Department of Biostatistics at Rochester

SAS Helpful Hints

Proc Glm

General Linear Models. The heart of statistics. Here is the basic structure and a few of the many options available in Proc Glm.

Procedure options:
    Anova: Analysis of variance.
    Npar1way: Nonparametric one-way analysis of rank scores.
    Reg: Simple linear regression.

Model options:
    Clm: Confidence limits of predicted value
    Clparm: Confidence limits of parameter estimates

Lsmeans options:
    Stderr: Standard Error
    Pdiff: p-values for differences

Proc Glm Data=work.A;
    Class treatment gender;
    Model dependent=treatment gender other_independents
        treatment*gender
        / Clparm ;
    Lsmeans treatment / Stderr Pdiff;
    Means treatment;
    Output Out=work.glmout Predicted=pred Student=stresid;

    *estimate the treatment effects of a & b, in a 2X2 factorial design;
    *NOTE: a & b must be in the Model & Class statements.
    Estimate 'EFFECT OF A' a -1 1;
    Estimate 'EFFECT OF B' b -1 1;

    *estimate the interaction of a & b, in a 2X2 factorial design;
    *NOTE: a, b & a*b must be in the Model & Class statements.
    Estimate 'A VS. Placebo' a -1 1 a*b -1 0 1 0;
    Estimate 'B VS. Placebo' b -1 1 a*b -1 1 0 0;
    Estimate 'A+B VS. Placebo' a -1 1 b -1 1 a*b -1 0 0 1;

    *estimate the treatment effects in a dosage trial;
    Estimate 'Placebo vs Low' treatment -1 1 0 0 0;
    Estimate 'Placebo vs Medium' treatment -1 0 1 0 0;
    Estimate 'Placebo vs High' treatment -1 0 0 1 0;
    Estimate 'Placebo vs Really High' treatment -1 0 0 0 1;
    Estimate 'Placebo vs All' treatment -1 0.25 0.25 0.25 0.25;

    *estimate the interaction of treatment & level;
    *NOTE: treatment, level & treatment*level must be in the model.
    Estimate 'Level 1' treatment -1 1 treatment*level -1 0 0 1 0 0;
    Estimate 'Level 2' treatment -1 1 treatment*level 0 -1 0 0 1 0;
    Estimate 'Level 3' treatment -1 1 treatment*level 0 0 -1 0 0 1;

Run;

*We often run the following plots for each Anova(Glm).;
Proc Plot Data=work.glmout;
    Plot stresid*pred;
    Label pred='Predicted' stresid='Studentized Residuals';
Run;
Proc Rank Data=work.glmout Normal=blom Out=work.rankplot;
    Var stresid;
    Ranks nscores;
Run;
Proc Plot Data=work.rankplot;
    Plot stresid*nscores;
    Label nscores='Normal Scores' stresid='Studentized Residuals';
Run;

Please send your comments and suggestions about this web page to A. Watts (watts@bst.rochester.edu)