stack up" vs. the predictions.

_________________
@audioidkid

believe me, you don't want to look at it and see how you measure up, in my pool it never breaks the top ten. It only has about a 75% accuracy rating, which in pools doesn't work because there are some people who luck out and have 85 to 90 percent accuracy...not to mention it CONSISTENTLY picks a 16 to beat a 1, or at least has for the past three years. I am proud to say that it usually is the 16/1 matchup that is the closest.

_________________

i have a few upsets as well, but i'm not going to share them until the brackets are in.
i guessed wrong on Florida A&M over Kansas, tough.

_________________
@audioidkid

they are done.
I'll wait until tomorrow to post, unless there are those that really want to see them.
they don't look bad, I'm not using they for my csfmb pool entry, but they has a shot in my other pool.

_________________

I love multivariate statistical analysis. Whenever I think about the subject, I recall the brainy babes from grad school. I love it when women talk regression to me.

Psycory, There is a statistician, Ken Pomeroy, who uses several different approaches to statistical analysis for college hoops. His main predictive system uses offensive and defensive efficiency (points scored and allowed per possession), and tempo/pace ratings. You might take a look at his formulas and predictive models. www.kenpom.com He has his predictions up for each team in the tourney. Click on each team's name and then "schedule" to see the prediction for the first round games.

I've checked it out, it does a good job, infact one of the other statistic nerds in my home pool uses his ratings (he never wins though). We started this as a project in a multivariate statistics course, and we have expanded it from there, now that we have students working for us, we want to get all the stats we can since 1985. We are not there yet, and our formulas are weighted way too heavy on offense statistics, but the predicted winners this year are pretty good picks. Unlike our first shot (12 seed winning it all).

_________________

Here it is,
whenever both teams either fit the winner profile or did not fit the winner profile we used two different decision rules, rule for the first bracket is who fit the winner profile better
Midwest - first round winners
florida arizona odu maryland nd oregon gt wisconsin

second round winners
florida maryland nd gt

third round winners
Florida gt

Midwest finalist
Georgia Tech

West First round winners
Kansas Villanova VT SIU Duke Pitt Indiana UCLA

Second round
Kansas VT Duke UCLA

Third Round
Kansas UCLA

West Finalist
Kansas

East first round winners
UNC MSU USC Texas GW WSU BC G-town

Second round
UNC USC WSU G-town

Third round
UNC G-town

East Finalist
Georgetown

South first round winners
OSU BYU Tenn Virgina Stanford Tx A & M Nevada Memphis

Second round
BYU Tenn Tx A & M Nevada

Third Round
BYU
Tx A & M

South Finalist
Texas A & M

Final Game
Kansas over Texas A & M

_________________

the second decision rule is the difference score between winner profile and loser profile and here are the results

Midwest First round
Florida Arizona, Butler Davidson ND oregon UNLV Wisconsin

Second round
Florida Butler ND UNLV

Third round
Florida UNLV

Midwest Finalist
Florida

West First round
Kansas Kentucky VT SIU Duke Pitt Indiana UCLA

Second round
Kansas VT Duke UCLA

Third round
Kansas UCLA

West Finalist
Kansas

East first round
UNC Marquette Arkansas Texas Vandy WSU BC G-town

Second round
UNC Texas WSU G-town

Third round
UNC G-town

East Finalist
Georgetown

South first round
OSU BYU Tenn Virgina L-ville Tx A & M Nevada Memphis

Second round
BYU Tenn Tx A & M

Third Round
BYU Tx A& M

South finalist
Texas A & M

Final Game: Kansas over Texas A & M

_________________

Here's my thoughts on the stats brackets
1. this is the first time that they both picked the same final game and the same winner

2. the stats analysis hates the big ten...which is not surprising b/c defense is under represented in our stats

3. they like BYU

4. Only the first three rounds had profiles that were significant...meaning they would be able to classify with some accuracy, after that they are not significant but that may be due to a statistical power issue (for example we only have a sample of 8 for predicting the final winner vs. a sample of 330 for the first round).
Go ahead and rip them apart.

_________________

Well when I did my statistical analysis, the first dart hit Texas....

That's good stuff Psycory. I actually really get into statistical analysis and am interested to see how well your model predicts. Thanks for posting it.

_________________
You know me like that.

for those wondering, here are the stats we used:
Strength of Schedule
Sagarin Ratings
RPI
Con Sagarin
Con RPI
winning pct
fg pct
3pt pct
ft pct
rebounds
assists
turnovers
blocks
points per shot
Assist/TO ratio
Def fg %
Points allowed
rebound margin
ppg
ppg - pts allowed

_________________

I rest my case on why this never wins.

_________________

"He uses statistics as a drunken man uses a lamp-posts....for support rather than illumination"

-Andrew Lang-

_________________
"In an insane world, only the sane will seem insane"

but bear in mind, the anaylsis doesn't have any big ten teams after the second round. So if they survive to the sweet sixteen, then it is wrong, again.

_________________