Sagarin says

**td577** · 10-23-2014, 03:55 AM

I find all the math kinda interesting. I see how one gets 45.1% for 12-0 by multiplying all the probabilities together, but how do they get the other percentages?

Sent from my SM-T230NU using Tapatalk

**Strategery** · 10-23-2014, 04:33 AM

Originally Posted by td577

I find all the math kinda interesting. I see how one gets 45.1% for 12-0 by multiplying all the probabilities together, but how do they get the other percentages?

Sent from my SM-T230NU using Tapatalk

Since the probability of winning any game is the percentage given, the probability of losing is 100 minus that percentage. Therefore, to get, for example, the probability of going 11-1, you need to add up all the combinations of multiplied percentages that include exactly one loss from remaining games.

Audit can correct me if I'm wrong but I worked it out for the 11-1, 8-4, and 7-5 scenarios and nailed 40.1%, 0.1%, and 0.002%. Would take a little longer for 10-2 & 9-3 since there are quite a few more combinations and I'm just using my phone calculator.

**bisonaudit** · 10-23-2014, 10:39 AM

Originally Posted by Strategery

Since the probability of winning any game is the percentage given, the probability of losing is 100 minus that percentage. Therefore, to get, for example, the probability of going 11-1, you need to add up all the combinations of multiplied percentages that include exactly one loss from remaining games.

Audit can correct me if I'm wrong but I worked it out for the 11-1, 8-4, and 7-5 scenarios and nailed 40.1%, 0.1%, and 0.002%. Would take a little longer for 10-2 & 9-3 since there are quite a few more combinations and I'm just using my phone calculator.

you got it dude.

**westnodak93bison** · 10-23-2014, 12:11 PM

Originally Posted by westnodak93bison

What are the odds of holding USD scoreless for three straight years?

Audit, how would you figure this one out?

**bisonaudit** · 10-23-2014, 04:14 PM

Originally Posted by westnodak93bison

Audit, how would you figure this one out?

A methodology that is relatively straight forward is discussed in Mathletics, but getting the data and the computing power to do it right is an issue.

If you had all the scores for all the games you could compute an offensive and defensive rating for each team and understand how much the model's predictions vary from reality which would then give you an idea of how likely a particular defense would be to pitch a shutout against a particular offense.

In college football you've got like 250 teams x 2 variables per team (off & def) + 1 for home field + whatever else you want to throw in, so 501 variables or more. Excel's solver can't handle that many variables. It can do a single conference or even the whole NFL (65 variables) but not 500+.

If we just use the MVFC conference games played to date (which is only 3 or 4 games per team so, sample size issues!, plus you're willfully ignoring a bunch of information you know is relevant!) such a model suggests that NDSU's defense is 13 points better than the conference average while USDs offense is 9 points poorer than the conference average. Those are top and bottom of the table respectively. The predicted game score: NDSU 39 USD 4 (which is way more than the 25 point margin suggested by Sagarin even though the conference only model's home field advantage is almost x2 Sagarin's). Comparing the model's predictions to actual results to date, the standard deviation of predicted points allowed compared to actual is 5.58 and the errors are normally distributed around the mean so we can use the normal distribution curve to estimate roughly a 20% chance of a shut out on Saturday.

Again, that's just based on performance in the 3 or 4 conference games played to date and it's completely ignorant of everything except the final scores and who's field the game was played on. But's that's a way you could get at it.

**Bison bison** · 10-23-2014, 06:48 PM