0 |
Subject: Statistical Question
Posted by: The Beezer
- [191202817] Tue, Mar 13, 10:10
Is there a good way to turn an expected point spread into an expected winning percentage? For example, if I think Tennessee is a 2-point favorite over Charlotte, what % of the time should I expect Tennessee to win based on that spread? Thanks in advance! |
1 | Sludge
ID: 18116195 Tue, Mar 13, 10:28
|
Yes.
You can "buy" a half point for, I believe, an extra 10% on the bet. (Been a while since I figured this.) You can then convert that to the probability of a team favored by "X" points winning. Those probabilities are:
0.0 0.500000 0.5 0.523810 1.0 0.545455 1.5 0.565217 2.0 0.583333 2.5 0.600000 3.0 0.615385 3.5 0.629630 4.0 0.642857 4.5 0.655172 5.0 0.666667 5.5 0.677419 6.0 0.687500 6.5 0.696970 7.0 0.705882 7.5 0.714286 8.0 0.722222 8.5 0.729730 9.0 0.736842 9.5 0.743590 10.0 0.750000 10.5 0.756098 11.0 0.761905 11.5 0.767442 12.0 0.772727 12.5 0.777778 13.0 0.782609 13.5 0.787234 14.0 0.791667 14.5 0.795918 15.0 0.800000 15.5 0.803922 16.0 0.807692 16.5 0.811321 17.0 0.814815 17.5 0.818182 18.0 0.821429 18.5 0.824561 19.0 0.827586
Unfortunately, I wouldn't trust them much past 10 points.
Another method is to collect historical data regarding outcomes for past basketball games and build a model to predict the probability of a team winning. Unfortunately, time is short, and I don't know if such data exists in a format that can be quickly used.
Or... you can just go to a site like Massey and look at his probabilities. Or you can go to Team Rankings and look at theirs. Or any number of sites.
|
2 | The Beezer
ID: 191202817 Tue, Mar 13, 10:39
|
Thanks, Sludge. Exactly what I was looking for! For what I'm doing, these factors should more than suffice. I'm doing 1 set of picks off the Sagarin ratings using expected returns, and then another based off the gut. Should be interesting to see who wins this "man vs. machine" battle...
|
3 | The Beezer
ID: 191202817 Tue, Mar 13, 11:05
|
1 more Q; do you have to formula to generate that? I know it has something to do with natural log, but I'm just missing something obvious and can't seem to nail it down. Thanks!
|
4 | Sludge
ID: 18116195 Tue, Mar 13, 11:31
|
Formula to generate what? The listings I gave? It doesn't involve natural log at all. I'm not even sure, now that I think about it, that these would even apply to point spreads in basketball.
|
5 | The Beezer
ID: 191202817 Tue, Mar 13, 11:34
|
That would explain why I can't find it. :) Thanks anyway.
|
6 | steve houpt
ID: 51291010 Tue, Mar 13, 12:13
|
Here is something I came up with a few years ago. Converted Massey or someone elses formula to use with Sagarin ratings. Do not ask how I came up with numbers. It was so long ago, I don't remember. Not sure of 'statistical' accuracy any more (if there ever was any). But it is relative when looking at all teams.
TEAM A=1/(1+10^((((SAGARIN TEAM[B]-17.79)^1.724) - ((SAGARIN TEAM[A]-17.79)^1.724))/400))
TEAM A = 95.00 TEAN B = 90.00
TEAM A = .755 (have to convert to 75.5%)
NOTE: 5 points favorite does not always mean 75.5% probability of winning in this method. Based on probability that Duke is less likely to blow/lose a game over a team they are favored by 5 points than a Winthrop is.
TEAM A = 75.00 TEAM B = 70.00
TEAM A = .711 (71.1%)
Does not work with rating below 17.79.
|
7 | Toral
ID: 452461218 Tue, Mar 13, 12:18
|
Thanx, Sludge! Whoops, that wouldn't work for football? I've been looking for a chart like this in football for a long time. Inhabited gambling usenet groups and couldn't get one (maybe no one would give it to me.)
Would a football chart be different?
Toral
|
8 | Sludge
ID: 18116195 Tue, Mar 13, 13:00
|
Toral -
That's one of the lists that I actually used for football pickoff. It didn't perform very well. The historical model has always been the winner (in the two years, anyway), even over the computer rankings. I suppose I could open up a bit and share that one. :) (It could be refined even more by looking at even more historical data. I believe the data I used covered only 5-6 years.
Spread Probability 0.0 0.500000 0.5 0.512255 1.0 0.529510 1.5 0.546473 2.0 0.563152 2.5 0.579558 3.0 0.595700 3.5 0.611585 4.0 0.627221 4.5 0.642617 5.0 0.657779 5.5 0.672715 6.0 0.687431 6.5 0.701933 7.0 0.716229 7.5 0.730323 8.0 0.744220 8.5 0.757928 9.0 0.771450 9.5 0.784792 10.0 0.797958 10.5 0.810952 11.0 0.823781 11.5 0.836446 12.0 0.848954 12.5 0.861306 13.0 0.873508 13.5 0.885563 14.0 0.897475 14.5 0.909246 15.0 0.920880 15.5 0.932381 16.0 0.943750 16.5 0.954992 17.0 0.966109 17.5 0.977104 18.0 0.987979 18.5 0.998737 19.0 0.999000
|
9 | The Beezer
ID: 191202817 Tue, Mar 13, 13:05
|
Thanks for the formula, Steve. Since I just finished the numbers using Sludge's numbers above, I'll save them for next year. :) BTW, I found the formula for Sludge's numbers, in case anyone wants to apply it without rounding spreads. It's (spread+5)/(spread+10). Ex: 5 point spread= (5+5)/(5+10)=10/15=0.666667 as above.
|
10 | Sludge
ID: 18116195 Tue, Mar 13, 13:19
|
steve h -
That looks like the formula given on Massey's page for computing probabilities using the Sauceda rating system. (Here)
It looks like you've just converted the Sagarin ratings into a Sauceda rating to do the computation. Recall how you did the conversion? Regression?
|
11 | Sludge
ID: 18116195 Tue, Mar 13, 13:22
|
Beez -
That would be it. I don't recall if I simplified it down to that simple formula when I originally computed them. I'm notorious for finding the formula (simple or not) and using it as is.
|
12 | steve houpt
ID: 51291010 Tue, Mar 13, 13:58
|
I have that formula in an 8 team bracket. Getting bigger was too hard for my spreadsheet (and brain).
Using that and Sagarin recent ratings (not full season) I came up with the following chances of making the Elite Eight (3 wins) out of a million tournaments.
Number of times making it to the Elite 8 has no 'direct' effect on chances against other teams that are predicted to make it in different brackets.
EAST (upper) Duke_____ 956,654 UCLA_____ 18,182 Ohio St__ 13,938 5 others_ 11,226
EAST (lower) KY_______ 578,758 BC_______ 311,829 USC______ 51,365 Iowa_____ 24,483 Creigh___ 23,170 3 others_ 10,394
MW (upper) Illinois__ 590,187 Kansas____ 330,815 Tenn______ 32,321 Charlotte_ 23,592 Syracuse__ 16,811 4 others__ 6,274
MW (lower) Arizona__ 756,186 Miss_____ 106,362 Wake F___ 90,664 Xavier___ 22,365 ND_______ 16,641 3 others_ 7,512
SOUTH (upper) Mich St__ 753,697 Okl______ 99,440 VA_______ 89,069 Calif____ 22,686 Fresno___ 20,076 Gonzaga__ 13,064 2 others_ 1,698
SOUTH (lower) Florida__ 501,794 UNC______ 310,788 Prov'nce_ 74,105 Temple___ 46,295 Texas____ 48,856 Penn St__ 11,504 2 others_ 9,658
WEST (upper) Stanford_ 770,548 Indiana__ 163,499 Cinci____ 31,577 St Joe's_ 14,958 4 others_ 19,418
WEST (lower) Maryland_ 712,919 Iowa St__ 123,111 Arkansas_ 96,764 WISC_____ 49,680 Georgt'n_ 13,419 3 others_ 4,107
READ AT OWN RISK.
|
13 | steve houpt
ID: 51291010 Tue, Mar 13, 14:11
|
Sludge - I cheated. Didn't have a spreadsheet that would do regression. Worked equation backwards, substituting different variables until I got lowest error for whatever I wanted at the time to get similar results at various points.
|
14 | The Beezer
ID: 191202817 Tue, Mar 13, 17:15
|
Had some free time this afternoon, so I played around with Steve's formula and used it to determine my winning percentages. I'm doing my expected values now, and I wanted to confirm my thinking on something. When I'm comparing team A and team B, if team A is the favorite, then I need to consider 3 factors: expected gain from team A if they win, expected gain from team B if THEY win, and the penalty to A if they lose to B. Since the second and third values are the same, I believe this means that the expected value of A winning should be twice the expected value of B winning. Does this sound correct? I started doing it just comparing expected A and expected B directly, and my bracket so far looks pretty normal. Is there a flaw in my thinking somewhere here? Thanks in advance!
|
15 | Sludge
ID: 18116195 Tue, Mar 13, 19:37
|
Expected Winnings of Team A = (Gain for A should they win)*(Probability A Wins) + (Gain for A should they lose)*(Probability A Loses)
Note that "Gain for A should they lose" will be a nonpositive number.
|
16 | Sludge
ID: 18116195 Tue, Mar 13, 20:16
|
Here ya go, steve h et al.
The regression equation to convert from Sagarin ratings to Sauceda ratings:
Sauceda = -424 + 19.8 * Sagarin
To compute probabilities, see the link to Massey's page I gave above.
Enjoy.
|
17 | The Beezer
ID: 191202817 Tue, Apr 03, 21:56
|
Well, I just wanted to thank Sludge and steve houpt for the help on the statistics. Using Sludge's table and s.h.'s formula with the Sagarin ratings, I managed a 5th place finish! Thanks tons guys!
|
18 | HooeyPooey
ID: 1631545 Wed, Apr 04, 05:17
|
Thanks you guys... some how I managed 611th place. :)
|
|
|
Post a reply to this message: (But first, how about checking out this sponsor?)
|