Run a loop by survey wave

[Alessandro Sovera]

Hello, I need to run the loop below separately for each wave of my survey as identified by the interview_nb variable. Since -foreach- does not admit the use of by, my super basic idea was to run each command preceded by -bysort interview_nb- but it does not seem the most efficient way to do that.

ds, has(char deflate)

foreach x in `r(varlist)' {
g `x'_99pct=`x'
g `x'_95pct=`x'
g `x'd_99=`x'_d
g `x'd_95=`x'_d
qui su `x',d
replace `x'_99pct=. if `x'>=r(p99)&`x'!=.
replace `x'_95pct=. if `x'>=r(p95)&`x'!=.
replace `x'd_99=. if `x'>=r(p99)&`x'!=.
replace `x'd_95=. if `x'>=r(p95)&`x'!=.
}

Is there a way to run this loop by survey way in a more concise way? Thanks

[Clyde Schechter]

Well, the following is more concise:

foreach x of varlist `r(varlist)' {
    quietly summ `x', detail
    foreach n of numlist 95 99 {
        gen `x'_`n'pct = cond(inrange(`x', r(p`n'), .), ., `x')
        gen `x'd_`n' = cond(inrange(`x', r(p`n'), .), ., `x'_d)
    }
}

Note: Not tested -- beware of typos.

That said, I'm not sure that the conciseness gained is a good thing--it's less transparent than the original code. I think the real gain here is replacing separate -generate- and -replace- commands with a single -generate- command using the -cond()- function. But the looping over 95 99 I think is not so advisable. A matter of taste, I suppose.

[Nick Cox]

Small detail:

In the code in #1 it is unnecessary to exclude the missing values with the condition

 
 `x' != .

They are missing already and you don't want to change that.