Logistic regression with dummy variables and regular variables

[ak1190]

Hello.

My logit regression for a sample of 182 has 4 significant explanatory variables. One of these variables is a dummy variable and the rest are regular ratio type data point variables.

Given this, how do I interpret my grand mean and intercept term? And how do I interpret the coefficient of the dummy variable? If I need to calculate my mean WTP, do I still use the formula:

-(b0+b2x2bar+b3x3bar+b4x4bar)/b1

where b0 is intercept term, b2 is the coefficient for the variable x2, x2bar is the sample mean of x2 and b1 is the coefficient of the bid value variable; without any caution against the fact that my regression includes a dummy regressor?

Thank you.

 

[Dason]

It depends on how the model was parameterized. What software did you use? And you're referring to logistic regression right - your outcome was binary?

 

[ak1190]

I used Stata. Yes, my outcome (i.e. actual values) would be binary. But the fitted values are arrived at with the help of the CDF, and are not binary.

 

[Dason]

I used Stata.

I don't work with Stata so I don't know how it does its default parameterization - somebody else will have to help there..

But the fitted values are arrived at with the help of the CDF, and are not binary.

Of course - it wouldn't really be logistic regression otherwise. I was just asking because you said "logit regression" which could be a non-linear regression using the logistic function as the response and just wanted to make sure we were on the same page.

 

[ak1190]

Ah ok. Thanks for looking anyway.

 

[caleb]

i dont have any idea

 

[noetsi]

The intercept is defined as it always is. The change in Y when all levels of X are zero (for the slope). In SAS at least there are two types of coding for dummy variables. In one case, reference I think, the level of the dummy is being compared to the level of the categorical variable you did not create a dummy for. The slope then is the mean difference between these levels. In effect coding, which is rarely used I suspect, the slope of the dummy is being compared to the mean of the means of the other levels (not the grand mean).

I would guess Strata does it the same way, you need to look at the documentation.