On Raoul's original question, my answer is neither
and both. A graphical "test", meaning examination,
should complement a "statistical test", meaning
hypothesis or significance test, and both should
be useful.
Nick
[email protected]
Mark Schaffer
> It sounds like you are interested in a RESET test. This is sometimes,
> rather misleadingly, called an omitted variables test, but it's better
> thought of as a test for neglected nonlinearities in functional form.
>
> Stata has a canned RESET test called -ovtest-. -ivreset-
> (downloadable
> from ssc-archives) will also work after -regress- and gives you more
> options.
> > What is the best way to test for a departure from linearity in a
> > regression model? Graphically or statistically? I ask this
> because when I
> > use, for example graphical methods like: .lincheck or
> .cprplot (component
> > plus residual plot), or .acprplot(augmented component plus
> residual) the
> > variable that I test (age) looks very linear. However, when I do a
> > statistical test for it, it is not linear!
> > I first create a regression model with a dummy
> variable(i.age) and then
> > compare it to a regression model with the same
> > variable but then as an ordinal variable (age). See below
> for an example
> > (BP=blood pressure)
> >
> > xi: regress BP i.age
> > estimate store Model1
> > regress BP age
> > estimate store Model2
> > lrtest M2 M2
> >
> > My sample size is very large (approx 20,000 subjects),
> could this be why
> > the test gives a significant result while the plots look
> very linear?
*
* 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/
