GLM in R (and specification of non-explanatory variables)

[Mr_Pag]

Hi all,

I am trying to fit a model via the glm function in R. I am wondering if R will differentiate between categorical and continuous (covariate) variables, or will I need to specify either in the model. From the in-built help it appears neither and I had a quick rake through the R archives to no avail.

From what I gather, in SAS you typically specify the categoricals, whereas in Minitab it is the covariates. e.g.




## MTB Code

GLM "Y" = X1 X2;
Covariates 'X3';
Brief 3.

## SAS Code

proc glm;
class X1 X2;
model Y= X1 X2 X3/p solution;

# Note: May not provide equivalent functionality



How are these stipulated in R, is glm the right tool for the job or should I be investigating ANCOVA functions or another library function? Many thanks in advance!

Kind regards,

Pag

[akniss]

Hi all,

I am trying to fit a model via the glm function in R. I am wondering if R will differentiate between categorical and continuous (covariate) variables, or will I need to specify either in the model. From the in-built help it appears neither and I had a quick rake through the R archives to no avail.

From what I gather, in SAS you typically specify the categoricals, whereas in Minitab it is the covariates. e.g.





## SAS Code

proc glm;
class X1 X2;
model Y= X1 X2 X3/p solution;

# Note: May not provide equivalent functionality



How are these stipulated in R, is glm the right tool for the job or should I be investigating ANCOVA functions or another library function? Many thanks in advance!

Kind regards,

Pag

It depends on how your variables are coded. If you have a variable that is text (say, "M","F") then it will automatically assume they are categorical. If your variable is coded as a number, (say, "1", "2") then R will assume they are continuous. To tell R explicitly that you want those numeric values to be used as categorical variables, you may use the as.factor() function.

Assuming variables a and b are numeric, and x and z are categorical:

as.factor(x)->x
as.factor(z)->z
glm(y~x+z+a+b)->fm1
summary(fm1)

should do the trick.

[Mr_Pag]

Nice one, thanks very much for your help!! I've give this a bash once I've updated to Karmic (I imagine a few people are in this boat today) - :D

All the best,

Alan