*****************************************************************
The SAS Proc GLM code and output for the rice/CO2/nitrogen
example 7.1.  The experiment was a split-plot
with whole-plots in an RCBD with random block effects. CO2 is the
whole-plot factor and nitrogen level is the split-plot factor.
*****************************************************************;

 proc glm data=subplots;
   class  block co2 nitrogen;
   model fertility = block co2 block*co2 nitrogen
                 nitrogen*co2;
   lsmeans co2*nitrogen;
   estimate 'low vs high N' nitrogen -1 0 1;
   estimate 'amb CO2: low vs high N'
        nitrogen -1 0 1 co2*nitrogen -1 0 1 0 0 0;
   test h=co2 e=block*co2;
   output out=glmout student=sresid p=yhat;
   run;



Dependent Variable: fertility
                       Sum of    Mean
 Source          DF   Squares  Square F Value  Pr > F
 Model           11    126.13   11.47    7.04  0.0011
 Error           12     19.54    1.63
 Corrected Total 23    145.67

                                 Mean
 Source        DF Type III SS  Square F Value  Pr > F
 Block          3       12.53    4.18    2.57    0.10
 CO2            1        5.51    5.51    3.38    0.09
 Block*CO2      3        5.39    1.80    1.10    0.39
 Nitrogen       2      100.29   50.15   30.80  <.0001
 CO2*Nitrogen   2        2.41    1.20    0.74    0.50

 Least Squares Means for Fertility
 CO2      Nitrogen           LSMEAN
 AMB        High              91.05
 AMB        Low               95.35
 AMB        Medi              94.19
 FACE       High              91.25
 FACE       Low               96.28
 FACE       Medi              95.94

 Tests of Hypotheses Using the Type III MS
            for Block*CO2 as an Error Term
                              Mean
 Source   DF   Type III SS  Square   F Value  Pr > F
 CO2      1        5.51       5.51      3.06    0.18

                       Standard
 Parameter             Estimate  Error t Value Pr> |t|
 low vs high N             3.91   0.64    6.13  <.0001
 amb CO2: low vs high N    3.14   0.90    3.48  0.0046



**************************************************************
Note that tests and standard errors of differences between
whole-plot treatment levels at specific split-plot treatment
levels must be computed by hand, using the analysis of
variance output.
**************************************************************