[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Is -predictnl , force- appropriate for out-of-sample prediction in fake data?

From   "Eva Poen" <>
To   Statalist <>
Subject   st: Is -predictnl , force- appropriate for out-of-sample prediction in fake data?
Date   Tue, 1 Jul 2008 13:14:27 +0100

Dear all,

this is, in some way, a follow-up question to the one I asked last
week. Austin Nichols explained the easiest way to graph a nonlinear
prediction after estimation, for set values of the covariates; this is
the link to his post:

What I want to do now is calculate the confidence intervals for these
predictions. To clarify, I created a series of fake data sets, with
very few observations each (10 or so). I use -predict- to obtain
predictions based on the estimates from the original regression (which
had more than 20000 observations), and plot them. Looking at the
manual entry for -predictnl- it appeared that this command is meant to
be used with the original data. If I try to use it in my little fake
data set, I receive the error message "expression is a function of
possibly stochastic quantities other than e(b)". The -force- option,
however, will do it anyway.

Looking at the methods and formulas section in the manual, it was not
quite clear to me at what point I (possibly) wander into dangerous
territory with this approach. Is it appropriate to use -predictnl- in
this way, i.e. to produce standard errors for a prediction with fake

Thanks a lot,
*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index