Just a thought on this random function. I personally would prefer 2 different random methodologies. 

The 2 options being a) random/variable and b) random/record. Meaning that if you aggregate 3 variables, in random/variable it would re-randomize which record to pick from for each variable.

a) Meaning if there are 3 variables being aggregated with this function. It could very well choose data from record 1 in var 1 and from rec 2 in var 2 and from rec 1 in var 3  again. Meaning it would be really random.

b) Keeping it random but structured would be to let it randomise for which record to use in it's entirety for all 3 variables when aggregating (as they may very well have some sort of link between them). Typically this would be more tricky to write, but I believe a fair assumption would be to let all in one of these commands be checked. If multiple variables are chosen with this option, do one random and use it.

I also note in examples above I use only 2 records as example, but note it might very well be more than that.

Thirdly but lastly. I note that there are sysmissing issues noted in the current dropdown functions list. I would prefer a tick box to say "Never include sysmis in function" as an option. This would mean that for the case above for instance, it would randomise between those that have data, or if not ticked in between all. Also being peculiar for option b as it would have to check all variables before deciding to choose whether a specific record can go into the choosing process of randomisation.