[R] Hickman models with two binary dependent variables in R

Arne Henningsen arne.henningsen at gmail.com
Mon Aug 29 14:43:34 CEST 2016

Dear Faradj

On 27 August 2016 at 16:11, Faradj Koliev <faradj.g at gmail.com> wrote:
> It actually worked with heckit() command as well, do I need to use
> selection()?

I suggest that you use the selection() command/function only. heckit()
is just a wrapper to selection() with the only difference that
heckit() estimates the model with the 2-step method by default, while
selection() estimates the model by the maximum likelihood method by
default (unless specified otherwise with argument "method").

I see no good reason for estimating the sample-selection model by the
2-step method instead of the maximum likelihood method.

> Also, I would be really grateful if you can suggest a package that would
> allow for estimation of heckman models with two ordered variables (0-1-2).
> Can sampleSelection handle this?

Do you mean that the dependent variable of the selection equation is
an ordered variable with three levels (0-1-2), that the dependent
variable of the outcome equation is an ordered variable with three
levels (0-1-2), or that both of the dependent variables are ordered
variable with three levels (0-1-2) each?

In any case: no, this is not implemented in the sampleSelection package. Sorry!

Anyway, I guess that it is not too complicated to derive the
likelihood function and implement the estimation yourself, e.g., using
the maxLik package. If you do this, I would be happy to help you to
implement this feature in the sampleSelection package.

Best wishes,

Arne Henningsen

More information about the R-help mailing list