I find doing these graphs `by hand' particularly useful when polynomials are combined with interactionterms. This method closely conforms to the way I make sense about those models myself: write the model down, and work through various contrasts. Of course it means more typing, but I find that I make less mistakes that way because this link between the syntax and the way I think about these models. Ultimately, I think it is just what you are used to doing (and how you were taught to do this stuff, I had a tough master who deliberately decided not to use Stata because it made this stuff too easy...)
Alternatively, since Tinna intended to use this after a 2sls model, I could imagine a polynomial occurring in an indirect effect. This too would be a case where I would prefer to first write it out, and the results would closely conform to the syntax in the do-file.
It is true that making graphs with complex models requires careful though, but I find it is very difficult to really explain the results without them. Interactions and indirect effects with a polynomial in it, are cases in point.
On 10/18/05, Nick Cox <[email protected]> wrote:
> As a model gets more complex, it is less likely that it can
> be presented by a graph. y = polynomial(x) is one exception, but
> in practice even cubics over a whole range are unlikely to be
> useful, in my experience.
>
> But, more to the point, a model gets more complex, you have _more_
> to type as you spell out each term as _b[varname] *
> varname. -predict- does this all for you. And this is error-prone.
>
> Other than the occasional pedagogic advantage of underlining
> what is being done, i.e. plugging estimates into a model
> formula, I can't understand your preference here.
>
*
* 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/