By Blaidd Drwg
The great thing about baseball is that you can generally use advanced metrics to make a prediction about the performance of a team in an upcoming season with reasonable accuracy. There are certainly things you can’t predict (injuries, guys significantly over/under performing, luck, etc.) but those metrics have been tested and tweaked to give you a pretty reasonable picture of what will happen in the upcoming season.
Football, for whatever reason, doesn’t seem to enjoy the same level of prediction accuracy. Back in August, ESPN had their NFL preview and included projected standings based on a computer simulation. Here is what we got:
| Team | Overall W-L | Home W-L | Road W-L |
| MIA | 10-6 | 7-1 | 3-5 |
| NE | 9-7 | 8-0 | 1-7 |
| BUF | 5-11 | 5-3 | 0-8 |
| NYJ | 4-12 | 4-4 | 0-8 |
| CIN | 11-5 | 7-1 | 4-4 |
| BAL | 9-7 | 8-0 | 1-7 |
| PIT | 9-7 | 7-1 | 2-6 |
| CLE | 5-11 | 4-4 | 1-7 |
| HOU | 11-5 | 8-0 | 3-5 |
| IND | 8-8 | 6-2 | 2-6 |
| TEN | 6-10 | 6-2 | 0-8 |
| JAX | 4-12 | 4-4 | 0-8 |
| DEN | 13-3 | 8-0 | 5-3 |
| KC | 10-6 | 7-1 | 3-5 |
| SD | 5-11 | 5-3 | 0-8 |
| OAK | 4-12 | 4-4 | 0-8 |
| WAS | 10-6 | 8-0 | 2-6 |
| DAL | 8-8 | 7-1 | 1-7 |
| NYG | 8-8 | 7-1 | 1-7 |
| PHI | 6-10 | 6-2 | 0-8 |
| GB | 11-5 | 8-0 | 3-5 |
| CHI | 9-7 | 8-0 | 1-7 |
| MIN | 7-9 | 7-1 | 0-8 |
| DET | 5-11 | 4-4 | 1-7 |
| ATL | 11-5 | 8-0 | 3-5 |
| TB | 9-7 | 7-1 | 2-6 |
| CAR | 7-9 | 7-1 | 0-8 |
| NO | 6-10 | 6-2 | 0-8 |
| SEA | 13-3 | 8-0 | 5-3 |
| SF | 13-3 | 8-0 | 5-3 |
| STL | 7-9 | 7-1 | 0-8 |
| AZ | 3-13 | 3-5 | 0-8 |
It wasn’t the most accurate prediction as they only got 4 out of the 8 division winners correct and 2 out of the 4 Wild Card winners correct. The win totals look reasonable on a cursory level until I actually looked at what made up the records. According to the simulation, 10 teams would go undefeated at home in 2013, 10 teams would go 7-1 and only 6 teams would be .500 or worse. On the flip side, they predicted that only 4 teams would be at least .500 on the road (with no one going better than 5-3) and 19 teams would be either 1-7 or 0-8 away from home.
Now I don’t know exactly what went into the programming of the simulation, but let me tell you, this just looks wrong. It seems that the programmers put too much emphasis on home field advantage and caused some whacky results. I am surprised that they let this be published, given that any average football fan would realize these numbers just look wrong. Just how wrong are they? Well, I decided to look at the road records and over the past 11 NFL Seasons (2002 – 2012), there have been 13 teams that have failed to win a game on the road, which is about 4% of the teams. The prediction for 2013 was for 10 teams to go winless away from home, or 37%. On the flip, over the same period, 46% of NFL teams played at least .500 ball on the road. The computer for 2013? Just 12%. Um, I am pretty sure that you have a significant error in the calculation here.
In some ways, I am comparing apples to oranges by looking at the historical numbers. How did the computer actually do with its predictions? Well, here you go:
| Number of Wins | Home Prediction | Home Actual | Road Prediction | Road Actual |
| 0 | 0 | 0 | 12 | 1 |
| 1 | 0 | 2 | 7 | 6 |
| 2 | 0 | 1 | 4 | 4 |
| 3 | 1 | 5 | 5 | 8 |
| 4 | 5 | 6 | 1 | 6 |
| 5 | 2 | 7 | 3 | 2 |
| 6 | 4 | 5 | 0 | 5 |
| 7 | 10 | 3 | 0 | 0 |
| 8 | 10 | 3 | 0 | 0 |
Those numbers look pretty bad in comparison, especially at the upper and lower ends of the spectrum.
How about total wins? Well, that looks a little better, but only because the increased number of bands flattens out the distribution:
| Number of Wins | Projected | Total |
| 13 | 3 | 2 |
| 12 | 0 | 3 |
| 11 | 4 | 4 |
| 10 | 3 | 2 |
| 9 | 5 | 1 |
| 8 | 3 | 7 |
| 7 | 3 | 4 |
| 6 | 3 | 1 |
| 5 | 4 | 1 |
| 4 | 3 | 5 |
| 3 | 1 | 1 |
| 2 | 0 | 1 |
The moral of this story is if you are trying to figure out how many games your team will win in 2014, take a look at their schedule, go through it game by game and predict a winner. My guess is that you will be more accurate than the computer.
FlatbreadGarden 
It looks to me like at least part of the issue is that they treated each individual game wins/loss as a strictly binary proposition… and that’s a big issue.
The more correct answer would be to simulate each teams’ chance to win each game, then simulate the season 10,000 times (or whatever). That would mostly get rid of the 8-0 home teams and 0-8 road teams.
LikeLike
Wait…there’s a whole _team_ that’s Missing In Action? How’d that happen?
LikeLike