Non parametric ANOVA and Dunnet adjust

[SASAlice]

Hi all,

I am currently analyzing a study where participants answered a questionnaire about how they perceived/graded two different stimuli (CS+ and CS-) after receiving either medication A, medication B or Placebo. I succeed to perform the non parametric anova using turkey adjust :

/* Ranks on Fear */
%macro rank_Fear(dataset);

proc rank data=&dataset out=&dataset;
var Fear;
ranks r_el_Fear;
run;

/* Non-parametric ANOVA on Fear ranks */
PROC MIXED DATA=&dataset METHOD=MIVQUE0 ANOVAF;
CLASS Subject CS Grp;
MODEL r_el_Fear=CS|Grp;
REPEATED/SUBJECT=Subject Group=Grp TYPE=UN;
lsmeans CS Grp CS*Grp/ adjust=tukey;
ods select ModelInfo Tests3 diffs lsmeans;
run;
%mend;

However since it is a medication study with a placebo I am thinking that it could make sense to use dunnett adj instead
PROC MIXED DATA=&dataset METHOD=MIVQUE0 ANOVAF;
CLASS Subject CS Grp_cat;
MODEL r_el_Fear=CS|Grp_cat;
REPEATED/SUBJECT=Subject Group=Grp_cat TYPE=UN;
lsmeans Grp_cat /PDIFF=CONTROL('Group 3'') adjust=dunnett;
ods select ModelInfo Tests3 diffs lsmeans;
run;
%mend;

Now I am a bit confused as I can't find a way to look at CS*Grp_cat interaction with dunnett adj, which would be really valuable for me. Is there a way to look at interaction using Dunnett in SAS?

Thank you for your time!

[sbxkoenk]

---

@SASAlice wrote:

Now I am a bit confused as I can't find a way to look at CS*Grp_cat interaction with dunnett adj, which would be really valuable for me. Is there a way to look at interaction using Dunnett in SAS?

---

Why can't you do something like?

lsmeans CS * Gtp_cat / stderr pdiff cl cov adjust=dunnett DIFF=CONTROL('abcxyz' 'Group 3');

What happens then?

BR, Koen

[SASAlice]

Thank you for your reply.

I tried and if I am not mistaken it seems that 'stderr' is not recognize

Would you have another suggestion?

[sbxkoenk]

Hello,

It seems 'stderr' is indeed a key-word that you cannot use as an option (after the forward slash) on the LSMEANS statement.

SAS/STAT 15.3 User's Guide
Shared Concepts and Topics
Syntax: LSMEANS Statement
https://go.documentation.sas.com/doc/en/pgmsascdc/9.4_3.5/statug/statug_introcom_sect041.htm

Just get rid of 'stderr' and submit again.

BR, Koen

[SteveDenham]

Beware using Dunnett's adjustment on rank based means. Dunnett's adjustment is fairly robust, but one thing that can really mess with it is a lot of ties. If your data is not heavily tied (less than 10% ties), you should be OK.

SteveDenham

[SASAlice]

Hi SteveDeham,
Thank you for the caution bout using Dunnett's adjustment on rank-based means, especially regarding the impact of ties. How would you consider making multiple group comparaison to a controle group then if more then 10% ties?
Alice

[SteveDenham]

For comparison of multiple groups to a single control for a nonparametric analysis (Kruskal-Wallis), Dunn's test is often recommended. While also influenced by ties (as is any rank based analysis), it is not as affected as Dunnett's test. SAS does not currently offer Dunn's as a multiplicity adjustment, so you would probably have to search for it on the internet.

However, SAS does offer the method of Edwards and Berry (ADJUST=SIMULATE), which is truly distribution-free, and provides superior family-wise error control to any of the other methods available for the LSMEANS or LSMESTIMATE statement.

SteveDenham

[SASAlice]

Hi Koen,

I also tried that before but I was having an error message, however with your suggestion I tried again few ways and it seems that the main structure of the code works but the order of the variables need to be carefully chosen like following:
lsmeans Grp_cat*CS / adjust=dunnett diff=control('CSMinus' 'Group 3') cl ;

This didn't work (which may be why it took some time to figure out how to do):

lsmeans CS*Grp_cat / adjust=dunnett diff=control('CSMinus' 'Group 3') cl ;

Thank you!!

Alice