Test of lack of fit for quadratic model without intercept

[Celine_France]

Hi everyone,

I made a quadratic regression thanks to the REG procedure. As I want to test a quadratic model without intercept, I specified the "noint" option.

Ex : proc reg data=tab; model Y = X X*X / noint; run; quit;

Now, I want to test the lack of fit of my regression. So, I know that I can do that with the RSREG procedure.

Ex : proc rsreg data=tab; model Y= X X*X / covar=1 lackfit; run; quit;

But, it computes the LOF with the intercept and I don't the intercept in the model.

I check in the SAS help but I've found no option to suppress the intercept. Does someone know how to do this ?

By advance, thanks a lot

Céline

[PGStats]

Well, it's about time somebody tries an answer. I suppose that you have a very good reason for not fitting an intercept. If you included the LACKFIT option in the MODEL statement of the REG procedure, you should see the test for lack of fit. Try for example:

data test;
call streaminit(4321);
do i = 1 to 4;
do x = 1 to 10;
  x2 = x*x;
  y = x + 0.1*x*x + rand('NORMAL',0,0.2);
  output;
end;
end;
run;

proc reg data=test;
model y = x x2 / noint lackfit;
run;

To perform this test, you need duplicate X values, because the lack of fit test (à la Draper & Smith) relies on those.

PG