[R] glm for nominal X ordinal

[David Winsemius]

Gang Chen <gangchen at mail.nih.gov> wrote in
news:4B1D9D53-7F9A-4337-912A-B7D29601DC25 at mail.nih.gov: 

> Suppose I have a two-way table of nominal category (party  
> affiliation) X ordinal category (political ideology):
> 
> party affiliation X (3 levels) - democratic, independent, and
> republic 
> political ideology Y (3 levels) - liberal, moderate, and
> conservative 
> 
> The dependent variable is the frequency (or count) for all the two- 
> way cells sampled from the voters. I want to test whether there is  
> any party affiliation effect, and, if there is, the pair-wise  
> contrasts. I have never used glm (I assume this the program I should
>  use) before, so I am not so sure how I can code the two independent
>  variables reflecting the fact that one is nominal while the other
> is  ordinal and how to formulate the model.

?factor

Set up affiliation as a factor and ideology as an ordered factor. The 
levels argument in factor sets the sort order.

> theo<-c("cons", "mod", "cons", "cons", "lib", "mod")
> table(theo)
theo
cons  lib  mod 
   3    1    2 

> theo<-ordered(theo, levels=c("lib", "mod", "cons"))
> table(theo)
theo
 lib  mod cons 
   1    2    3 

 The formula  in glm would be something similar to counts ~ theo + affil

Much more detail and worked examples regarding modeling count data would 
be found in:
Thompson, LA;  (2004) R (and S-PLUS)Manual to Accompany Agresti’s (2002) 
Catagorical Data Analysis;
https://home.comcast.net/~lthompson221/Splusdiscrete2.pdf