intEff {DAMisc}  R Documentation 
Norton and Ai (2003) and Norton, Wang and Ai (2004) discuss methods for calculating the appropriate marginal effects for interactions in binary logit/probit models. These functions are direct translations of the Norton, Wang and Ai (2004) Stata code.
intEff(obj, vars, data)
obj 
A binary logit or probit model estimated with 
vars 
A vector of the two variables involved in the interaction. 
data 
A data frame used in the call to 
A list is returned with two elements  byobs
and
atment
. The byobs
result gives the interaction effect
evaluated at each observation. The atmean
element has the marginal
effect evaluated at the mean. Each eleement contains an element int
which is a data frame with the following variable:
int_eff 
The correctly calucalted marginal effect. 
linear 
The incorrectly calculated marginal effect following the linear model analogy. 
phat 
Predicted Pr(Y=1X). 
se_int_eff 
Standard error of

zstat 
The interaction effect divided by its standard error 
The X
element of each returned result is the Xmatrix used to
generate the result.
Dave Armstrong
Norton, Edward C., Hua Wang and Chunrong Ai. 2004. Computing
Interaction Effects and Standard Errors in Logit and Probit Models. The
Stata Journal 4(2): 154167.
Ai, Chunrong and Edward C. Norton. 2003. Interaction Terms in Logit and Probit Models. Economics Letters 80(1): 123129.
Norton, Edward C., Hua Wang and Chunrong Ai. 2004. inteff: Computing Interaction Effects and Standard Errors in Logit and Probit Models, Stata Code.
data(france)
mod < glm(voteleft ~ age*lrself + retnat + male, data=france, family=binomial)
out < intEff(obj=mod, vars=c("age", "lrself"), data=france)
out < out$byobs$int
plot(out$phat, out$int_eff, xlab="Predicted Pr(Y=1X)",
ylab = "Interaction Effect")
ag < aggregate(out$linear, list(out$phat), mean)
lines(ag[,1], ag[,2], lty=2, col="red", lwd=2)
legend("topright", c("Correct Marginal Effect", "Linear Marginal Effect"),
pch=c(1, NA), lty=c(NA, 2), col=c("black", "red"), lwd=c(NA, 2), inset=.01)