Date: Sun, 4 Oct 2009 15:19:21 -0700
Reply-To: j1een <jess.nagel@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: j1een <jess.nagel@YAHOO.COM>
Organization: http://groups.google.com
Subject: Re: Help with RBCD and SAS code
Content-Type: text/plain; charset=ISO-8859-1
On Sep 28, 11:12 am, Warren.Schlec...@TPWD.STATE.TX.US (Warren
Schlechte) wrote:
> Although no expert, I will add my understanding as well.
>
> With a fixed effect, you are assuming the inference will only apply to
> the set of objects sampled. With a random effect, you are assuming that
> the set of objects investigated are a sample from a larger population.
> Thus, you have a sample, and need to account for the fact that your
> sample has variance, and that a sample can only tell you so much about
> the larger population. Because of the narrower inference with the fixed
> effect idea, you might see a treatment effect that is significant, but
> when you broaden the inference space, your variance increases, meaning
> your treatment effect may no longer be strong enough to be significant.
>
> Proc GLM assumes a narrow inference space (Fixed Effects); Proc MIXED
> assumes a broad inference space, but narrower inference spaces can be
> defined by reducing the number of random effects in the model.
>
> Warren Schlechte
>
> -----Original Message-----
> From: Steve Denham [mailto:steve...@YAHOO.COM]
> Sent: Monday, September 28, 2009 9:07 AM
> Subject: Re: Help with RBCD and SAS code
>
> I would guess that GLM is giving significant results in the type 3
> table, but what do you find in the random section--particularly after
> changing to:
>
> PROC glm data=dataset;
> class block treatment;
> model samplevar = block treatment block*treatment;
> random block block*treatment/test;
> run;
>
> You need the /test option on the random statment to test against the
> "proper" mean square.
>
> By the way, the GLM documentation says this about random effects:
>
> Note:PROC GLM uses only the information pertaining to expected mean
> squares when you specify the TEST option in the RANDOM statement and,
> even then, only in the extra tests produced by the RANDOM statement.
> Other features in the GLM procedure-including the results of the LSMEANS
> and ESTIMATE statements-assume that all effects are fixed, so that all
> tests and estimability checks for these statements are based on a
> fixed-effects model, even when you use a RANDOM statement. Therefore,
> you should use the MIXED procedure to compute tests involving these
> features that take the random effects into account; see the section PROC
> GLM versus PROC MIXED for Random-Effects Analysis and Chapter 56, The
> MIXED Procedure, for more information.
>
> Steve Denham
> Associate Director, Biostatistics
> MPI Research, Inc.
>
> ----- Original Message ----
> From: j1een <jess.na...@YAHOO.COM>
> To: SA...@LISTSERV.UGA.EDU
> Sent: Saturday, September 26, 2009 7:50:09 PM
> Subject: Help with RBCD and SAS code
>
> Hey all.
>
> I have a randomized complete block design for which I am trying to
> decide on the correct Proc statement. In a nutshell, I have a block
> (3 levels), treatment (4 levels), plot (3 reps within each treatment),
> rep (2 within each plot) and samplevar. It's been a while since I've
> used SAS. I assume that I can use either GLM or MIXED but when I
> tried both, I received different output (i.e., glm outputs highly
> significant treatment effect, mixed says no effect). In my design,
> treatment is the only fixed variable. Here's the code:
>
> PROC glm data=dataset;
> class block treatment;
> model samplevar = block treatment block*treatment;
> random block block*treatment;
> run;
>
> OR....
>
> PROC MIXED data=dataset;
> class block treatment;
> model samplevar = treatment;
> random block block*treatment;
> run;
>
> Anyone have any idea what I've done wrong? Again, my SAS knowledge is
> rusty so I'd appreciate any help! Thanks!
Thanks everyone!
| 
