Date:         Thu, 8 Sep 2005 22:17:23 -0700
Reply-To:     David L Cassell <davidlcassell@MSN.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         David L Cassell <davidlcassell@MSN.COM>
Subject:      Re: Analysis of an experiment
In-Reply-To:  <200509081055.j88AjV61013416@malibu.cc.uga.edu>
Content-Type: text/plain; format=flowed

dopijeme@YAHOO.COM wrote:
>   Perhaps I can tap some expertise to help knock the
>rust off as I try to get my mind around the concepts
>involved in an analysis of data from a lab.
>
>   In this experiment, the investigators applied
>several treatments prepared in potentially two
>different manners.  As far as I know, these are fixed
>effects, i.e. only their levels are of interest.  Each
>subject received on one treatment/preparation
>combination.  They did not prepare all treatments in
>the two different manners.  The number of subjects per
>treatment/preparation varies from a low of 2 to a high
>of 23.
>
>   I have plotted the means for each treatment and for
>some of the points for which I have both preparations
>there appears to be a difference.
>
>   Assuming that the outcome is normally distributed,
>can I pool these data and develop indicator variables
>for treatment and preparation?  By this I mean, is it
>appropriate to have data concerning
>treatment1/preparation1 if there is no corresponding
>treatment1/preparation2? Perhaps yes, if I use GLM and
>either contrast or lsmeans statements to test specific
>combinations?  What are your suggestions?

Here are some suggestions:

[1] You ought to be able to combine data from both investigators, but
then you WILL have a random effect: INVESTIGATOR.

[2] Depending on your dependent variable(s), your model(s) and your
assumed error structure, you're probably looking at PROC MIXED
and/or PROC NLMIXED.  Since you have missing cells, you ought to be
looking at PROC MIXED anyway, even if you decide that you want to
treat INVESTIGATOR as a fixed effect.  (I wouldn't.)

[3] If you're seeing patterns in your data when you plot the points,
then you ought to be using that information to design the models.
Is the 'investigator' effect additive or multiplicative or neither?  Do
the treatments show additive behavior, or are there interactions?

[4] Using investigator as a random effect, you can forget the
treatmentA/preparationI stuff.  Simply look at the means for the
treatments, and/or the contrasts between treatment levels.

HTH,
David
--
David L. Cassell
mathematical statistician
Design Pathways
3115 NW Norwood Pl.
Corvallis OR 97330

_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today - it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/