# Forecasting percentages

#### haz33

##### New Member
I'm looking for the proper way to forecast shooting percentages in a basketball game.

Say the average shooting percentage in a league was 50% and we decided to rate all teams against the hypothetical opponent X, who shoots 50% on offense and allows 50% on defense.

Now let's say that we've calculated the following:
* On offense, Team A shoots 52.1% vs. the average opponent X.
* On defense, Team B allows the average opponent X to shoot 54.2%.

How would Team A be projected to shoot on offense w/ Team B on defense?

One could go this route and come up with a value: .50 + (.521 - .50) + (.542 - .50) = .563 = 56.3%

That would be a reasonable estimate. However, now let's try this example w/ the average shooting percentage still at 50%:
* On offense, Team A (a great offense) shoots 74% vs. the average opponent X.
* On defense, Team B (a very poor defense) allows the average opponent X to shoot 76%.

If we use the same equation, you get the following percentage for Team A on offense: .50 + (.74 - .50) + (.76 - .50) = 1.00 = 100%

I find that to be unlikely and, therefore, the method to be unreliable. Is there a statistical method to accurately forecast the offensive shooting percentage of Team A in these situations?

#### obh

##### Member
Hi Haz,

Try to understand the basic ...
Is "shooting percentage" the percentage of shooting go into the basket?
How can you shoot in "defense" ?

#### haz33

##### New Member
Hi Haz,

Try to understand the basic ...
Is "shooting percentage" the percentage of shooting go into the basket?
How can you shoot in "defense" ?

As for the second question, it's the percentage of your opponent's shots that go in when you are playing defensive against them.

Last edited:

#### haz33

##### New Member
I'm a bit fuzzy on that. How would one use logistic regression in this particular situation?

#### obh

##### Member
In logistic regression, the dependent variable (Y) is a categorical variable, for example, succeed shot or missed a shot.
And the regression will calculate the probability for evet value od Y.

But I'm not sure if you will have enough data to create the regression especially when the dependent variable (x) is also a categorical variable. (opponent group)