gen _var4=(_var1!=0)+(_var2!=0)+(_var3!=0)

may be better in some cases where you have missing values or string values in your observations.

EX:
your command above still counts missing values, if you wanted to run the same command but not count missing values you could run:

egen _var4 = rownonmiss(_var*) if _var1!=0 | _var2!=0 | _var3!=0

Another approach is that the rononmiss() option can count the number of non-zero, non-missing observations in a row, even if there are string values present by using the "strok" option, e.g.,

egen _var4 = rownonmiss(_var*) if _var1!=0 | _var2!=0 | _var3!=0, strok

That unab vars : *var* is another good trick. Anytime STATA syntax accepts a "varlist" you should be able to use the different wildcard or shorthand operators listed under the varlist helpfile.

Another quick thing that I ran into when implementing this code was that my vars_sum macro was getting cut off by running it through the substr to get rid of that last plus sign. As you probably know, STATA's max string length is 244 characters. So if you have long variable names or many variables to check for non-zero values, your formula might hit the max character length for strings. Since the maximum characters allowed in a macro is much larger, I kept my formula in the macro domain like so:

unab vars : _var1-_var39;
local vars_count : word count `vars';
local z = 1;
foreach x of local vars {;
if `z' == `vars_count' {;
local vars_sum "`vars_sum' (`x' != 0 & `x' < .)";
continue;
};
else {;
local vars_sum "`vars_sum' (`x' != 0 & `x' < .) +";
};
local z = `z' + 1;
};
gen _var_count = `vars_sum';

My code is delimited with a ';', and I needed to count all non-zero non-missing values, hence the addition of the `x'<. statement to your formula.

As it turns out, unab is quite flexible in its pattern recognition capabilities. For example, unab vars: *var* would pick out all the variable names that contained the string "var" anywhere. So, sorry about the code bloat.

Thank you for visiting. I'm glad you like the place.

To get all the variables that start with "_var", couldn't you just write:

unab vars : _vars*

I use that all the time when working with numerically named variables with a common stub.

reshape long _var, i(id) j(varnum)
gen _varnonzero = (_var!=0) // are you missing "& (_var!=.)" ?
bysort id: egen count = total( _varnonzero)
drop _varnonzero
reshape wide _var, i(id) j(varnum)
rename count vars_sum