effect statement to generate interactions by group

[liyongkai800]

I am wondering is there any way to generate the interaction by groups.

say the following program generates the effects x1,x2 and x1*x2 :

proc glmselect;

effect poly = polynomial(x1-x2/degree=2 mdegree=1);

model y = poly/selection=none;

run;

However, if I have x1,x2,y1,y2 and want to generate the effects x1,x2,x1*x2,y1,y2 and y1*y2, how to modify the effect statement? Thanks.

[arthurcavila · Accepted solution]

You can add a second effect statement, but it is a lot easier to modify your model statement instead.

proc glmselect data = sashelp.baseball;
    effect poly = polynomial(nhits nrbi / degree = 2 mdegree = 1);
    effect poly2 = polynomial(nouts nerror / degree = 2 mdegree = 1);
    model logsalary = poly poly2 / selection = none;
run;

proc glmselect data = sashelp.baseball;
    model logsalary = nhits|nrbi nouts|nerror / selection=none;
run;

[liyongkai800]

Thanks, it's good to know