Adjusting "player unit" based on position
The scheme for rating injury values from the official NFL injury report released each week, was described in detail in the Evaluating Injuries article in the main research section.
Now one premise of that scoring mode for trying to quantify the value of an injury is that all positions are counted equally. Of course, a lot of people would have you believe that the value of an injury at one position is NOT likely to be equivalent to another, even in a general, league-wide sense.
It's easy enough for us to implement adjustments to the basic setup based on position, and so we have developed such a program. However, the key issue remains, what adjustments should be made?
The betting public typically reacts the strongest to an injury to the Quarterback, but for just that reason we elected to "de-value" the QB injuries. Since it is commonly believed that the passing game is the key to winning in the NFL, it would make sense we thought to promote the value of the players most immediately involved in the passing plays (other than the QB).
Consequently we arbitrarily set the weighted values as follows:
- 1.5x value - WR, T, DE, CB, S
- 1.4x value - C
- 1.3x value - DT
- 1.2x value - RB, TE
- 1x value - all other positions
Any number of arguments could be made against the scheme, but that's what we started with, pre-testing. In addition, we didn't look at the "This Week" ratings only, since of the three (This Week, Change Week, a combination of the two) it has shown the least signs of being effective.
"Change Week" score difference
Player Units
Edge |
0.1 - 0.9 |
1.0 - 1.9 |
2.0 - 2.9 |
3.0 - 3.9 |
4.0+ |
ALL |
| HomeFavs |
25 - 25 |
17 - 27 |
9 - 6 |
8 - 6 |
10 - 5 |
69 - 69 |
| HomeDogs |
13 - 10 |
14 - 5 |
9 - 7 |
6 - 4 |
10 - 1 |
52 - 27 |
| AwayFavs |
11 - 9 |
6 - 11 |
7 - 5 |
7 - 3 |
5 - 2 |
36 - 30 |
| AwayDogs |
20 - 23 |
21 - 15 |
8 - 10 |
11 - 3 |
8 - 9 |
68 - 60 |
| Favorites |
36 - 34 |
23 - 38 |
16 - 11 |
15 - 9 |
15 - 7 |
105 - 99 |
| Underdogs |
33 - 33 |
35 - 20 |
17 - 17 |
17 - 7 |
18 - 10 |
120 - 87 |
| Home Teams |
38 - 35 |
31 - 32 |
18 - 13 |
14 - 10 |
20 - 6 |
121 - 96 |
| Away Teams |
31 - 32 |
27 - 26 |
15 - 15 |
18 - 6 |
13 - 11 |
104 - 90 |
| ALL PICKS |
69 - 67 |
58 - 58 |
33 - 28 |
32 - 16 |
33 - 17 |
225 - 186 |
| WIN % |
51 % |
50 % |
54 % |
67 % |
66 % |
55 % |
| Spread Range |
10+ points |
5 to 9.5 |
0 to 4.5 |
| Favorites |
7 - 11 |
37 - 36 |
61 - 52 |
| Underdogs |
11 - 9 |
51 - 32 |
58 - 46 |
Analysis: Can't complain about the look of the 3+ difference plays -- good for a 65-33 (66%) mark. Of course, the straight weights (all positions count for 1x) showed a 41-17 record at 3+ so the win percentage actually dropped slightly, but you added another 24-16 (60%) to the total, giving a greater level of confidence that this isn't just an aberration. Re-running the test using a minimum of 3.0 reveals the following spread ranges for the prime zone:
| Spread Range |
10+ points |
5 to 9.5 |
0 to 4.5 |
| Favorites |
4 - 3 |
7 - 7 |
19 - 6 |
| Underdogs |
5 - 3 |
15 - 6 |
15 - 8 |
| All |
9 - 6 |
22 - 13 |
34 - 14 |
That's a strong record in the games with small favorites (0 to 4 1/2 points), but a 71% effort is less significant than the 66% mark over 98 games.
Looking at the combined method (play teams with healthy edges in both This Week and Change Week scores) shows an actual decline when adjusting by position versus treating all players as the same.
Another idea for a weighting scheme was to value the linemen as more important than the other positions, in that the linemen are generally lesser known players to the football fan public and probably are more readily overlooked by the "squares" as a result. Giving all linemen a 1.5x value produced some interesting data: the Change Week results were 57-30 at the 3+ difference threshold, and 94-57 (62%) at the 2+ level. That's good going, given we're talking about 4-5 plays a week on average over the two year span of the database.
Still, without more data to look at, it's hard to make a strong case that tweaking the numbers like this will offer much better long term results.
|
