[Mansoor Afzali]

Sorry for the confusion here. There are 80,000 new directors who are available to be picked. Some are picked by one firm and some by other. I have to retain the director who was chosen and randomly add 1,000 directors from the 79,999 left as "unpicked" directors. It will lead to an 80 million pairs and that's exactly what I am looking for.

[Clyde Schechter]

I don't understand. There are 80,000 directors that are "picked" and the same list is also the "not picked" directors? I don't get it at all.

And in any case, you do realize that when you start with 80,000 picked directors and you match each of them with 1,000 other directors (regardless of where those come from), you end up with 80 million pairs of directors. Are you sure that is what you want?

[Mansoor Afzali]

The list of picked directors is around 80,000 for 14 years but for each director selected, I have to assign 1,000 other directors randomly from the 80,000 as not picked ones. I am trying to replicate a paper which uses this technique to proxy for other potential candidates that could've been picked but weren't.

[Clyde Schechter]

How large is the file of directors? How many are picked? How many are unpicked?

[Mansoor Afzali]

Thank you so much Clyde. This works like a charm. One more question, how would I assign 1,000 random directors (not picked) to every director who was picked. I have a big list of all directors along with their information and want to know how to pair the picked director with 1,000 random directors and then run a logit model.

[Clyde Schechter]

Thank you for using -dataex- on your first post!

I believe you want the following:

tempfile copy
save `copy'

//    PAIR UP EACH OBSERVATION WITH ALL OF THE SAME DIRECTOR'S OBSERVATIONS FROM
//    THE PRECEDING YEAR
rangejoin fyear -1 -1 using `copy', by(directorid) prefix(prior_)

//    DETERMINE WHETHER ANY WERE IN THE SAME INDUSTRY OR FIRM OF SAME SIZE TERCILE
by directorid fyear, sort: egen byte same_industry = max(ind == prior_ind) if !missing(prior_ind)
by directorid fyear: egen byte same_size = max(tsize == prior_tsize)if !missing(prior_tsize)

//    CALCULATE MEAN PERFORMANCE IN PRIOR YEAR
by directorid fyear: egen mean_prior_perf = mean(prior_perf)

//    REDUCE TO ONE OBSERVATION PER DIRECTOR PER YEAR
drop prior_*
by directorid fyear: keep if _n == 1

-rangejoin- is written by Robert Picard and is available from SSC. To use it, you must also install -rangestat-, by Robert Picard, Nick Cox, and Roberto Ferrer, also available from SSC.