Question

… and add columns for differences and t-statistics.

I learned how to make a quantile by quantile table of means and how to add a column/row of differences here (thanks to @lejohn).

Now instead of each row as one quantile of one variable, I would like each row to be a different variable and each cell would be the mean value for each variable for the individuals that fall in each column for the quantile of a given variable.

I can calculate the cell entries easily with tabstat, but I would like the variables in the rows and the quantiles in the columns (tabstat produces the transpose). I would also like the ability to difference columns (as in my first question) and calculate t-statistics for the cell differences.

I feel like the intermediate step is to reshape to long data with three columns: id (here acc_d), variable name, and variable value. But I can’t figure out how to do this and I may be stuck in an R paradigm.

Here is an example of the type table I would like to make

and here is some code with which I have been (unsuccesfully) tinkering

* generate data
clear
set obs 2000
generate acc = rnormal()
generate r1 = rnormal()
generate sar1 = rnormal()
generate arbrisk = rnormal()

* generate quantiles for for a and b
xtile acc_d = acc, nquantiles(10)

* form table (at least my attempts)
* w/ tabstat (but transposed and can't manipulate columns)
tabstat acc r1 sar1 arbrisk, stat(mean) by(acc_d) nototal 

* my attempts to reshape fail, but I would want something like to following to use tabulate
* acc_d   variable    value
* 1       acc         0.01
* 1       r1          1.03
* 1       sar1        -0.03
* 1       arbrisk     0.05
* 2       acc         1.01
* 2       r1          2.03
* 2       sar1        0.03
* 2       arbrisk     1.05

Thanks!

Answer 1

Here I would proceed a bit differently. I would first of all gather the information required to compute the difference and the t statistic

foreach v of varlist acc r1 sar1 arbrisk {
    summarize `v' if acc_d == 1
    local m_`v'_1 = r(mean)
    local var_`v'_1 = r(Var)
    local n_`v'_1 = r(N)
    summarize `v' if acc_d == 10
    local m_`v'_10 = r(mean)
    local var_`v'_10 = r(Var)
    local n_`v'_10 = r(N)
}

Then I would proceed by collapsing and transposing the data

collapse (mean) acc r1 sar1 arbrisk, by(acc_d)
xpose, clear varname 
drop if _varname == "acc_d"
order _varname
forvalues n = 1 / 10 {
    rename v`n' acc_d`n'
}

In a last step, I would add the difference and the t statistic:

generate diff_d10_d1 = . 
generate tstat_d10_d1 = .
foreach v in acc r1 sar1 arbrisk {
    replace diff = `m_`v'_10' - `m_`v'_1' if _varname == "`v'"
    replace tstat = (`m_`v'_10' - `m_`v'_1') / sqrt((`var_`v'_10'/`n_`v'_10') + (`var_`v'_1'/`n_`v'_1')) if _varname == "`v'"
}

And finally print the results:

list, abb(12) noobs

Hope this helps.