[nilkn]

He's using rich people as an example because in order to draw any inferences here you have to have a large sample set of people who could conceivably live in either a large house or an apartment, with no financial difficulties either way. Rich people are the simplest sample set to look at which trivially satisfies these criteria.

His basic point is that if your sample set consists of people who don't actually have the choice of buying a large home, then that's a major confounding variable whose influence you cannot reliably determine on the results. Beyond that, the sample set also must be large enough for it to not be anecdata. A single couple doesn't imply much about societal trends at large, but the behavior of an entire class of people -- the wealthy -- does, to at least some degree.