## Game Probabilities - Week 9

Weekly game probabilities are available now at the nytimes.com Fifth Down. This week I also lead-in with break down of the Chargers-Giants matchup.

### 6 Responses to “Game Probabilities - Week 9”

1. Happy says:

I was really suprised to see how many people have NYG as strong favorites. I have the game as more or less a toss up. I see that you do too. By the way, I've converted your probabilities to spreads (in the NY Times comments). I'll put it here too. We more or less agree, though I don't have NE covering against MIA or Seattle covering against Detroit.

2. Anonymous says:

JMED 2 in GA
are you gonna post your college picks? Would like to keep seeing them if you don't mind? If you are posting them in different arenas could you share where?
Thanks

3. Guy says:

Brian: I know you adjust home field advantage based on the talent spread of the two teams, with HFA being smaller as the teams are more unequal. But I worry that this is a function of your looking at this retrospectively, using game outcomes based on two teams' record in that same season. The chance of a team with a better record winning that game is not the same as the chance that team would win a future game against the same opponent, because the outcome of this game is part of the team record you're looking at.

As a result, I think if you look at the outcomes of these games, you will find that the favorite wins more often than it "should." For example, a 12-4 team almost never loses against a 3-13 team, because the probability this game accounts for both one of the weak team's rare victories and the strong team's infrequent losses is very low -- even lower than the probability of an upset between these 2 teams in the future. And the huge probability of a win by the favorite swamps HFA, making HFA appear smaller.

Conversely, consider two 8-8 teams. That is really a match between a 7-8 team and an 8-7 team. BUT, we know that the weaker team must win this game (to get to 8-8). So it must be more than a 50% chance that the 7-8 team is playing this game at home (both because we know they will win this game, and because they have the worse record outside of this game).

It would be interesting to repeat your analysis, but using only the team records for games other than the game being played. I suspect you will then find a more uniform HFA.

Another way to look at this is to look at HFA by team record. Do teams that win, say, 7-9 games actually have a higher HFA than those that win fewer or more games? If your model is right, than I think they should (because they play more games against closely-matched opponents).

4. Brian Burke says:

Guy-I don't deliberately build an adjustment in the size of the HFA based on team strength. It's just a function of a logistic probability function for any predictor. When there are multiple predictors for an outcome, and most of them even out, any one remaining predictor will be more influential. It's just a natural result of the logit math.

HFA in the NFL is about 7%. Consider a mismatch where the game is a 95/5 matchup before we consider HFA. If we just added 7% we'd have a >1 probability, which isn't possible. The size of HFA (in terms of %) must decrease as the degree of mismatch increases.

The graph I pointed to is based on each team's ultimate season record, not record to date, so the effect you suggest isn't possible. Also, see the graph in this post on the NBA. (Darn, firefox won't let me paste the URL here.) Go to the search box and enter 'bias nba'. Look at the last result. The NBA, with a much larger schedule, shows the same phenomenon very clearly.

5. Guy says:

Brian: Thanks for reply. Now that I've read more of your posts and the comments, I see what you're saying: HFA is worth a certain point advantage (about 2.5), and that point advantage has more win value when the teams are closely matched. Essentially, you're saying there is a diminishing return for points scored (or points prevented on defense), which must be true.

FYI: I looked at teams 2006-2008 based on season record -- 10+ wins, 7-9 wins, 0-6 wins --and found these results:
Wins/Win%/Home win%/HFA
10-16 .724 .771 +.047
7-9 .502 .570 +.068
0-6 .248 .293 +.045
Seems consistent with your data, though perhaps suggesting a narrower range for HFA.

I still think that comparing HFA and same-season win differential (as in your NFL graph) is potentially skewed a bit, because the season record includes the matchup you are looking at.
Isn't it true that the favorite wins "too many" games at each point differential (compared to what you would predict for a future game between such teams)? In the NBA, where each game is only 1/82nd of the team record, I'm sure this is trivial. But in NFL, I would think it could be a small problem.

6. Anonymous says:

Any college picks for next week?