And OP's point is that you don't necessarily need to have representation for all of those occupation/age group combinations. It's highly unlikely there isn't high amount of overlap in perception among those occupations. There's going to be supersets of those occupations that are fairly homogenous. So you really don't need a huge sample of respondents and coverage of all of those possibilities. You just need the makeup of the sample to be reasonably representative, and that's the problem with this poll. Seems like it was conducted in a way that the respondents are not representative.