Regression validation

[matt23]

What is the easiest way to take around 80% and data and test it against the rest 20%? I'm using regression and want to find the best model and RMSE and R-squared and AIC etc. are very close. And by easiest I mean as for someone new to SAS.

Or Am I should I just pick the lowest RMSE or the best model from the SELECTION option? I have over 80,000 observations and the numbers are really close

[PaigeMiller]

a simple way to generate a validation data set and a training data set with 20/80 split

data rand;
    set have;
    random=ranuni(0);
run;
data train_plus_validate;
    set rand;
    original_y=y;
    if rand>0.8 then y=.;
run;
proc reg data=train_plus_validate;
    model y=x1 x2 x3;
    output out=predicteds p=predicted;
run;
    

By setting y to missing in the validation data set, these observations are not used in creating the regression model, but you do get predicted values for each of them (that's what PROC REG does if the y is missing); you can then see how close the predicted values are to variable ORIGINAL_Y.