IYH Dear Elena Tagliabue

Edit full disclosure and thank you to Elena: I attempted to have the sas code patch automatically generated based on general modification instructions. I do not have access to a SAS code executor or linter, hence was not able to test it.

A. To calculate the p-value for the post hoc test in the first provided .sas macro, you can use the PROC NPAR1WAY procedure with the WILCOXON option. You can request the p-value by specifying the WILCOXON option in the PROC NPAR1WAY statement and using the /TEST option in the CLASS statement.

* PERFORM THE STANDARD KRUSKAL WALLIS TEST WITH P-VALUES; PROC NPAR1WAY data=&DATANAME wilcoxon;output out=KW_MC_TMP5; CLASS &gpname / TEST; VAR &OBSNAME; TABLE &gpname*&OBSNAME / BONFERRONI; * Use Bonferroni correction for multiple comparisons; RUN;

This will add the p-values to the output dataset KW_MC_TMP5 in a new variable called p_kw. You can then use this variable in the rest of the macro to determine significance.

Kindly note that this modification uses the Bonferroni correction for multiple comparisons. If you prefer a different correction method, you can change BONFERRONI to eg DUNNETT or SCHEFFE

B. To add the p-value calculation to the second (Dunn) SAS macro in ch. 6 of your zip file, you can use the PROC MULTTEST procedure to perform the multiple comparisons and calculate the p-values. Here's the modified section of the macro that you can replace:

* Perform the pair-wise large sample approximation mcp; data nparmcp; %do u=1 %to &ngrps; %do v=%eval(&u+1) %to &ngrps; %let rsum&u=%scan(&srvec,&u," "); %let n&u=%scan(&nvec,&u," "); %let rsum&v=%scan(&srvec,&v," "); %let n&v=%scan(&nvec,&v," "); c+1; difflbl="%scan(&grpvec,&u," ")-%scan(&grpvec,&v," ")"; avdiff=abs((&&rsum&u/&&n&u)-(&&rsum&v/&&n&v)); cutoff=probit(1-(&alpha/((&ngrps*(&ngrps-1)))))*(((&n*(&n+1))/12)**0.5)*((1/&&n&u + 1/&&n&v)**0.5); symbol=" "; if avdiff>=cutoff then do; symbol="**"; end; output; %end; %end; label difflbl="Group@comparisons" avdiff="Difference@in@average@ranks" cutoff="Cutoff@at@alpha=&alpha" symbol="Significance@difference = **" c="Comparison@number" ; title1 "Large sample approximation multiple comparison procedure"; title2 "designed for unbalanced data"; title3 "%trim(%left(&ngrps)) groups: %trim(%left(&grpvec)) (respective sample sizes: %trim(%left(&nvec)))"; title4 "Alpha = &alpha"; * Print out the results of the test; proc print label split='@' noobs; run;

Replace it with the following section:

* Perform the pair-wise large sample approximation mcp using PROC MULTTEST; data nparmcp; %do u=1 %to &ngrps; %do v=%eval(&u+1) %to &ngrps; %let rsum&u=%scan(&srvec,&u," "); %let n&u=%scan(&nvec,&u," "); %let rsum&v=%scan(&srvec,&v," "); %let n&v=%scan(&nvec,&v," "); c+1; difflbl="%scan(&grpvec,&u," ")-%scan(&grpvec,&v," ")"; avdiff=abs((&&rsum&u/&&n&u)-(&&rsum&v/&&n&v)); call symput("avdiff",avdiff); call symput("difflbl",difflbl); run; ods output ParameterEstimates=pvalues; proc multtest data=&dataset.2 pvalue=&avdiff; class &group; var &var; twosided; adjust=none; output out=mcp_&difflbl pvalues=pvalues; run; data nparmcp; set nparmcp; set mcp_&difflbl; if _n_=1 then set pvalues; pvalue=p; label pvalue="P-value"; keep c difflbl pvalue; run; %end; %end; * Print out the results of the test; proc print label split='@' noobs; run;

This will add the p-value calculation to the macro and print it out in the results.

Again kindly note that the adjust=none option is used to calculate the unadjusted p-values. You can replace it w other options such as adjust=bonferroni or adjust=sidak to calculate the adjusted p-values.

Thank you Daniyel Yaacov Bilar .

I tried to implement your second point in the Dunn SAS macros, but something seems to go wrong. See SAS log comments I reported below within the code (all errors are like xxx is not valid or it is used out of proper order):

proc multtest data=&dataset.2 pvalue=&avdiff;

class &group;

var &var; *var statment is not valid;

twosided; *twosided statment is not valid;

adjust=none; *adjust statment is not valid;

output out=mcp_&difflbl pvalues=pvalues; *output statment is not valid;

run;

IYH Dear Elena Tagliabue

Kindly accept me apologies: I attempted to have the sas code patch automatically generated based on your needs w general modification instructions. I do not have access to a SAS code executor or linter, hence was not able to test it.

Let me see if I can get hold of a specialist w more resources and provide a working fix What version of SAS are you using?

Again I apologize for this not working ad initio.