Mathematics Of Football Upsets

[Almighty Colin]

This is pretty interesting if you follow the point spreads or participate in the pools. The chances of an upset in a game fall exponentially as the point spread increases.

Here is a table that demonstrates the likelihood of an upset.

Points Upset%
1 47%
2 44%
3 42%
4 39%
5 35%
6 33%
7 31%
8 28%
9 26%
10 24%
11 21%
12 19%
13 18%
14 16%
21 6.7%
28 2.3%

A 7 point underdog wins 31% of the time.
A 14 point underdog wins 16% of the time.
A 21 point underdog wins 6.7% of the time.
A 28 point undersog wins 2.3% of the time.

You can see the effect of doubling the spread from 7 to 14 decreases by about 1/2 the chance of an upset. It decreases by about another 1/2 when the spread goes from 14 to 21.

[Vick]

Colin is this mathematical probability or done on statistical data from actual NFL games?
Is the win including the point spread or is in the outcome of the game without the point spread?

[Almighty Colin]

Vick -

I thought you'd be the first to post in this thread.

Probabilty AND data. There are white papers that show the error on NFL point spreads is calculated from the standard deviation for football games (14 points) using the bell-curve. The table I displayed is from theory but is very close to the real data.

The win is not including the point spread. In other words, since the Cardinals are 11 point underdogs to the Eagles this weekend, there is about a 21% chance they will win the game outright.

More accurately, 11 point underdogs win 21% of the time.

Since the error is distibuted symetrically - there is also a 21% chance that an 11 point favorite will win by 22 points.



Last edited by Colin at Nov 14 2002, 10:48 AM

[JerryW]

One thought on this is that when picking the spreads, the handicappers are not trying to pick the probability of the winning spread, but rather to insure that equal amounts are bet on each side of the proposition. They do not want to pay the winners from their pockets, but rather the loosers pockets. This will tend to skew the results somewhat from the mathematical curve. :matey:

[Almighty Colin]

Jerry,

It still works out that the error distribution is "normal" because the attempts to equalize the betting on each side of the spread refects the knowledge of the "smart money". What it means is that there is a 67% chance that a game will be within 14 points of the line. Well, that's sort of a lot but the interesting thing is it is symmetrical about the spread. It's still the best bet.

The same thing happens when one used a mathematical betting system. such as Massey ratings - see http://www.mratings.com

The errors are distributed normally because the error in the predictions is reflective of imperfect knowledge and chance with an equal amount on each side of a bet. If enough people catch on to a new idea that is a better predictor than the spread, the spread will move to reflect it.