[R] generalised linear models

[Prof Brian Ripley]

On Mon, 7 Mar 2005, Michael Gray wrote:

> To whom this may concern,
>
> I would be very grateful if someone could give me some advice on where I 
> am going wrong with a logistic regression I am trying to run. I am 
> trying to run a logistic regression on an aggregated data set and have 
> input the command:
>
> logistic.mod<-glm(x~Frequency+Location+Sex+Age.Group,family=binomial(link="logit"),data=earsag1.dat)
>
> where x is the count of my response and frequency, location, sex and 
> age.group are other variables. However, R gives an error because my 
> values of x are not between 0 and 1. Therefore to compensate for this I 
> denoted y= x/n where n is the number of people in each group and ran the 
> regression
>
> logistic.mod<-glm(y~Frequency+Location+Sex+Age.Group,family=binomial(link="logit"),data=earsag1.dat)
>
> At this point R told me that my y values were not integer values and 
> hence there was an error. I do not know how, therefore, to set up the 
> logistic regression properly, so I would be very grateful if someone 
> could give me some pointers. If I haven't explained anything well 
> enough, let me know. Thanks very much for your help,

You have forgotten to set the weights=n.  The alternative is to have
response cbind(x, n-x).

See MASS4, p.190.

-- 
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