Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# st: Standard error of the marginal effect or predicted probability in a conditional logit

 From Glenn Hoetker To statalist@hsphsun2.harvard.edu Subject st: Standard error of the marginal effect or predicted probability in a conditional logit Date Thu, 19 Aug 2010 12:42:51 -0500

```Hi all.

Can anyone point me towards a way to calculate the standard errors of
either the marginal effects or the predicted probabilities of a
conditional logit model (I suppose in the new Stata terminology, I could
also mean an alternative specific conditional logit).

Since this is a non-linear function of the coefficients, the Delta
method seems like it might be appropriate.  That would require
calculating

dF(x'b)                            dF(x'b)
--------    Var(b_hat)     ----------
db                                      db

which equals (I think)

f(x_i b) x_i Var(b_hat) x_i  f(x_i b)     {the x in x_i should be x' , but that's too hard to read in ascii math)

But, I'm struggling to understand how I would calculate f(x'_i b) and which x's correspond to x_i, since there are x's for each of the =
possible choices in a choice set.

Any leads at all would be very, very welcome.

Thank you.

Glenn
Glenn Hoetker
Julian Simon Faculty Fellow in Business
Associate Professor (Law, Institute for Genomic Biology)
Director, Center for International Business Education and Research
University of Illinois
217-265-4081
ghoetker@illinois.edu