I merely explained statistics and statistical sampling, not election polls, which are their own animal. Election polls are very complex. They are a combination of a preference poll and a turnout model (predicting who will actually show up to vote on a date certain). Moreover, in the US, they are exponentially more complex than that, as there are 51 distinct elections that determine the president. When there is a bust in the election polls, its usually the turnout model, and its usually over the youth vote. That part they haven't figured out. The preference polls have been pretty accurate.
Moreover, polling individual states is pretty difficult. Most of these polling companies are really good at the national polling, as evidenced by the fact that it has pretty spot on, certainly in 2016 and 2020. Its the state polling that has not been quite as precise.
The other really important thing in understanding election polls is momentum. The survey of these polls is conducted historically, reflecting how things were several days ago. Then, they are averaged with previous polls, some of one, two or three weeks ago, to give you what the purport to be a current view, but its not. It does not reflect how things of changed. For example, because Harris has momentum, her actual lead as of today is actually higher than is shown. If momentum should shift, you will see the opposite effect.
BTW... no one, to my knowledge, every has a Margin of Error (MoE) of less than 3%. An MoE of plus/minus 5 is more typical. If a candidate has a 5% or better lead in a particular state, they are very, very, very likely to win that state barring something that causes a momentum shift.
news.vanderbilt.edu
fortune.com
[/ISPOILER
debatepolitics.com
