Let me try and be careful and clear here.

What I meant by "aggregation" is that when we have to choose between X and Y once, we may have unclear intuitions, but if we have to choose between X and Y multiple times (given certain conditions), the choice is clear (and is Y, for example).

There are two intuitive examples of this. The first is when X causes a definite harm and Y causes a probability of harm, as in http://lesswrong.com/lw/1d5/expected_utility_without_the_independence_axiom/ . The second is the example I gave here, where X causes harm to a small group while Y causes smaller harm to a larger group.

Now, the "certain conditions" can be restrictive (here it is applied repeatedly to a fixed population). I see these aggregation arguments as providing at least some intuitive weight to the idea that Y>X even in the one-shot case. However, as far as I can tell, this aggregation argument (or anything similar) is not available for creating populations. Or do you see an analogous idea?