You actually have two criteria that possibly creates an impossible solution: 12 months (per year) and 10% per month. 10% times 12 months = 120% so there will be duplicates if you are indeed sampling every month. A second complication is 129 courses with 12 teachers so as many as 9 of your 12 teachers teach at least one more class than others. You may need to decide which is more important "evenly distributed" or "not duplicated".   I would be very tempted to randomly order the data and create partitions numbered 1 to n, where n is the actual total number evaluations. Then for the first evaluation use group 1, second use group 2 and so forth.   This would insure no duplicates assuming no changes in the numbers of teachers or classes. But is your data actually that stable?     ... View more