Bookmark and Share

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

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Selection with an endogenous ordinal variable

From   Austin Nichols <>
Subject   Re: st: Selection with an endogenous ordinal variable
Date   Tue, 8 Feb 2011 22:37:41 -0500

Filipe Silva <>:
I don't see any theory motivating a selection model, and correlated
residuals in a misspecified -heckman- are not evidence in favor of a
selection model.
Firms with mean R+D conditional on X vars close to zero will often
have realized R+D of zero or so small an amount that it will be
measured as zero in your data.
Sounds more like -glm- with a log link, or better, given an instrument
for FC, a -gmm- model:

On Tue, Feb 8, 2011 at 9:43 PM, Filipe Silva
<> wrote:
> Thank you for the suggestion on the -cmp-
> I apologise for not being as clear as I should have been
> I intend to measure the impact of firms' financial constraints (FC)
> upon R&D investment:
> y: R&D investent, whith 74% of zeroes,   €{0}U]0, +00[
> main x: FC, which is ordinal,  €{0;1;2;3}
> Since there are so many zeroes, I hypothesized that there is first the
> decision to invest or not (RD) followed by the decision on the amounts
> invested if RD==1. Accordingly, if these decisions are assumed to be
> independent (conditional on observable vars.),  I'd estimate a 2part
> "hurdle" with no need to jointly specify errors.
> However, I tested a -heckman- not accounting for endogeneity and it
> appears that there are unobservables affecting both errors.
> As a result I thought of a selection model, but please correct me if
> wrong (I have just recently started to deal with such kind of data).
> Additionally, there are reasons to believe that FC is endogenous in
> the RD_I decision, which further complicates.
> Thank you very much,
> Filipe
> 2011/2/8 Austin Nichols <>:
>> Filipe Silva <>:
>> I don't understand your problem statement--can you clarify what the outcome and
>> endogenous ordinal var are, and what form of selection is hypothesized?
>> You may also want to look at -cmp- on SSC for MLE options.
>> On Tue, Feb 8, 2011 at 5:57 AM, Filipe Silva
>> <> wrote:
>>> Dear all,
>>> I am currently trying to estimate a model of selection, where the main
>>> explanatory variable in this model is endogenous (and ordinal!). All
>>> of them observed.
>>> This should be easily done with -heckman-, if not for the endogenous regressor.
>>> I wonder if there is any package that can handle (MLE) estimation of such model?
>>> Alternatively I have considered a two-step procedure (probit, obtain
>>> mills, use in ivreg2 to account for endogeneity- as far as I've
>>> understood this is the procedure described in Wooldridge's textbook,
>>> 2002 pp. 567-570 as an extension to "heckit", even though I'm not
>>> totally sure it applies to an ordinal endogenous var.) but the the
>>> Variance correction in the second step is rather hard to compute. I
>>> have considered using the bootstrap for such correction.
>>> Could anyone please advise me if this reasoning is correct or if there
>>> is an alternative way to do it?

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index