## Week 7 Game Probabilities

Win probabilities for week 7 NFL games are listed below. The probabilities are based on an efficiency win model explained here and here with some modifications. The model considers offensive and defensive efficiency stats including running, passing, sacks, turnover rates, and penalty rates. Team stats are adjusted for previous opponent strength.

 PWIN Game PWIN 0.71 SD at BUF 0.29 0.35 NO at CAR 0.65 0.24 MIN at CHI 0.76 0.73 PIT at CIN 0.27 0.84 TEN at KC 0.16 0.18 BAL at MIA 0.82 0.14 SF at NYG 0.86 0.82 DAL at STL 0.18 0.09 DET at HOU 0.91 0.47 IND at GB 0.53 0.54 NYJ at OAK 0.46 0.05 CLE at WAS 0.95 0.14 SEA at TB 0.86 0.63 DEN at NE 0.37

### 18 Responses to “Week 7 Game Probabilities”

1. Anonymous says:

On paper I thought this was going to be a tough week, but your model has a lot of heavy favorites. (i.e. SD-Buf, Minn-Chi, Bal-Mia)

2. Brian says:

Yeah there's usually a good amount of upsets, but I only see two this week, Denver(+3) and Green Bay(+1.5), at least according to current lines.

3. Anonymous says:

Hey Brian, went back and reread your signal vs. noise post and was wondering if you might improve predictive power by combining Osacks + Oints into a generic negative pass play (Onpp). Many, though not all, INTs are caused by line pressure and quarterback indecision (and the ones caused by DBs seem to be random) which is also true for sacks.

4. Brian Burke says:

bmoore-I don't think that's a bad idea at all. That might be the key to identifying what drives def ints other than the qb throwing the ball. But I can only count sack data once, either on its own, together with passing eff, or together with 'Onpp.' It's just a matter of which is more predictive. If I can get some time I'll look at that.

5. Anonymous says:

It looks like this equation is overfitting. On any given Sunday the most a team should be a fav is by 80%. Just by the nature of turnovers, officials, etc.

6. Brian Burke says:

I would disagree. If that were right, teams would only very rarely have more than 10 wins or less than 6 in a season. It would be more like baseball, where a team with a .600 win% would have the best record in the league.

Additionally, the vast majority of games would be 60/40 or 55/45 propositions. There would be no way for models like this one and others to have better than 55 or 60% accuracy. So I don't think that could be right.

7. Brian Burke says:

Another note of caution. Last week I warned that we would eventually see some large upsets, and sure enough we did. In fact we saw 2. Keep in mind that even if CLE faced NYG and STL faced WAS again, NYG and WAS would still be heavy favorites, (although slightly less heavy). The 2 most common words used in newspaper articles about both games were "shock" and "stun," as in Browns Shock Giants or Skins Stunned by Rams.

Last week, the model was 8-6, which was ok for a week full of upsets including 2 very surprising ones. Mathematically, the model should be between 70-75% accurate, which in a week of 14 games is about 10-4. Some weeks will be 12-2 and some will be 8-6.

I don't care about absolute accuracy. Because some weeks feature more upsets than others, I compare accuracy with the "consensus" favorites signified by the Vegas spreads. The consensus was only 7-7 last week. So given the topsy-turvy upsets, I'll take 8-6 any day.

8. Anonymous says:

Do the win probabilities factor in home field?

9. Brian Burke says:

Yes. Home teams are favored by about 7% (57% win percentage).

10. Anonymous says:

What is your record so far this season?

How does it compare with Vegas?

Thanks.

11. Brian Burke says:

This might sound like a cop-out, but I promised myself I wouldn't track it week by week this year. I tend to obsess, gloat when it does well, and make excuses when it doesn't. I'm also trying to focus on some other stuff this year. But over the past 2 seasons, this model has been the most accurate I can find, averaging a game or 2 better than Vegas SU. Guys who are regular readers will back me up on that.

12. Brian says:

I've been using the picks in my office pool so I've been keeping track. Starting in Week 5, Brian's record is 26-16, and the consensus favorites are also 26-16.

He had a chance to take a two game lead last night, but the system picked Denver, and Vegas had New England.

13. Anonymous says:

Mr. Burke started his picks in Week 4, though, and here are the weekly results:

Week 4: 10-3 (76.9%)
Week 5: 10-4 (71.4%)
Week 6: 7-6 (53.8%) & one 50/50 game (no-play)
Week 7: 9-5 (64.3%)

TOTAL for 54 games: 36-18 (66.7%)

This is okay, I guess. I'm a little disheartened though, and I ran into this problem last year: Mr. Burke's highest probability winners will lose a couple times each week. Which for betting purposes makes ... and really who gives a crap who wins the game if its not useful for betting purposes ... it an inconsistent starting point.

I'm having a difficult time sticking with Mr. Burke's system over the course of a season, it killed me this past week. I use the following site to convert the probability of winning into cover % (I love this site): http://www.smartcapper.com/cgi-bin/basic/index.pl?cmd=tools&tool=nfl_point_spread_value_calculator&pom_ps=minus&ps=7.5&pom_ml=minus&ml=105&winning_pct=95.00&at=home&period=G.

If he'd only picked New England to win, I'd be sitting pretty. How can Nutshell Retro pick every game winner just like your's except for New England, which his system predicted would win and is killing your system this year in percentage correct? I don't know, I'm thinking of giving up on this site. - Andy

14. Anonymous says:

Shouldn't home teams be favored by 14 percentage points, 57-43, rather than 7 (which is actually a 32.5 "percent" advantage)?

thanks for all the great stuff!

15. Brian Burke says:

Regarding the model's record this year--you guys got me obsessed all over again!

Wk 4: Me 10-3 Vegas 9-4
Wk 5: Me 10-4 Vegas 8-6
Wk 6: Me 8*-6 Vegas 8-6
Wk 7: Me 9-5 Vegas 10-4

Vegas favorites from this site.

That's 2 winning weeks and 1 losing week, for a total of 37-18 (67.2%) vs. 35-20 (63.6%). That's all I can promise about this model--it will be slightly better than the Vegas odds. But don't kid yourself. That's very, very hard to do.

According to predictiontracker.com, this is the most accurate system over the combined past 3 seasons. A 4% edge is just about right for what I've come to expect.

(Remember that this is not primarily designed to be an ATS betting tool.)

*I'm claiming the 50.3% GB vs SEA game! If I got that wrong you guys would rightfully count that against me!

Anon-yes, it's 57-43 'percentage points.' +7 is how I think of it in my head.

16. Anonymous says:

That's right Mr. Burke, come to the dark side!!! But do you think the Smart Capper website _http://www.smartcapper.com/tool_nfl_point_spread_value_calculator.html, which uses the probability of winning (that's where you come in) and average margin of victory from 2002-2006 seasons to predict expected point spreads and money lines and cover % is a reasonable and accurate method of converting your winning probabilities into point spreads? It seems reasonable to me, but I'm neither a genius statistician nor a former air force pilot, so I'd like your opinion.

17. Brian Burke says:

:) Well that makes 2 of us--because I'm not a genius, nor an actual statistician, nor was I an air force pilot.

I was however an F/A-18C pilot in the Navy, and the only probability we ever worried about was a thing called Pk. You don't want to be on the wrong end of that!

Your method seems reasonable to me. But the only method I've ever tested was simply using the SU win predictors as ATS bets. Over the past few years it would yield a 59% win rate ATS. Here's why it works.

18. Anonymous says:

Since I have a tendency to want to fit data to make a model work, I've been playing with Brian's probabilities. There was a previous post which suggested the extreme probabilities might not hold. So I've discarded the 90% + picks as well as games where Vegas makes the favourite more than 10.5. Comparing this ATS gives us a good result.

Also note that Brian is not very good ATS on games that are televised by themselves. (This goes into the category of stretching. lol) For week's 5, 6, and 7 (I've not looked at week 4) Brian is 0 - 6 on Sunday night games and Monday night games. So if we discard all picks over 90%, over 10.5 favoutires and stay away from Sunday and Monday night games we have an excellent record. (And before anyone gets too upset, I am aware of how bizarre this suggestion is.) But it is interesting.