Read this for a description of APR
Introduction
The following is an analysis of APR pick reliability for regular season games.
APR new margin power vs. classic margin power
You may remember at the beginning of the season I announced that I would be using a new margin power function for the APR power rankings this year. APR's picks for this year finished at 152-88. This turns out to be 6 picks weaker than the original 'classic' margin function for the same 2009 games.
The new and classic versions of APR were actually mostly in agreement on game picks; they only picked 8 games differently (correct picks are shown in bold):
| Week | Game | Pick (current) | Pick (classic) | |||
| 3 | Falcons | 10, | Patriots | 26 | Falcons | Patriots |
| 4 | Jets | 10, | Saints | 24 | Jets | Saints |
| 5 | Patriots | 17, | Broncos | 20 | Patriots | Broncos |
| 10 | Bills | 17, | Titans | 41 | Bills | Titans |
| 11 | Chargers | 32, | Broncos | 3 | Broncos | Chargers |
| 12 | Giants | 6, | Broncos | 26 | Broncos | Giants |
| 12 | Steelers | 17, | Ravens | 20 | Steelers | Ravens |
| 14 | Redskins | 14, | Raiders | 13 | Raiders | Redskins |
Over the course of the season, I've become increasingly dissatisfied with the new margin power function; this tends to confirm my suspicion that it's not really an improvement.
APR: Home vs Visitor
Here's the breakdown of pick accuracy for picking the visiting or home teams.
| Visitor Won | Home Won | Total | |
| Picked Right | 66 | 86 | 152 |
| Picked Wrong | 37 | 51 | 88 |
| Total | 103 | 137 | 240 |
APR picked the home team to win 123 times (86+37), again noticably less than the 137 home teams that did win.
As with last year, this indicates that something needs to be changed so that the home team is picked more often. This year, I think I know what that change needs to be: each team needs to have separate home and away power rankings, and the games picked according to the appropriate power.
If you look at the regular season standings, there are some teams that have a noticable difference between home and road records:
| Team | Home Record | Road Record | Difference |
| Patriots | 8-0 | 2-6 | 6 |
| Vikings | 8-0 | 4-4 | 4 |
| 49ers | 6-2 | 2-6 | 4 |
| Ravens | 6-2 | 3-5 | 3 |
| Steelers | 6-2 | 3-5 | 3 |
| Falcons | 6-2 | 3-5 | 3 |
| Bears | 5-3 | 2-6 | 3 |
| Seahawks | 4-4 | 1-7 | 3 |
I will be very surprised if separating power rankings into home and away categories doesn't yeild at least some improvement in pick reliability.
APR vs. Pythagorean Projection
For week 1 of the season, I posted game picks based on the Pythagorean projections from last year. The Pythagorean picks for this year were 164-92-0 (64.1%), which beats APR (and even APR classic) this year. A 64.1% successful pick rate is best Pythagorean Projection has done at least since 1993 (see results for previous years here, along with an important update).
Conclusions
APR continues to be a work in progress. For next year, I will implement the home/away power rankings for game picks (the weekly power ranking posts will still show a combined power ranking, though).
I'm going to abandon the current margin weight function, either in favor of the classic function, or something better (if I can find something better).
I'm also going to adopt the Pythagorean Projection pick method for week 1 games, so that APR can pick the full season of games.
