How to skip merging an empty master dataset in a 500 loops?

[Jae Li]

Dear Stata experts,

I'd like to ask a question about merging datasets in a loop when no observations merged and please see my code here:

forval i = 1/500{
preserve
bsample

reg y x1 x2 i.sic3 i.year, vce(robust)

gen b_x1 = _b[x1]
gen se_x1 = _se[x1]
gen t_x1 = abs(b_x1 / se_x1)

* keep significant coefficients
keep if t_x1 >= 1.65

* if no observations left, exit
 if c(rc) != 0 {
  exit (c(rc))
 }

 else {
   gen index = `i'
    keep gvkey fyear
    
    merge m:1 gvkey fyear using "base.dta"
    keep if _merge == 3
    drop _merge
restore
}
}

However, it stops and shows an error message after a few times of merging. I want it to skip merging if no observations left in the master dataset after

keep if t_x1 >= 1.65

and start another loop, and keep going until 500 times. Is there any advice on how to resolve this issue?

    Result                      Number of obs
    -----------------------------------------
    Not matched                        49,576
        from master                         0  (_merge==1)
        from using                     49,576  (_merge==2)

    Matched                                 0  (_merge==3)
    -----------------------------------------
(49,576 observations deleted)
no observations
r(2000);

end of do-file

Many thanks!
Jae

[Clyde Schechter]

To get around the immediate problem with -merge- you can do something like this:

else {
   gen index = `i'
    keep gvkey fyear
    if _N > 0 {
        merge m:1 gvkey fyear using "base.dta"
        keep if _merge == 3
        drop _merge
    }
    restore
}

That said, I see some other questionable aspects of your code. After completing the -merge-, your next command is a -restore-, which will obliterate the results that you just created with the -merge-. Don't you want to put a command in there to save the results somewhere?

Also, further up in the code, -if c(rc) != 0- is pointless there. And, in particular, it does not tell you whether there are any observations left at that point. -c(rc)- is the system variable that contains the return code of the immediately preceding command, which is -keep if t_x1 >= 1.65-. The only way that command can fail is if variable t_x1 does not exist. But if that is the case, Stata will halt before even executing -if c(rc) != 0-. So if you even reach that command at all, c(rc) is necessarily 0. So that -if- guard does nothing at all.

Finally, just to make the code a bit briefer and easier to follow, the three-command sequence of -merge ...- -keep if _merge == 3- and -drop _merge- can be combined into a single command: -merge m:1 gvkey fyear using base, keep(match) nogenerate-. (Note also that neither the quotes nor the .dta are necessary in specifying the file base.dta here.)