Bookmark and Share

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


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

Re: st: lincom vs nlcom


From   "Airey, David C" <[email protected]>
To   "[email protected]" <[email protected]>
Subject   Re: st: lincom vs nlcom
Date   Thu, 3 Mar 2011 08:43:12 -0600

.

I liked that answer. It was helpful even though I didn't ask the question.

Thanks, Maarten.

--- On Thu, 3/3/11, Vincenzo Carrieri wrote:
> I am trying to estimate an ordered probit model like this:
> 
> W=age+agesquared+height+height*age+height*agesquared
> 
> I am interested in the significance and the sign of the
> interaction between age and height. I would do:
> lincom  height+height*age+height*agesquared
> 
> I have two questions on that:
> 1. is lincom the right command to do this? Or I need nlcom?
> (I think that even if the model is non linear, the combination I
> want to test it is linear, so lincom should work well)

By adding a square terms it has become logically impossible to 
answer the question what "the" interaction effect is, as there is 
no longer one interaction effect but many. 

On top of that comes the fact you are using an ordered probit, so
even if you did not add the square term, you would still have the
same problem.(*)

> 2. The output of lincom (and eventually of nlcom) is also a
> marginal effect? And, if not, how to get a marginal effect where
> there are interaction effects?

No it is not a marginal effect.

One way to resolve this problem (for as far as it is possible) is
really pin down the hypothesis that you want to test. "Testing an
interaction effect" can mean too many things in this type of 
models. What is your dependent variable? A latent score or a 
probabilty of choosing a category, or the odds of moving one 
category up? Do you want to see how the effect of height changes 
with age or how the effect of age changes with height? Do you 
define the effect in absolute terms (difference) or in relative 
terms (ratios). Once you have figured that out draw a lot of 
graphs.(**) That way you get a feel for how variable the effect 
is and thus whether it would make sense to try to summarize it in 
one number. Once you have finished that stage you will have a
sufficient idea what you want to test exactly and you will know
how to implement that in -lincom- or -nlcom-.

Hope this helps,
Maarten



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


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