[R] Selecting Best Model in an anova.

[Greg Snow]

This really depends on what question you are trying to answer, and for some questions the "Best" model is not one of the 2 you show.

-- 
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 Hector Guilarte
> Sent: Thursday, March 25, 2010 1:36 AM
> To: r-help at r-project.org
> Subject: [R] Selecting Best Model in an anova.
> 
> Hello,
> 
> I have a simple theorical question about regresion...
> 
> Let's suppose I have this:
> 
> Model 1:
> Y = B0 + B1*X1 + B2*X2 + B3*X3
> and
> Model 2:
> Y = B0 + B2*X2 + B3*X3
> I.E.
> Model1 = lm(Y~X1+X2+X3)
> Model2 = lm(Y~X2+X3)
> 
> The Ajusted R-Square for Model1 is 0.9 and the Ajusted R-Square for
> Model2 is 0.99, among many other significant improvements.
> 
> And I want to do the anova test to choose the best one:
> 
> H0: B1 = 0
> H1: B1 != 0
> 
> Test = Anova(Model2,Model1)
> 
> How do I know what model wins? (I'm using a confidence level of 0.1)...
> 
> My guess is that:
> If p-value of summary(Test) is greater than 0.1 then I don't reject H0
> so Model2 is better and otherwise I reject H0 so Model1 is better?
> 
> My teacher once said: "If p-value is greater than 0,5 we choose the
> short model and otherwise we choose the long model", but she never said
> how the p-value and the significance level were related in this test...
> Actually she never talked about significance level...
> 
> In short: Should I consider the significance level or always use 0.05
> for this kind of test?
> 
> Thanks a lot!
> 
> Hector Guilarte
> Enviado desde mi dispositivo movil BlackBerry(r) de Digitel.
> ______________________________________________
> 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.