Message-ID: <4F96B127.7060205@stats.ox.ac.uk>
Date: 2012-04-24T13:56:55Z
From: Brian Ripley
Subject: nobs.glm
In-Reply-To: <CANQBBMsUcKHZeLpN5a1mCnzHhTMNNeSBwiPTUdydc=2jp8netw@mail.gmail.com>
On 24/04/2012 14:36, Wincent wrote:
> Hi all,
>
> The nobs method of (MASS:::polr class) takes into account of weight,
> but nobs method of glm does not. I wonder what is the rationale of
> such design behind nobs.glm. Thanks in advance. Best Regards.
>
>> library(MASS)
>> house.plr<- polr(Sat ~ Infl + Type + Cont, weights = Freq, data = housing)
>> house.logit<- glm(I(Sat=='High') ~ Infl + Type + Cont, binomial,weights = Freq, data = housing)
>> nobs(house.plr)
> [1] 1681
>> nobs(house.logit)
> [1] 72
>
Well, the interpretation of 'weights' for a GLM depends on the family.
They may be equivalent to duplicated observations for a binomial GLM,
but they are not for a Gaussian one. The nobs method for class "glm"
(there is no visible nobs.glm) follows the "lm" method.
--
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