Code and Football

Well, the system went 0-2, and probably would have gone 1-1 had the refs been able to call pass interference in the last two minutes of the Saints-Rams game. The story as I gather it, is the lack of experience in the current crop of referees and the lack of good positioning during the infraction. Count me among the folks who think “the booth” should be able to overrule this kind of blatant miss on the field.

The other factors that don’t seem to change is that New England outplays its strength of schedule and that Kansas City underperforms its playoff predictions. All that said, the formula predicts a close game with the Rams emerging with a victory.

Not much to say, other than my system went 4-0 predicting winners. It did predict a bigger margin of victory in the Kansas City game, and closer games than most would have expected in the Saints game and Rams game. I don’t think in all honesty, that my odds were that much different from Vegas odds.

Once again, my data favor the home team, and by more than HFA. In both cases the home teams faced tougher competition throughout the year than the challenger.

 
In this instance the old and new formulas are close in terms of their predictions. That is because the strength of schedule adjustments between the teams are a little larger in the old formula.

In terms of picking winners, the system went 2-2, unable really to deal with tough underdogs such as the Chargers and Colts. It picked Philadelphia, which by traditional means was the most in favor of the home team, though that game was one foot from being a Chicago win. So it went 2-0 in the NFC and 0-2 in the AFC.

In this round the home teams are favored in all four contests, but by varying amounts compared to the spread.

The methodology of how we pick is given here. The 2018 worksheet is given here. And as an aside, Doug Farrar’s new football book is very very good and I recommend that hard core fans buy it.

In the worksheet below, the factor 0.66 is the logit of home field advantage as calculated by the logistic regression. That’s equivalent to a HFA of 4.9 points. The playoff HFA of 62.7% is equivalent to 3.8 points. So, if you prefer 3.8 or even 3, just subtract 1.1 points or 1.9 points from the points margin respectively. Just for yucks we calculated the Rams and Cowboys odds both with the 0.66 factor of the fitted formula and the 0.518 factor of actual results, the latter in parentheses.

Whether I stick with this new formula is up in the air. I have an older formula that is much the same but not inclined to generate 15 point advantages, a bit tamer, if you will. We’ll see. I don’t do this for a living, just for fun, and the methodology link above gives the old formula.

That said, the second round worksheets.

 
Update: decided to add the old formula predictions, and also use the measured HFA factor.

 

It’s a new playoff season, and another time to try our new playoff formulas. Methodology of this work is described in depth here.

The playoff formulas like New Orleans and Kansas City. They like Baltimore, but Baltimore, which will lose home field after the first round, is unlikely to be favored after that point. The formulas place a substantial penalty on the lack of playoff experience, and so does not favor Chicago, the Chargers, or the Colts. Update: Baltimore has not been in the playoff since 2014, and so the results have been amended.

 
The total score of a particular team is used as a base. Subtract the score of the opponent and the result is the logit of the win probability for that game. You can use the inverse logit (see Wolfram Alpha to do this easily) to get the probability, and you can multiply the logit of the win probability by 7.4 to get the estimated point spread.

Because the worksheet above can be hard to decipher, for the first week of the 2018 playoffs, I’ve done all this for you, in the table below. Odds are presented from the home team’s point of view:

But to summarize, the formulas used here were found by logistic regressions and each element in the formula has a playoff significance of 95%. I promise if the more common offense metrics could say that, they would. I’ll also note that in vogue stats like FPI don’t really give answers markedly different from other common offensive metrics, such as Pythagorean expectation.

That said, offensive metrics like Pythagorean Expectation favor Seattle over Dallas by about half a point, or 52% win probability for Seattle. Offensive stats still favor Baltimore, but not as much. Simple Ranking stats favor Chicago by around 8 points, circa 75% WP. Houston-Indianapolis have approximately even offensive stats, so the difference between the teams is about 3 points. HFA is worth a bit more in the playoffs, circa 63%.

Final games of the season! Philly makes it, and Minnesota does not. Time now to crunch some playoff numbers.

Global Statistics:
Games  Home Wins HwPct Winning_Score Losing_Score Margin
256       153     59.8      28.70        17.60     11.09

Calculated Pythagorean Exponent:  2.86


Rank  Team    Median  GP   W   L   T  Pct   Pred   SRS    MOV   SOS
------------------------------------------------------------------------
1     NO       8.0    16  13   3   0  81.2  73.5  10.07   9.44  0.63
2     LA       6.0    16  13   3   0  81.2  71.2   8.49   8.94 -0.44
3     KC       7.0    16  12   4   0  75.0  69.9   8.89   9.00 -0.11
4     CHI      7.0    16  12   4   0  75.0  75.7   6.28   8.62 -2.34
5     LAC      4.0    16  12   4   0  75.0  68.0   5.95   6.19 -0.23
6     NE       9.5    16  11   5   0  68.8  69.9   5.18   6.94 -1.76
7     HOU      3.0    16  11   5   0  68.8  66.6   3.85   5.38 -1.53
8     BAL      5.5    16  10   6   0  62.5  70.5   7.02   6.38  0.64
9     IND      3.0    16  10   6   0  62.5  65.9   3.36   5.56 -2.21
10    SEA      3.0    16  10   6   0  62.5  64.6   4.51   5.06 -0.56
11    DAL      2.5    16  10   6   0  62.5  53.2   1.09   0.94  0.15
12    PIT      3.0    16   9   6   1  59.4  62.1   5.57   4.25  1.32
13    TEN      3.0    16   9   7   0  56.2  51.6   0.23   0.44 -0.21
14    PHI      2.5    16   9   7   0  56.2  53.8   1.73   1.19  0.55
15    MIN      1.0    16   8   7   1  53.1  53.9   0.59   1.19 -0.60
16    CLE     -1.0    16   7   8   1  46.9  43.7  -0.33  -2.06  1.73
17    ATL     -2.0    16   7   9   0  43.8  48.5  -0.11  -0.56  0.45
18    CAR     -2.0    16   7   9   0  43.8  48.9   0.93  -0.38  1.30
19    WAS     -5.0    16   7   9   0  43.8  33.2  -4.95  -4.88 -0.07
20    MIA     -6.5    16   7   9   0  43.8  29.4  -8.83  -7.12 -1.71
21    GB      -2.5    16   6   9   1  40.6  45.6  -2.74  -1.50 -1.24
22    DEN     -2.5    16   6  10   0  37.5  45.8  -0.51  -1.25  0.74
23    DET     -2.5    16   6  10   0  37.5  42.5  -3.01  -2.25 -0.76
24    CIN     -4.0    16   6  10   0  37.5  35.3  -3.41  -5.44  2.02
25    BUF     -5.5    16   6  10   0  37.5  28.0  -6.90  -6.56 -0.34
26    NYG     -2.5    16   5  11   0  31.2  42.2  -2.17  -2.69  0.51
27    TB      -3.0    16   5  11   0  31.2  38.9  -2.56  -4.25  1.69
28    JAX     -3.0    16   5  11   0  31.2  32.6  -4.02  -4.44  0.41
29    SF      -4.5    16   4  12   0  25.0  33.5  -5.54  -5.81  0.28
30    NYJ     -7.0    16   4  12   0  25.0  30.9  -7.84  -6.75 -1.09
31    OAK    -14.0    16   4  12   0  25.0  20.4  -9.31 -11.06  1.75
32    ARI    -11.0    16   3  13   0  18.8  14.0 -11.50 -12.50  1.00