[Nick Cox]

Niels: Some confusion here. My comment in #13 refers to program varlast. It is entirely relevant to that program if used by itself. Gerald's post in #12 was not what I was replying to. It wasn't visible to me when I posted my comment.

Anyone: Note further that varlast's criterion of not missing is not equal to system missing. It therefore counts extended missings .a to .z as non-missing.

[Niels Henrik Bruun]

Sorry Nick. I misunderstood. And wrt

 st_numscalar("r(last)", max(select((1..rows(st_data(.,.)))', st_data(., varname) :!= .)))

in varlast you are of course correct. But I think that it would be more fair to mention that by changing != to < that would be solved, ie

 st_numscalar("r(last)", max(select((1..rows(st_data(.,.)))', st_data(., varname) :< .)))

This can also be seen by:

. mata
: x = 1,2,.,.a, .g

: x :< .
       1   2   3   4   5
    +---------------------+
  1 |  1   1   0   0   0  |
    +---------------------+

[Nick Cox]

Niels: You are of course right that using < rather than != gets you non-missings in Stata's usual sense, but I have no idea whether Gerald was deliberate in coding that way. I just pointed out how the program behaves and (perhaps wrongly) took it that the change to get other behaviour was evident.

[Gerald Gjini]

Thanks guys, I incorporated that change. Thank you for the explanation, I am atill long ways from getting any decent at mata so this was much appriciated.

[Fahad Mirza]

Just a quick question on this. How do we keep the last non missing maximum row by group.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(group value)
1   10
1   20
2   10
2   30
3   10
3   30
3    .
4   10
5 49.5
5   50
end

Edit: Just to add context. I would like to keep the row that contains the highest value for each group. As there is a missing value I would like to ignore that and keep the last non-missing row in that group. Hope this helps

[Mike Lacy]

It sounds like you want:

egen mvalue = max(value), by(group) // does not treat missing as max
gen tiebreaker = runiform()
bysort group mvalue (tiebreaker): keep if _n == 1

[Nick Cox]

#20 is a non-Mata question. See https://www.stata.com/support/faqs/d...t-occurrences/ in any case. In addition as you don't want to keep missings, you can simplify things by dropping them directly and then going

drop if missing(value) 
bysort group (value) : keep if _n = _N

As Mike Lacy hints, what happens with ties for maximum may need some thought.

bysort group (value) : keep if value == value[_N]

keeps the ties for maximum.