Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


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

st: Re: st: Re: st: Re: st: Re: st: Re: st: Re: st: RE: Truncated sample or Heckman selection‏


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: Re: st: Re: st: Re: st: Re: st: Re: st: Re: st: RE: Truncated sample or Heckman selection‏
Date   Fri, 5 Oct 2012 17:51:32 +0100

Much of the point of this thread is that those answering you
_disagree_ on certain key things and are at least in part_unclear_ on
your situation.

I already said "If you are expecting that Statalist can act as a
collective oracle that can tell you the _correct_ thing to do in your
project, that is I suggest more than the list can deliver."

So, you still think you can get definitive advice out of us?

0. What do your advisors/supervisors/colleagues say who know more
about your dataset and objectives?

1. I think that's overstating it. It just sounds like a natural model
but no-one has access to your data or awareness of your full
objectives to be able to say how well it will work.

2. That's overstating it too. Whether you _need_ to do Heckman is your
decision.

3. I don't know what this means. See #0.

4. You need to try it. Answering this question would mean going back
to Stata 10 and looking at the documentation and I can't answer off
the top of my head. (As you didn't say, contrary to advice, that you
are using an old version of Stata, no-one will have been thinking
about that.)

5. I don't see why not in principle, but you may need to write a
program in practice. All depends on how the panel aspect is handled.

Nick

On Fri, Oct 5, 2012 at 5:38 PM, Ebru Ozturk <ebru_0512@hotmail.com> wrote:
>
> Thank you very much for your help. I just would like to make sure that I understand you correctly. Also I am here to learn so please forgive me for asking many questions.
>
> 1. I restrict sample to innovating firms and the dependent variables that I am interested in are percentages therefore, you agree on that I should use fractional logit model.
>
> 2. I do not need to do Heckman sample selection even though I restrict my sample. I also want to mention that yes I do not have some Xs but I still do have other Xs for non-innovators which I use them as control variables.
>
> 3. Do you think using a different model puts me trouble?
>
> 4. I use Stata 10 so can I use fractional logit model?
>
> 5. What about if I have a panel data? Is fractional logit model still applicable?
>
> Ebru
>
> ----------------------------------------
>> Date: Fri, 5 Oct 2012 16:21:40 +0100
>> Subject: st: Re: st: Re: st: Re: st: Re: st: Re: st: RE: Truncated sample or Heckman selection‏
>> From: njcoxstata@gmail.com
>> To: statalist@hsphsun2.harvard.edu
>>
>> Isn't this the same question?
>>
>> It seems to me that you have two issues, which are independent of each other.
>>
>> 1. Your response is bounded [0,1] or [0,100], dependent on units.
>>
>> Joerg, Maarten and I all seem to agree that this is not censoring in
>> any strong sense, as values outside the bounds are impossible in
>> principle. Thus we don't think that tobit makes sense here.
>>
>> Daniel's argument appears to be that there is a important sense in
>> which this can be considered as censoring. Thus he seems more open to
>> the idea that tobit makes sense.
>>
>> This is not an election and what is right is not determined by
>> counting votes. As I don't understand Daniel's argument I certainly
>> can't refute it.
>>
>> Personally I pay little attention to what's popular in the literature
>> if I think it's wrong. It's good news for me: I have a chance to try
>> to improve practice.
>>
>> 2. There is a question of what determines which firms do and which do
>> not innovate. This is interesting and important and you should model
>> it _if you have data to do so_. In your earliest posts in this thread,
>> you seemed (to me at least) to imply that you have no other data on
>> non-innovating firms with which to model selection, but you should
>> know.
>>
>> This is just my summary. The others who have contributed to the thread
>> may want to correct or complement it. Other people of course may have
>> other views.
>>
>> If you are expecting that Statalist can act as a collective oracle
>> that can tell you the _correct_ thing to do in your project, that is I
>> suggest more than the list can deliver.
>>
>> Nick
>>
>> On Fri, Oct 5, 2012 at 3:29 PM, Ebru Ozturk <ebru_0512@hotmail.com> wrote:
>> > My question was whether I should use Heckman sample selection as I restricted my sample to innovating firms only.
>> >
>> > After discussing with you, I also confused as I said before many papers used tobit regression and did Heckman correction and never seen fractional logit model in these papers such as Grimpe & Kaiser, (2010); Cassiman & Veugelers, (2006).
>> >
>> > These papers are saying that, the data is left censored as it includes many zeros. Also when I search for fractional logit model it says it is bounded between 0 and 1 but my dependent variable includes for instance, 0, 11, 13, 15, 25 values and so on.
>> >
>> > Maybe I am not very good at statistics but when I read papers they use tobit regression. Whey then?
>> >
>> > Ebru
>> >
>> > ----------------------------------------
>> >> Date: Fri, 5 Oct 2012 15:02:22 +0200
>> >> Subject: st: Re: st: Re: st: Re: st: Re: st: RE: Truncated sample or Heckman selection‏
>> >> From: maartenlbuis@gmail.com
>> >> To: statalist@hsphsun2.harvard.edu
>> >>
>> >> On Fri, Oct 5, 2012 at 1:53 PM, Ebru Ozturk <ebru_0512@hotmail.com> wrote:
>> >> > I think there is a misunderstanding here. Yes, I have valid and meaningful values on my explanatory variables but when I restrict the sample to innovating firms. By restricting the sample I just exclude the firms that no activity toward innovation at all which I am not interested in them. So, in the restricted sample I can also observe Xs for 0s as my dependent variables are radical and incremental innovation.
>> >> >
>> >> > I hope it is clear now.
>> >>
>> >> Now I really don't understand your problem: Is it that you have a
>> >> dependent variable that is a proportion and some of these proportions
>> >> are 0? Than just use a fractional logit model (just -glm- with the
>> >> -link(logit)- and the -vce(robust)- options). For such problems
>> >> neither -tobit- nor -heckman- are appropriate nor is there any
>> >> truncation going on.
>> >>
>>
>> *
>> * For searches and help try:
>> * http://www.stata.com/help.cgi?search
>> * http://www.stata.com/support/faqs/resources/statalist-faq/
>> * http://www.ats.ucla.edu/stat/stata/
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index