Balance Table using Ranksum and Chi2

[Arto Arman]

Hi Everybody!

I'm currently trying to produce a balance table for two different samples. The variables I'm looking at are either continuous (e.g., an outcome variable between 0 and 6) or binary (e.g., socio-economic characteristics such as gender). I was able to produce a code for another project, in which I use a t-test to compare the two samples and produce a nice table for it (see code below). Now, I'd like to have the same table, but report the p-value for a Ranksum test for continuous variables and the p-value for a Chi2 test for the binary variables.

Can anyone help me or has done this before?
Thanks a lot for your help!

// set globals
global nonbin        outcome_1 outcome_2 outcome_3
global bin            gender tertiary_education convicted
global testvar1        $nonbin $bin

** Sample 1
estpost sum $testvar1 if rep==0
matrix cgmean=e(mean)
matrix cgsd=e(sd)
local cgN = e(N)

** Sample 2
estpost sum $testvar1 if rep==1
matrix tgmean=e(mean)
matrix tgsd=e(sd)
local tgN = e(N)

** exclusion off all regressors 
reg rep $testvar1
test $testvar1
scalar pvalue = (int(r(p)*1000)/1000)
local p1 = pvalue

** difference in samples
gen     cg = rep==0 
replace cg = .     if rep==.
estpost ttest $testvar1, by(cg)
eststo cg_tg 
estadd scalar pvalue: cg_tg
estadd matrix cgmean: cg_tg 
estadd matrix cgsd: cg_tg
estadd matrix tgmean: cg_tg 
estadd matrix tgsd: cg_tg

global varlist1        outcome_1 outcome_2 outcome_3 gender tertiary_education convicted

esttab cg_tg using "$tables/rcomparison.tex", replace ///
        prehead("\begin{tabular}{lcccccc}" /// // 
        "%" "\toprule" "%" /// 
        " & \multicolumn{2}{c}{Sample 1} & \multicolumn{2}{c}{Sample 2} &         &  \\" ///
        " & \multicolumn{2}{c}{N = `cgN'} & \multicolumn{2}{c}{N = `tgN'} &         &  \\" ///
        "   \cmidrule(r){2-3}  \cmidrule(r){4-5}  " ///
        " & Mean & SD & Mean & SD & \it{p-val}        & \# obs \\" /// 
        "%" "\vspace{-0.3cm}" "%" ) ///
        cells("cgmean(fmt(3) label((1))) cgsd(par fmt(2) label((2))) tgmean(fmt(3) label((3))) tgsd(par fmt(2) label((4))) p(star fmt(2) label((6))) count(fmt(0) label((7)))") ///
        nonumber noobs star(* 0.1 ** .05 *** 0.01) /// 
        keep($varlist1) ///
        order($varlist1) /// //     
        varlabels($labellist1) ///
        postfoot( /// //                "Joint test (\it{p}-value)     &        & & & & `p1'  \\" ///
        "\bottomrule" ///     
        " \addlinespace " ///
        "\end{tabular}" ///
        "\parbox{\textwidth}{\textit {Notes:} Joint significance test (P-value) = `p1'., * \(p<0.10\), ** \(p<0.05\), *** \(p<0.01\)}")

[George Ford]

I wouldn't use a t-test for balance analysis, since the results depend on sample size (which is not the problem). Use standardized differences. covbal will do it.