Adjusted means from Randomized complete block experiment where block effect is highly significant

[mclaughj]

What is the best way to obtain adjusted means from a Proc GLM model (file attached) such as:

proc glm data=invivo;

      class block treatment;

      model invivo=block treatment;

weight treatment;

      lsmeans treatment/ ADJUST=TUKEY;

Contrast 'Compare stx1-0hr with stx2-0hr' treatment 1 0 0 0 -1 0 0 0 0;

Contrast 'Compare stx1-1hr with stx2-1hr' treatment 0 1 0 0 0 -1 0 0 0;

Contrast 'Compare stx1-2hr with stx2-2hr' treatment 0 0 1 0 0 0 -1 0 0;

Contrast 'Compare stx1-3hr with stx2-3hr' treatment 0 0 0 1 0 0 0 -1 0;

   run;

The block effect is highly significant but I am having trouble identifying a method to report the adjusted means which accounts for the significant block effect.  Does this require the ESTIMATE function?

Any suggestions?

Thank you.

[SteveDenham · Accepted solution]

The best way is to analyze the data using PROC MIXED or PROC GLIMMIX, and consider block as a random effect.  In this case, standard errors of the treatment means will correctly incorporate the variance due to block.  I also notice in your contrasts that there is a time factor.  If the time reflects repeated measurements on the same experimental unit, then the analysis should also incorporate that. Until that is clear, consider the following:

proc glimmix data=invivo;

class block treatment;

model invivo=treatment;

random block;

lsmeans treatment;

LSMESTIMATE treatment

'Compare stx1-0hr with stx2-0hr'  1 0 0 0 -1 0 0 0 0,

'Compare stx1-1hr with stx2-1hr'  0 1 0 0 0 -1 0 0 0,

'Compare stx1-2hr with stx2-2hr'  0 0 1 0 0 0 -1 0 0,

'Compare stx1-3hr with stx2-3hr'  0 0 0 1 0 0 0 -1 0/adjust=simulate(seed=1);

/*Tukey adjustment is not available for the LSMESTIMATE as it is not defined */

run;

Steve Denham

[mclaughj]

Many thanks Steve.  I appreciate your time and expertise.

Sincerely,

John