Problem with dividing a data set randomly but consistent with a random variable

[Steffen Pluetzke]

Dear Statalisters,

I generated a dataset and want to randomly divide it into 70/30. In the beginning of the do-file I used the command set seed, but somehow the division changes. Therefore a regression I ran on the 70% gives different estimates everytime. How do I get a consistent division based on a random variable?
The code is below. ZV2 is the random variable I generated and OOS should divide the data set OOS=0 (70% of the observations) and OOS=1 (30% of the observations). However, if I sum my dependent variable, the summary statistics of the subgroups are different. How can I divide the data set randomly, but with the same observations in the same subgroups everytime I run the do-file?
I really appreciate your help.

set seed 100
gen ZV2 = runiform()
label var ZV2 "random variable"
xtile OOS = ZV2, nquantiles(10)
replace OOS = 0 if OOS <=7
replace OOS = 1 if OOS >7
label var OOS "indicator"
global y depvar

sum $y if OOS==1
sum $y if OOS==0

Kind regards

Steffen

[William Lisowski]

Is it at all possible that earlier in your program you are doing something that leaves your dataset in a different order? For example, if you are sorting your data and there are ties — that is, multiple observations with the same values for the sort key variable(s) — the output of help sort tells us

Without the stable option, the ordering of observations with equal values of varlist is randomized.