er size of the sample ( so long as it is representative - that is not drawn from a vets reunion or a AMA poss-up or the nat rifle association meeting)
is discussed here
https://statisticsbyjim.com/hypothesis-testing/sample-size-power-analysis/
and 2000 will detect on expectation a 49-51 split or larger
Yeah technical, but it comes up regularly.
To know reasonably certainly in a pop of 330m people think you need a sample of millions - you dont.
a million sample will detect a 49.999 - 50.001 % split ( but they cant all be deep south afro-americans).
This was why the NHS was in the forefront of Covid research - the NHS cd dial up 100 000 pts ( assorted) no probs, and no other health ( insurance) sector could.