|Date: ||Sun, 23 May 1999 17:29:55 +0200|
|Reply-To: ||Torben Haslund <Torben.Haslund@VBIOL.SLU.SE>|
|Sender: ||"SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>|
|From: ||Torben Haslund <Torben.Haslund@VBIOL.SLU.SE>|
|Subject: ||Split-plot in PROC MIXED - MISSING values|
|Content-Type: ||text/plain; charset="us-ascii"|
I could not have made my work without all the help I have got from all of
you - unknown friends. When I finish I will make a summary to the list of
the most fruitful tips I got.
Until now I have been "kneading" my data set for a long time, thinking I
was close to an end. But then a serious question re. missing values turned
As some of you already know, I am analyzing a split plot experiment with
covariates using PROC MIXED. Now I am wondering how to treat missing
values. There are some of them here and there. This question also relate to
the treatment of a few extreme outliers, as - if I exclude an outlier - it
will appear as a missing value.
The field experiment consist of three blocks (replicates). Eight manuring
treatments as main plots within each block. Each main plot is split into
two subplots treated and not treated with some preparations. Each subplot
is characterized with some measurements which are used as covariates -
eventually used as interaction terms with treatments.
NOW THE PROBLEM: If I have one missing value in the response variable (i.e.
representing a result from one sub plot (a)) how will this be handled by
PROC MIXED. As I have understood PROC MIXED will handle this situation. But
- now my question - should I exclude the value in the same response
variable representing the neighbor subplot (b) within the same main plot?
If I do not make that exclusion will the covariate "run away" with the
value from the measured subplot (b) without having possibility to account
for the missing value from its neighbor (a)? - Or will the keeping of this
value (b) make it possible to estimate (a) - which should be the same as
letting PROC MIXED take care of the problem?
Looking forward to fruitful comments
Department of Plant Biology
Swedish University of Agricultural Sciences
P.O. Box 7080
S - 750 07 UPPSALA, Sweden
E-mail address: Torben.Haslund@vbiol.slu.se
Home and postal address: Ostre Paradisvej 7B, DK-2840 Holte, Denmark
Home phone: +45 4541 1198