- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hello friends,
I have a question regarding selection of statistical test. Here is the situation: we have two treatments A & B and would like to compare differences between A & B. We have 100samples, we made two aliquots from each sample and applied both treatments. How should I treat this data - matched pair or independent two groups?
Which SAS procedure will work for me?
Thanks in advance.
DJ
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I would treat the data as matched pairs :
proc glimmix data=test plots=all;
class treat sampleId;
model x = treat / solution;
random intercept / subject=sampleId;
run;
look carefully at the residual plots for any pattern.
PG
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Hi PG,
Thanks for the prompt reply.
Is there any non-parametric test appropriate to this problem?
Thanks
DJ
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
You could do a Wilcoxon signed ranks test on the difference between treatments (A-B) for each sample. It is available in PROC UNIVARIATE.
PG
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I am very curious as to why the OP would want to do a non-parametric test. Are the response variables really non-normally distributed? If the response variable is a concentration, perhaps a lognormal distribution can be assumed, and the GLIMMIX approach appropriately modified with a dist= option. On the other hand, if there is a problem with responses being above or below limit of quantitation for the assay, I can understand wanting to move to a nonparametric test. However, if this is not the case, the detectable difference for the nonparametric test will be much larger than for the parametric test.
And of course, once those plots are examined, it may be really obvious why a nonparametric approach would be needed.
Steve Denham