2 Comments

I think i prefer pure random sampling to stratified sampling.

Let's say there's a population with a deep ethos of non-participation in the larger society (e.g bedouin, certain Jewish groups, and perhaps Amish). If you use stratified sampling to pick those of them who do believe in participation you will necessarily get someone unrepresentative.

Further, stratified sampling requires making arbitrary decisions about which traits are relevant, and i believe it's bound to result in distortions and under-representation of other traits.

Expand full comment

Straight random sampling needs to be tied to quasi-mandatory service (like court juries), otherwise the self-selection bias is likely to generate unacceptably unrepresentative bodies in most societies (relatively fewer young people, women, and minorities who have historically been oppressed, will agree to serve). I advocate such quasi-mandatory lotteries (not stratified sampling) for final law-adopting juries. Stratified sampling is useful for intermediary bodies that will serve for much longer duration to improve both diversity and representativeness, When stratified sampling uses some specific criteria, that does NOT harm the representativeness of other traits. Indeed, the closer the sample is in terms of education (if that is a criterion with a quota) other traits such as age and income will also become closer to that of the population. The stratified characteristics are not in competition, they are supportive.

Expand full comment