test of difference in median

[LanMin]

May I ask you if my interpretation of SAS results are correct.  thanks !!!

my data has a class variable u_rank, it takes on two values 1, and 4,

my test variable is disclose (a continuous variable).

I use proc npar1way to test difference in  median of  disclose between  u_rank=1 and u_rank=4.

see below for code (attached)

my  interpretation is :

median is zero for both u_rank=1 or 4

I am not certain:

The NPAR1WAY shows z value to be significant.

could I conclude : even though median value of disclose is 0 for both u_rank=1 and u_rank=4  , but median test still shows significant difference between the class variable u_rank?

thanks !!!

Lan

[PGStats · Accepted solution]

Yes you could. Your data likely has a fair proportion of disclose=0 observations in both u_rank groups. Run the following simulation:

data test;
call streaminit(987676);
do u_rank = 1, 4;
     do i = 1 to 2520;
          if mod(i, 2) then disclose = 0;

               else disclose = rand("NORMAL", u_rank*0.05);
          output;
          end;
     end;
drop i;
run;

proc univariate data=test;
class u_rank;
var disclose;
histogram;
run;

proc npar1way data=test median wilcoxon;
class u_rank;
var disclose;
run;

PG

[LanMin]

Thanks so much, PG .