topic Re: glm procedure in Statistical Procedures

glm procedure

marica14 — Wed, 24 Feb 2016 14:08:50 GMT

hello,

i am doing a glm procedure to determine the factors exposure on fetal growth. ( pesticides, season )

i have to adjust it with sex and diabetes (for example). ( i have make dummy variables for qualitatives variables and put it on class statement)

i don't know the right procedure :

is it

proc glm data = work;

class diabetes;

model weightnewborn = pesticides * sex*diabetes season*sex*diabetes;

run;

proc glm data = work;

class diabetes;

model weightnewborn = pesticides * season*sex*diabetes

run;

thank you for your response

Re: glm procedure

Rick_SAS — Wed, 24 Feb 2016 14:30:03 GMT

Both sex and diabetes are categorical, so both go on the CLASS statement:

CLASS Sex Diabetes;

If you want a model that uses only the main effects, use

MODEL weightnewborn = pesticides season sex diabetes;

If you believe that there are interactions between the explanatory variables, then use the "*" operator to specify the interaction effects. For example,

MODEL weightnewborn = pesticides season sex diabetes sex*diabetes;

is a model that assumes that the birth weights depend on whether the females also have diabetes (or diabetic mothers?).

Re: glm procedure

marica14 — Wed, 24 Feb 2016 14:32:38 GMT

thank you, so we mix exposition factors and confusion factors ?

Re: glm procedure

Rick_SAS — Wed, 24 Feb 2016 14:44:25 GMT

There is a nice section of the GLM documentation that discusses how to specify effects by using the SAS "stars and bars" notation.

I've told you how to specify the syntax, but deciding on the correct model is more difficult because it involves the data. Many analysts start by fitting a main-effect model and then use graphical disgnostic plots and statistical techniques to investigate whether that initial model is sufficient, or whether the data support a more complex model.

Re: glm procedure

marica14 — Wed, 24 Feb 2016 14:55:15 GMT

I read the documentation, so if i want to adjust a multiple regression on the sex, i have to put them on the model like others variables ?

Re: glm procedure

Rick_SAS — Wed, 24 Feb 2016 14:58:35 GMT

Yes, that is correct. "Adjusting for a variable" or "controlling for a variable" means that those effects are in the model, even if they are not the primary quantity that you are investigating.