I don't think it makes much difference what
your models are. You have a choice with missing
data between living with them and using some
method to impute non-missing values, after
which your model may show improved performance,
but you should worry how genuine that improvement
might be. Thus neither approach is problem-free,
but there is an enormous literature on the matter
and some Stata implementations of imputation methods,
among which -impute-, -hotdeck- and -mice- spring to mind.
-findit- will give further pointers.
Nick
[email protected]
Quang Nguyen
> I am applying the Heckman procedure to estimate the role of "Share
> Contract" in longline fishery. In the first step, I run a Probit model
> where the dependent variable is the binary variable "Share Contract or
> something else", the independent variables include economic factors
> such as variable cost, risk sharing attitude... In the second step, I
> run OLS where dependent variable is revenue and several independent
> variables including the binary variable "Share Contract...". My data
> includes about 95 observations.
>
> After running the Heckman procedures, no variable seems statisitically
> significant. Some coeficinces have unxpected sign. I think one of the
> reason is probably due to many missing observations. Or it could be
> some other reasons.
>
> I would highly appreciate if anyone could help me to fin dthe way
> dealing with this matter, especially with respect to missing
> observations. Also, any idea on how to improve the model would be very
> much appreciated.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
