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 on April 23, and its replacement, is already up and running.

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

Re: st: Simple slope analysis for non-linear models

From   Maarten Buis <>
Subject   Re: st: Simple slope analysis for non-linear models
Date   Mon, 25 Mar 2013 10:30:22 +0100

On Sun, Mar 24, 2013 at 11:40 AM, Ebru Ozturk wrote:
>I followed your suggestion and it seems that I want this derivative E( y |X, y ≥ 0).


> In my model, I have one interaction term (dummy X continuous-ranges from 0 to 10) and other Xs.

This is sentence full of contradicotry statements: Do you have one or
many interaction terms? Is your key variable a single indicator
(dummy) variable, 10 indicator variables, or continuous?

>I would like to plot the values of true interaction effect and implied z-statistic value at each observation. After doing that I would like to show the value and significance of X’s marginal effect at selected values of the moderator Z (low, mean and high).  I use Stata 10.

There is no such thing as a "true interaction effect". I presume you
are looking for a cross partial derivative or discrete difference
(depending on whether your variables are continous or categorical). As
far as I know there is no program that does this kind of computation
for -tobit-, so you'll need the general purpose commands -predictnl-
and -adjust- for that in Stata 10. So, you'll need to look up the
appropriate formulas and probably do the derivatives yourself. For
such computations I often combine doing the derivations by hand and
using <>. After those computations you can
feed the results to -predictnl- or -adjust-.

I realize you would have liked the answer to be in the form of a
command rather than some general tips on how to write a program that
does what you want to do. But if no one wrote the program before, then
that is the only answer possible. I could have written the program for
you, but that is too big time investment on my part. I only do things
like that if I am also interested in that problem. In this case I
consider this way of thinking about interaction terms a dead end, so I
am not going to invest time in it.

> But the problem is in journals I use they never mention which Tobit intepretation they implement so that's why I am struggling.

So they report differences/changes in predicted outcome without
defining what the outcome is? That seems a bit problematic to me.

Hope this helps,

Maarten L. Buis
Reichpietschufer 50
10785 Berlin

*   For searches and help try:

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