A multinomial model requires multiple responses. So there can be no
`multinomial family', in glm let alone in gee or yags. All assume a
univariate response variable.
On Mon, 25 Nov 2002, Paul, David A wrote:
> In the documentation for the "family" function which
> generates an object of type "family", it is stated that
>
> "...'quasi' can be used to combine any available link and
> variance function... The function 'power' can also be used
> to generate a powerlink function object for use with 'quasi';
> 'power' takes an argument 'lambda'. Users can construct their
> own families, as long as they have compatible components
> having the same names as those, for example, of binomial.
> The easiest way is to use 'quasi' with home-made link and
> variance objects; otherwise 'make.family' can be used, or
> else direct construction of the family object. When passed
> as an argument to 'glm' or 'gam' with the default link, the empty
> parentheses () can be omitted. There is a print method for
> the class 'family'."
>
> As I read it, this indicates that the multinomial family
> could be constructed from scratch. Is this true? If so,
> has anyone implemented GEE or YAGS with a multinomial family?
--
Brian D. Ripley, ripley@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 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
|