Bug (?) in collapse percentiles with iweight

[Mauricio Caceres]

In the most recent version of Stata (17) the behavior of collapse percentiles with iweight may have a bug (or perhaps this is an intentional change; is this documented anywhere?) To illustrate:

. clear
. qui set obs 3
. gen iw = -1
. gen x  = _n
. collapse (median) x [iweight = iw]
. disp x
1

In previous versions of stata, the above snippet returns 2 instead of 1.

[Hua Peng (StataCorp)]

Thanks for the example. It looks like a bug and we will try to get it fixed in a future update.