SAS vs. R... round two: FACTORIAL DESIGN WITH MULTIPLE TREATMENTS

That's right, folks, the great statistical champions are battling it out again in "SAS vs. R: Round 2!" (or, for extra intensity, "round ii"). Today I would like to discuss... FACTORIAL DESIGN WITH MULTIPLE TREATMENTS and SAS's great item for this, PROC MIXED. What is PROC MIXED? PROC MIXED is a SAS command that allows us to do, essentially, a ton of partial regressions within a multilinear regression in order to test statistics about the mean and the variance. What does R have to offer for this same process? I am not sure...

But let me tell you a bit about the SAS. Suppose, for example, we wanted to look at individual means and variances for four treatments blocked on one factor in a RCBD. SAS would say:

PROC MIXED;

CLASS TREATMENT;

MODEL RESPONSE VAR= TREATMENT / DDFM=SATTERTH OUTP=SUMMARY;

LSMEANS TREATMENT / CL;

PROC UNIVARIATE NORMAL PLOT DATA = SUMMARY;

VAR RESID;

PROC GLM DATA=SUMMARY;

CLASS TREATMENT;

MODEL RESID = TREATMENT;

MEANS TREATMENT / HOVTEST= LEVENE(TYPE=ABS);

RUN;

Notice that there is no mention of said means, variances, or amounts of treatment factors in the above SAS. That is because of the "magic" of PROC MIXED. PROC MIXED is able to use the class statements to create the treatment variables. BEST OF ALL... if you have interaction, PROC MIX can be instructed to create variables for the interaction, and test their significance...this is AWESOME, because even with 3 treatments and 3 blocks (1 factor) you are looking at 3^3 possiblities. Do we REALLY want to write all those variables?

I am not sure what R has to offer to combat this... R? Anyone?

I admit that discovery of PROC MIXED has reduced homework time from several hours down to about 30 minutes. Ah.. sigh...

(By the way, every time I have to write "mu" <-- the Greek letter < "mu2" I can't stop laughing. "mu" is not less than "mu2," at least, according to a certain video game that shall not be *pokemon* mentioned.