.
Math decides, assuming a representative sample. The larger your sample becomes the better, as long as it is representative. The margin of error is not nearly as important as the confidence interval quoted, that is when they say 19 times out of 20 or so. It is expected the results will not be accurate in that 1 time out of 20. As I keep saying this relies on a representative sample who answer honestly which is very hard to get and it is why outcomes often don't match polls.
Thank you for your helpful reply ITpro. I appreciate it. It other words it's kinda complicated. Happy Friday! :)



This is an extremely helpful reply Rpf for which I thank you. I get your point. That would be assuming people tell the truth. Ever hear about "the Bradley" effect? There was an African American running for political office and polling showed him winning. But he lost. That was because people did not want to admit they wouldn't vote for someone of color so they lied. I suppose the lie could go in the opposite direction aa well. Some folks say they won't vote for Trump but they might just be afraid to say so and when they vote he will get those votes. So it's a crapshoot isn't it? Just what we need. More uncertainty! :(



