How come it’s never a nice and straightforward slam dunk? To predict the superbowl winner. As it turns out the expected scores for this one are barely different. And what little separation exists appeared only because I added a little touch of (sensible) ad hoc juice.
The FULL New Haven Economics prediction model says: Patriots 33 – Rams 32.
The early model, based on regular season stats favored the Rams 28.4 to the Patriots 27.6
Here’s how it went.
The closeness of the score is interesting because the teams are significantly different when it comes to offense and defense –in opposite directions. Opposite in the sense that it’s a wash. Their respective strengths are offset by their respective shortcomings.
The Rams boast an exceptional offense – almost 40 percent over league average, ranked 2^{nd} in points scored per game. The Patriots have a brutal defense, 15% percent under league average in points scored against them, 7^{th} best.
This combination of a great offense going up against a spectacular defense offset each other.
The Patriots are not too shabby on offense, 4^{th} in the league, coming in at 17 percent above average. The Rams are fair to middling in terms of their defense. They come in at a mere 2 percent over the average, ranked 20^{th} in the league. Again, the combination of an OKish offense facing a soso defense offset each other.
The result of the first iteration of the model based on regular season data alone predicted the Patriots 28.4 to the Rams 27.6.
Conundrum. This is almost impossible to call.
So I included the results from the playoffs into the mix. I did this to use all the data available but realistically to try to achieve a little distance. I know it’s arbitrary – and once you start assuming things there is no end to that rabbit hole. I justify it by saying I am using all the data available.
But as it turns out – in the playoffs the Patriots improved considerably in their offense– but let their defense stall a bit (relatively to the respective averages). And the Rams, did the opposite: they improved their defense and dropped a bit on their offense (again, relative to the season averages).
So I ended up factoring these additional data points into the little model and here ‘tis: what I call the full model.


% Probability of scoring this many points 


Expected Points 
20 
25 
26 
27 
28 
29 
30 
patriots 
33 
0.48 
2.88 
3.63 
4.41 
5.18 
5.86 
6.41 
rams 
32 
0.57 
3.20 
3.98 
4.78 
5.52 
6.16 
6.65 


% Probability of scoring this many points 


Expected Points 
31 
32 
33 
34 
35 
36 
40 
45 
50 
patriots 
33 
6.79 
6.96 
6.92 
6.68 
6.27 
5.72 
3.02 
0.79 
0.12 
rams 
32 
6.94 
7.02 
6.88 
6.55 
6.06 
5.45 
2.73 
0.66 
0.09 
These tables are based on a Poisson distribution and present the probability of each specific score occurring. For instance, the probability that the Patriots score exactly 28 points is 5.18 percent. Given these numbers the chance of the Patriots winning is 56%. And just so you know, FiveThirtyEight give the Patriots a 53 percent chance of winning.
The oddsmakers are posting a 2.5 spread in favor of the Patriots, and a total of 57.5; they are also giving the Patriots a 58 percent chance of winning. The New
Haven Economics model has 3 points, a 55 point total and a 56 percent chance of winning.
Speaking of – the NHE model is an adaptation of one used by David Spiegelhalter for the Premier League and adapted by us here. It is beautiful in its simplicity. To obtain the expected score it starts from the league average and multiplies by offensive prowess of the particular team and also multiplies by the defensive talents of the opposing team. I used data from The Football Database. I also had help and great insights from the students in my Introduction to Business Analytics classes.
So there you have it.
Write to me.
arod
arodriguez@newhaven.edu
Comments
How was the adaptation from soccer to football done with the NHE model? And according to all these models, we should have ourselves a good Super Bowl. Good work arod