Converting NFL Predicted Scores Into Win Probabilities

[Sportslover]

If I predict an NFL game will end up with Team A scoring 24 points and Team B scoring 21 points - how do I calculate the probability of Team A winning the game?

Also, how would I calculate the probability of Team A covering a spread of -7?

I've read that you need to take into account the total score, not just the difference between the 2 teams, in order to get an accurate probability. Can anybody tell me how to do this?

[Jaug]

You could assume that scoring is poisson distributed in steps of 4 (?) points and use your estimates as mean values. This is probably not the optimal way to do it but could be a start.

Poisson dist will give you for each team respectively:

Score prob
0 5%
4 7%
8 10%
12 12%

And with such data it is easy to estimate probability of team A covering spread of 7.

Not sure this is a great way to do it but it will probably help start a discussion.

[Justin7]

Poisson isn't really very good for predicting score distributions, at least not in this manner.

One approach is to turn your result into a spread. If you predict 21-24 for the home team, your spread is home -3. You could then convert that to a moneyline or win percentage, which suggests around 60%.

[Sportslover]

Poisson isn't really very good for predicting score distributions, at least not in this manner. One approach is to turn your result into a spread. If you predict 21-24 for the home team, your spread is home -3. You could then convert that to a moneyline or win percentage, which suggests around 60%.

Thanks for your reply.

Is this done by knowing the value of the -3 in NFL to be roughly around -160?

Or is this done by doing a normal distribution of the chances of covering 0.5, when you expect the score difference to be 3, with a standard deviation of "score error against the spread"?

Basically what I'm asking is which of the above methods would be most accurate for getting the Moneyline/Win Probability estimation, if using the spread as the value to determine this? Or, is there another method that is more accurate?

Also, wouldn't turning the result into a spread be inaccurate because the win probability for a -3 spread would be different for a 17-14 score than it would be for a 24-21 score? Or is this inaccuracy balanced out by the fact that it works for you sometimes and works against you other times?

Is there a more accurate method that also takes into account the expected total points of the game?

[uva3021]

pythagorean:

24^2.37/(24^2.37+21^2.37)

for a one game situation the exponent might be a little higher

[WendysRox]

I was googling for the correct exponent to use in pythagorean method for football and found this thread. I've read that 1.85 is pretty good for baseball and I remember reading somewhere that NFL was over 2. Thanks for posting this and if you happen to have references for this exponent, I'd greatly appreciate it.

[buby74]

How did you arrive at the 24-21 score initially? As that same method should be used to determine the variance in your prediction (assuming your initial method is accurate, if it is not then your 24-21 estimate can also be dumped)

On a related issue I have found very little evidence that totals actually affect pointspread in reality as opposed to theory but I would love to see data proving me wrong