## Weekly Game Probabilities

Weekly game probabilities are available now at the nytimes.com Fifth Down. I take a detailed look at the NYG-NE match-up. Also, there are a couple very kooky games this week. Please see my comment under the main post for a brief explanation.

### 18 Responses to “Weekly Game Probabilities”

1. Steve Freeman says:

Brian: Your model predicts Washington over San Francisco, when of course San Francisco is a solid favorite. You wrote recently, though, that the Redskins stretched too thin and a late season fade is to be expected.
Does your model take “Redskin fade” into account? Do you weight recent games more heavily?

2. Justin says:

Is there a chance Indy favored over Atlanta is a mistake?

3. Dan Whitney says:

Brian, I have noticed that your model seem to heavily favor home teams, as in in most weeks only 1 or 2 road teams will be favored in your model. Does this perhaps indicate in issue in regressing all of the other numbers towards the mean for the season while home field does not regress to the mean? As a result there would be less of a gap between other numbers and home field would be overly emphasized. Once again thanks for all your great work, Dan

4. Dan Whitney says:

Steve, Brian has said on many articles that his models do not take into account injuries or emphasize recent performance. Thus the models will not account for the redskins injuries or the "redskin fade". Dan

5. Mo says:

Brian, I have been saying for a almost a week that Indy is a lot better than it seems. I think this may be the game that shocks people. If people think ATL is going to have a blowout, what do you think ATL thinks of themselves? Picking Indy .....

6. Anonymous says:

I had virtually the same question/issue as Dan. The model only has one road team, GB, winning this week. Even in that game, the model gives the home team, SD, a 48% chance of winning (which is a much greater chance of winning when compared against the Vegas lines). A gambler using the model's winning percentages would be betting home teams almost exclusively. Has the model's view on home field advantage been changed or tweaked in the last year? If so, what was the reason for that change? Thanks for the great work. Jason

7. Anonymous says:

While the model has only 1 away team as a favorite, Vegas lines only have 3 this week - it's only two games difference.

8. Anonymous says:

Yeah, but its a systemic thing, as Dan previously pointed out. It is pretty apparent right now that the model has a bigger home field advantage than the Vegas lines. I'm not saying that the model is right or wrong about how much home field advantage means but its clear that something is going on there. Jason

9. James says:

As Brian posted earlier this year, the updates to the model lowered the highest GWP and strengthed home field advantage due to past results.

The slightly weaker road teams and stronger home field advantage mean home teams will be favorites much more often than before.

10. Anonymous says:

Honestly, I have a problem with most of these probabilities. I don't know what goes in to the model, but I have a hard time believing the Jets win only 32% of the time (I'd consider this perhaps a 45-55% games; Fred Jackson can't even be just a very good back because Buffalo's mostly pedestrian wide receivers will be shut down in large part). I think Tampa Bay should have about twice as much of a chance of winning, 30-35%, against New Orleans, and I can't imagine the Bears being so unfavored against the Eagles.

And of course the Redskins are going to lose to the Niners and the Colts will probably lost to the Falcons.

I'd just tinker with the home field advantage a bit...

11. Brian Burke says:

HFA is only very slightly stronger than in recent seasons. You really couldn't notice the difference without looking at the thousandths place on the coefficient.

Actual home team win rates vary widely from year to year. I don't chase the HFA of the last couple years. It's an average that goes through 2002.

The reason it appears stronger is because in a few recent matchups, the model has the two teams much closer in strength than most other people. ATL is very close to IND in the rankings, and WAS is ranked too high given their current injury issues.

12. Anonymous says:

So far this season, home teams are 72-44 (a 62% winning percentage). The model, however, has so far predicted home teams to go 53-15 (a 78% winning percentage). The difference in games is due to the fact that the model does not start picking until week 4. I'm not sure what the winning percentage has been historically for home teams but I'm sure it isn't 78%. Although certainly not conclusive, these numbers suggest to me that the model may be overweighing home field advantage. If I'm wrong about this, I'm sure that a reader or Brian can point out what I'm missing. Thanks. Jason

13. Steve Freeman says:

jason: you need to compare apples with apples: what is home team record since week 4?

14. Dan Whitney says:

I still think that the inherent problem with over-emphasizing home field advantage is the regression to the mean. I don't know when Brian stops regressing the numbers, but by making every team close to the league average, all the differences in other stats will be lessened, and thus home-field will play a bigger role in the prediction analysis. It would be interesting to go back and look at past years and see if the percentage of home teams favored by this model decreases as the season goes on (which I think it will) I don't think their is an easy fix, so it may just be an inherent flaw in the system (maybe you could somehow regress the 1 for home-field towards the mean by the same factor to keep it consistent with the other inputs.) Anyways that's my thought on this, keep up the great work-Dan

15. Mike B says:

Jason - Pretty much any model is going to "overweight" home field advantage in the sense that the actual home winning percentage will be less than the predicted home winning percentage.

I looked at thepredictiontracker.com csv file for 2010. The "line" had the home team as the favorite 68% of the time, but the home team won 56% of the time. This doesn't mean the line was flawed. It's more of a reflection of our relative ignorance.

Home field advantage is a consistent, persistent phenomenon that stays relatively constant from year to year. It's based on the outcomes of literally thousands of games.

On the other hand, we've seen most 2011 NFL teams play 7 games so far. We know far less about any particular team's "true strength" than we do about home field advantage. Our relative ignorance between the two leads us to overpredict home winners.

16. Andrew Foland says:

If I take the GWP table from this week, I can reproduce the game probabilities for this week very accurately (RMS difference well under roundoff error) with a logit-advantage for the home team of 11.2% (i.e. between two evenly-matched teams, the home team would win 61.2% of the time.)

Historically, about 57% of home teams win. There is indeed some convexity induced by the logit function among mismatched teams, so that the average observed HFA is not exactly equal to the HFA between two evenly-matched teams. Using a standard deviation among teams in GWP of 0.1 (taken from this week's GWP table), and generating many fictional matchups with a logit calculation, the effect is about 1%--namely, to generate an average home win percentage of 57%, the logit HFA needs to be about 58%.

In an earlier column, Brian mentioned that evenly-matched teams can average up to 63% home wins, which is reasonably consistent with the 11% I find. These are, respectively, 6% and 4% larger than the observed HFA.

But I don't really see how that's consistent with the fact that overall the home team wins 57%, even once the convexity of mismatches is taken into account. I find I cannot really make the effect as large as 4%, let alone 6%, even for absurdly large values of the standard deviations among teams (and correspondinly high frequencies of large mismatches.)

17. Brian Burke says:

Dan's explanation above is correct. The regression of the input stats is beginning to recede at this point in the season. This is when the SD of most of team stats stabilize at the season-ending spread.

18. TB says:

I compared Brian's numbers with the late money line, and this season Brian's model favored the home team against the money line in 43/81 (53%) games. So there's no obvious or overall discrepancy between the model and Vegas regarding HFA.