[R] how to verify gauss-markov hypothesis for linear model validity?

Greg Snow Greg.Snow at imail.org
Wed Jun 17 19:14:48 CEST 2009


I don't think that your questions are stupid, but they probably are the wrong one(s).

There are 2 questions (or sets of questions) when thinking about your data for doing statistical inference.  

The first question is "does this assumption hold exactly?" e.g. "are the residuals exactly normal?".

The second question is "is the assumption close enough to holding that I will get reasonable results when I do my inference?" e.g. "Is the data normal enough?" or "Is it close enough to normal?".

The first set of questions is easier to answer (the answer is "No"), but generally the answer is uninteresting and sometimes misleading.

The second set of questions is more important/useful to answer, but requires more thought/work by the researcher.

Hope this helps,  

-- 
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 jose romero
> Sent: Tuesday, June 16, 2009 7:27 PM
> To: r-help at r-project.org
> Subject: [R] how to verify gauss-markov hypothesis for linear model
> validity?
> 
> Hello list:
> 
> (This is probably a stupid question).  Is there a "quick and easy" way
> to confirm the gauss-markov conditions of a linear multiple regression
> model?  That the mean of the residuals is 0 can easily be tested for.
> The normality of the residuals as well (shapiro-wilk?).  But what about
> homoscedasticity? And independence of residuals with respect to the
> model variables?
> 
> Thanks in advance
> 
> 
> 	[[alternative HTML version deleted]]




More information about the R-help mailing list