Eli5 – How are news channels able to project a winner in elections with a very small percentage of votes counted?
In: 4
Election polls are done from long out until right before the election (and then actually after people do vote too!). They already know the outcome of most elections before the election ever takes place for the most part anyways, and the votes coming in just confirms what they already knew or they make slight changes to their predictions, but generally they already know pretty close to what to expect and what the outcome is
They know how they expect people to vote, they know how people and areas have voted in the past, and they expect those all to be pretty similar again to how they are going to vote now (along with polls to confirm, yup its happening).
Using some statistics and info they expect to be true, they can generally get a pretty high confidence in the outcome without a large amount of votes actually counted, as they see a ew pieces falling into place, they can see where the rest of the pieces are highly likely to fall as well.
As a side note, generally most elections aren’t all that close either. Most elections strongly favor one candidate winning.
There’s a lot of granular statistical analysis that goes on behind the scenes that can give a better vision than just “percentage of votes counted.” For example, while 5% of a state’s votes might be recorded, 40% of that could come from a typically conservative district and show a sharp swing towards a progressive party, indicating other seats will be even more demonstrative.
Also, few respectable outlets will put their cards down and “call” a result until a majority of votes are counted. They make projections based on probability and refine it as it goes on, mostly because most outcomes preserve a status quo and so that assumption is a good place to start.
They look at a bunch of different factors. County election officials post regular updates on vote totals and, importantly, where those votes are from. They also have access to voter registration information like party affiliation and voter location. Combining that data can give you a pretty good idea of the likelihood of a particular outcome.
The most basic thing is statistical analysis. I don’t need to flip a coin until it wears out to know if it’s a fair coin or not: about 100 flips will give me a good guess if it’s close to fair – and if it’s obviously not fair, that will show up. In the same way, it only takes less than 1000 people to get a good idea of how an election will go if you randomly sample people.
However, you can do better by NOT randomly sampling people, and instead intelligently sample people. This takes a lot more work, but the general idea is that you want to ask people who represent the people who are going to vote. I can’t explain this well because I haven’t looked into it as much, but there are ways to do this.
The last thing you can do is look for trends. If you’re seeing a lot more women vote, that’s going to mean something in close elections where one candidate has a lot more women voting for them. If you see moderate voters voting for one candidate over another, that’s going to mean something – and it’s going to mean more if you see moderate voters in multiple elections all leaning the same way. These trends can also help guide your intelligent sampling: maybe you need to ask more women how they’re voting if more women are voting.
And there’s a lot of math behind all of this – some of which is still new, and not really well understood (even by the people doing it – AI generates results that are like that some times).
Latest Answers