Dear all and Mikis I have the opportunity to compare fits with the 'classical' glm and gamlss and no smoother of any kind just the same model formula (both with the binomial family). I get exactly the same coefficients but very different residuals, gamlss giving residuals which are extremely close to 'normal' and glm very far... How can this be ? Thanks in advance for any indication or threat... Patrick
On Tue, 25 Mar 2008, pgiraudo at univ-fcomte.fr wrote:> Dear all and Mikis > > I have the opportunity to compare fits with the 'classical' glm and > gamlss and no smoother of any kind just the same model formula (both > with the binomial family). I get exactly the same coefficients but > very different residuals, gamlss giving residuals which are extremely > close to 'normal' and glm very far...What type of residuals are these? See ?residuals.glm and its first reference, and ?residuals.gamlss. The defaults for glm and gamlss are not comparable.> How can this be ?Comparing 'apples and oranges', perhaps?> > Thanks in advance for any indication or threat... > > Patrick > > ______________________________________________ > 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.PLEASE do as we ask. -- Brian D. Ripley, ripley at stats.ox.ac.uk Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/ University of Oxford, Tel: +44 1865 272861 (self) 1 South Parks Road, +44 1865 272866 (PA) Oxford OX1 3TG, UK Fax: +44 1865 272595
<p>Mikis' answer below... As guessed by Pr. Ripley, I was comparing apples and oranges indeed, and would have been better inspired looking at methods !</p><p>i Patrick <br /> <br /><a href="mailto:pgiraudo at univ-fcomte.fr" class="moz-txt-link-abbreviated">pgiraudo at univ-fcomte.fr</a> wrote: <br /><blockquote type="cite">Dear all and Mikis <br /> <br />I have the opportunity to compare fits with the 'classical' glm and gamlss and no smoother of any kind just the same model formula (both with the binomial family). I get exactly the same coefficients but very different residuals, gamlss giving residuals which are extremely close to 'normal' and glm very far... <br /> </blockquote>GAMLSS is? calculating the (randomized) quantile residuals which if the distribution is correctly specified should be normal. <br />The nice think about the quantile residuals is that they can be applied to any distribution. <br />An additional complication arises for discrete distributions where also a randomization is taken place because of the discrete nature of the data. <br />See the rqres.plot() function for more details. <br />Another way of looking at the gamlss residuals is that they are the fitted z-scores for the individual observations. <br /> <br />The glm() function is using either deviance or Pearson residuals<br></br> <br /> <br />Mikis <blockquote type="cite">How can this be ? <br /> <br />Thanks in advance for any indication or threat... <br /> <br />Patrick <br /> <br /> <br /> <br /> <br /> <br />---------------------------------------------------------------- <br />This message was sent using IMP, the Internet Messaging Program. <br /> <br /> <br /> </blockquote> <br /> <br />Companies Act 2006 : <a href="http://www.londonmet.ac.uk/companyinfo" class="moz-txt-link-freetext">http://www.londonmet.ac.uk/companyinfo</a> <br /> <br /> <br /> </p>