The simplest way I found to explain that the representative sample size is independent of population is via coin flips and dice:
We can flip a coin an infinite number of times. But to get a good handle on whether it's biased or not, we only need to sample a very finite number of flips.
My point was that sample size is too small. According to Standard Occupational Classification [https://www.bls.gov/soc/] there are 459 broad occupations.
Multiply that by age group and you still haven't tossed a coin even a single time for every combination of occupation/age group.
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.
IF done right, 500 can be representative. Although considering the number of jobs out there combined with age groups and gender, I guess you'd need a few thousand carefully selected participants to get valid results.
No need to carefully select participants. If you do random sampling with a large enough sample, the proportion of all subcategories should reflect the population at large.
Well, a survey usually tries to extrapolate numbers from a sample smaller than the total. They do this by trying to get a representative sample. Also if you were to ask each and every American about their stance on remote work I think that would be called a vote and not a survey.
That is a VERY rough (actually false) sentance, considering they have only surveyed ~500 Americans out of 205'950'000+ population at working age.
> We surveyed over 500 Americans...