[R] Comparing non nested models with correlation coefficients (inspired from Lorch and Myers )

[Greg Snow]

Well this starts veering towards fortune(194) and fortune(218).

Though I did one time receive a response from an editor and reviewers asking for fewer p-values in favor of more confidence intervals.  I was excited that someone was willing to move in that direction (unfortunately, that particular study was one for which the p-values were more meaningful than the confidence intervals).

I have successfully talked some researchers out of presenting p-values in favor of more meaningful presentations, though it is often harder than it should be (Had a stanza from "The Hunting of the Snark" posted on the wall for a while after one researcher finally gave in after the 3rd person told him the same thing).

I am somewhat curious what the editor thinks the null and alternative hypotheses are, what the real null and alternative hypotheses are, and what the interpretation of a failure to reject will be (especially if there is low power).

If you cannot come up with meaningful/compelling answers to the above questions, but the editor still insists on a p-value, then you can always use SnowsCorrectlySizedButOtherwiseUselessTestOfAnything (TeachingDemos package).  Though that could prove the exception to Dieter's rule.

-- 
Gregory (Greg) L. Snow Ph.D.
Statistical Data Center
Intermountain Healthcare
greg.snow at imail.org
801.408.8111


> -----Original Message-----
> From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-
> project.org] On Behalf Of Boris New
> Sent: Thursday, March 24, 2011 12:14 AM
> To: Thomas Lumley
> Cc: r-help at r-project.org; Brian S Cade; r-help-bounces at r-project.org
> Subject: Re: [R] Comparing non nested models with correlation
> coefficients (inspired from Lorch and Myers )
> 
> Thank you very much for your answers.
> 
> The problem is that the editor wants a formal test. So I guess that the
> Vuong test should be ok...
> 
> 2011/3/24 Thomas Lumley <tlumley at uw.edu>
> 
> > On Thu, Mar 24, 2011 at 8:26 AM, Brian S Cade <cadeb at usgs.gov> wrote:
> > > As a follow-up to Greg's suggested graphical presentation, it seems
> like
> > > the Vuong test is sometimes used to compare fits of non nested
> models.
> > >
> >
> > There is a nice practical example of this with code in R, by Cosma
> > Shalizi and coworkers, at
> > http://cscs.umich.edu/~crshalizi/weblog/491.html
> >
> > The application is the tendency of physicists and people who read
> > Wired to see power law distributions everywhere, and Vuong's test is
> > used to determine whether a power law fits better than a lognormal
> > (which it typically doesn't).
> >
> >     -thomas
> >
> > --
> > Thomas Lumley
> > Professor of Biostatistics
> > University of Auckland
> >
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-
> guide.html
> and provide commented, minimal, self-contained, reproducible code.