drop obeservations according to some rules

[Flora Yin]

Hello!

I have a set of firms for the sample period 2010-2019 (some firms may have less than 10 years' records), there is a variable "first", for each firm-year observation. It is an indicator variable that =1 if it is the first time the firm meet some condition. A firm can only be "first" once throughout the sample period, and it is possible that some firms would never be "first" in the sample period. I want to delete all the observations after the year the firm coded 1 for "first", how can I achieve this?

Here is some data:

year    firm    first
2010    0035    0
2011    0035    0
2012    0035    0
2013    0035    0
2014    0035    1
2015    0035    0
2016    0035    0
2017    0035    0
2018    0035    0
2019    0035    0
2013    0042    1
2014    0042    0
2015    0042    0
2016    0042    0
2017    0042    0
2018    0042    0
2019    0042    0
2010    0050    0
2011    0050    0
2012    0050    0
2013    0050    0
2014    0050    0
2015    0050    0
2016    0050    0
2017    0050    0
2018    0050    0
2019    0050    0

And after deletion I would like the data looks like this:

year    firm    first
2010    0035    0
2011    0035    0
2012    0035    0
2013    0035    0
2014    0035    1
2013    0042    1
2010    0050    0
2011    0050    0
2012    0050    0
2013    0050    0
2014    0050    0
2015    0050    0
2016    0050    0
2017    0050    0
2018    0050    0
2019    0050    0

Thanks a lot for any kind help!

[Flora Yin]

I think I solved it after reading https://www.statalist.org/forums/for...ce-of-an-event (sorry I searched but didn't see this post before I ask this question)

by firm (year), sort: keep if sum(first) <= 1 & sum(first[_n-1]) == 0

Although I tried and think it works, can anyone confirm me this is correct or not? Thanks!

[William Lisowski]

I agree that what you have is correct, and also note that an alternative, at least for your example data, is

by firm (year), sort: keep if sum(first) == 0 | first == 1

[Flora Yin]

Thanks William, your suggested code is neater