EViews.com

Hi,

I am trying to get the marginal effect of an x-variable (a dummy) in a probit regression. I.e. if the x-variable I'm interested in is x1, I need to find

G(B0 + x1 + B2.x2 +...Bk.xk) - G(B0 + B2.x2 +...Bk.xk).


I understand that usually, the other x-variables (other than x1, that is) would be held constant in both G() expressions by substituting a significant value like the mean of each x-variable for that x-variable (i.e. B2.x2 would become B2.(mean of x2), B3.x3 becomes B3.(mean of x3) etc.). However, several of my x-variables are also dummies - i.e. I'm not sure if using their means would be meaningful.

I have tried getting the marginal effect for x1 by setting all other dummy x-variables to 0, and I have also tried setting all other x-variables to 1, but both times, I get a figure for the marginal effect of x1 that is negative (e.g. -1.54), which is troubling because the coefficient of x1 is positive - I would think that the marginal effect should have the same sign as the coefficient.

So right now I'm not sure what to set the non-x1 dummy variables to in order to hold them constant.

I'm rather new to econometrics, any help at all would be much appreciated.

Thanks in advance! :)

Note that there isn't a definitive answer as to what to hold your other values. People often use means, but I'm not a particularly big fan of that...Evaluating at data points and then computing descriptive statistics over the set of values isn't bad, but the results are obviously quite dependent on the sample.

The EViews User's Guide chapter on limited dependent variables has some discussion of the mechanics of doing this. If you are going to evaluate at data points, then you can use the substitution methods that I outlined in our other post. If you are going to evaluate at specific points, I'd recommend you use the model object to do the simulation. Note that the Version 7 manual has an updated example which is a bit better than the earlier one...