Can the sampleSelection package estimate selection models,
where the second stage is binary?
The latest versions of the sampleSelection package (>= 0.7)
can estimate models,
where the second stage is binary.
Thilo Klein informed us
that this model can be also estimated by the vglm function
and the constraints option
in the VGAM
package:
Selection: C = 1 if Xa + U > 0; else C = 0
Outcome: Y = 1 if Zb + V > 0; else Y = 0
const <- list()
const$"(Intercept)" <- diag(3)
const$x <- as.matrix(c(1,0,0))
const$z <- as.matrix(c(0,1,0))
const$C <- as.matrix(c(0,1,0))
fit <- vglm(cbind(C, y) ~ x + z + C,
binom2.rho(zero=3),control=vglm.control(maxit=maxit),constraints=const)
constraints(fit)
Furthermore, he made us aware of
that the SemiParBIVProbit function in the
SemiParBIVProbit
package also can estimate such models.
