Instrumental variable in logistics regression in R (ivreg in AER)

Question

Does ivreg in AER support a logit regression with instrumental variables?

For example:

IV = ivreg (Mort ~ Age + Sex + APACHE + PART_SendImmed + ICU_AdmImmed + ICU_LOS | Age + Sex + APACHE + PART_SendImmed + NurseOCC_Adm + NurseOCC_Disch,
    data = test)

Where, Mort is a binary variable ICU_AdmImmed and ICU_LOS are endogenous variables, and NurseOCC_Adm and NurseOCC_Disch are two instrumental variables.

Answer 1

I would suggest to use a probit rather than a logit, which will both improve the interpretability as well as allow you to use ivprob.

data("mtcars")

library(ivprobit)

ivprob(I(mtcars$vs==0), mtcars$cyl, I(mtcars$disp>150), mtcars$hp)

$coefficients
          [,1]
[1,] -62.48480
[2,]  17.09827
[3,] -74.66346

$se
[1]  93844628  28724155 134217728

$tval
              [,1]
[1,] -6.658325e-07
[2,]  5.952576e-07
[3,] -5.562862e-07

$pval
numeric(0)

$df
numeric(0)

$names
[1] "(Intercept)" "X1"          "yhat"

I don't know of a canned package to do it in R using a logit, in fact I'm fairly sure there isn't one as this was the conclusion of an econjobrumors.com thread on the topic, however if there's some reason you'd prefer a logit to a probit, then you could always carry out the process manually.