Testing eqality of medians between two variables/separate collumns

[Jacob Ghotbi]

Hello, how should I go forth in testing eqality of medians between two variables/separate collumns? I am trying to compare the median ages between two variables that aren't grouped together. I unerstand that I should use the Wilcoxon signed-ranks test, but for me this only works when they are grouped into the same variable.

Many thanks in advance.

[Rich Goldstein]

for -signrank- or for -median-, the data need to be in one variable (column); depending on your data setup, about which you say nothing, you might want to use -reshape- or -stack- (or possibly something else) to achieve this; if you were to give a data example, using -dataex- and posting within CODE blocks (see the FAQ), more specific advice could be given

added in edit - you could probably use the bootstrap on your data as is, but,, again, without a data example, one can't be sure and can't give good advice

[Nick Cox]

Further, that test is only a test of equality of medians with extra assumptions.

[George Ford]

I don't think Wilcoxon-Mann-Whitney is not really a test of medians. There's a few papers on that issue.

ranksum is an option.

qreg would work and is straightforward, but again the data would need to be stacked.

sysuse auto, clear
tabstat mpg, by(foreign) stats(mean p50)
reg mpg foreign
qreg mpg foreign  //defaults to p50

[George Ford]

This might work. Just replace "mpg if foreign==0/1" with the variables of interest.

sysuse auto, clear
program diffmed, rclass
          quietly summarize mpg if foreign==0, detail
          local med0 = r(p50)
          quietly summarize mpg if foreign==1, detail
          local med1 = r(p50)
          return scalar difference = `med0'-`med1'
  end

bootstrap r(difference), rep(1000): diffmed