model market share with proc logistic?

[magalineven]

We need to model the market share of a certain brand in a certain store at a given time. According to what we have learned, logistic models are often used when the dependent variable is marketshare as the natural maximum is 100%. However, when we use proc logistic in sas, we get impossible outputs. The reason for this might be the fact that in order to use proc logistic, you need an event and a non-event.

Do you have any idea which sas procedure to use instead? Thanks in advance!

[Reeza]

However, when we use proc logistic in sas, we get impossible outputs. The reason for this might be the fact that in order to use proc logistic, you need an event and a non-event.

Can you elaborate further here? What do you mean by impossible outputs?

Logistic regression takes two forms, where the Y is either 0/1 or a # of success/failures. 

<label:> MODEL variable<(variable_options)> = <effects> </ options>;

<label:> MODEL events/trials = <effects> </ options>;

If you used the first method that wouldn't fit your data but the second may. You would need to provide more information on your data and design for more informative suggestions.

---

@magalineven wrote:

We need to model the market share of a certain brand in a certain store at a given time. According to what we have learned, logistic models are often used when the dependent variable is marketshare as the natural maximum is 100%. However, when we use proc logistic in sas, we get impossible outputs. The reason for this might be the fact that in order to use proc logistic, you need an event and a non-event.

Do you have any idea which sas procedure to use instead? Thanks in advance!

---

[magalineven]

I guess we cannot use the proc logistic than... we should obtain a marketshare instead of a binary outcome. 

As for the impossible outcomes: we had some very large odds ratio estimates (around 400).

The goal is to become the logistic model of form ln(marketshare/(1-marketshare))=B0 + B1X1 + B2X2 + ... 

How can we obtain this? We changed the marketshare inputs to ln(marketshare/(1-marketshare)) inputs and used proc reg, could that be correct? 

[Ksharp]

Try Possion Regression .
Possion and Logistic has very close relationship.

Logistic model : log(p/1-p) =
when p~0 log(p/1-p) ~ p
-> log(p) -> log(n/total) -> log(n)= log(total) + .......... (it is just a Possion model)


http://support.sas.com/kb/24/188.html