3 ms·
That is not how sample sizes work. If the sampling was actually random, a sample size of 1000 is way more than enough to measure a 25/75 split in a population.
by krastanov 4y ago
That is not how sample sizes work. If the sampling was actually random, a sample size of 1000 is way more than enough to measure a 25/75 split in a population.
The sample size gives you the "random error" in your estimate, and that scales roughly as 1/sqrt(sample size). So 1000 is plenty. But it is reasonable to worry whether the sample was actually random, or biased in some way (e.g. sampling only Gen Z-ers that use a certain website) -- that would lead to a systematic error. However, for a systematic error a bigger sample size would not help much.
- Myrmornis 4y agoI like your comment! However, isn’t it plausible that the underlying population (or the distribution resulting from the ascertainment process) is structured in such a way that N=1000 might well have over-representation of certain subgroups, whereas N=10000 would be likely to have equal representation of subgroups?
- epistasis 4y agoIf your sampling strategy gets a biased group of 1000, it's almost certainly going to be biased in the same way as you pick 10,000. 1000 is plenty of individuals, the real concern is how they were chosen.
- danjoredd 4y agoThat was my concern. How many states did they test in the US? How many people from each state? LA and NYC are famous for being the cities every theater kid wants to live in, so if too many responses come from them, it seems obvious a huge portion want to be influencers. What about the midwest and south? Do they get the same results there?