Genuine questions, not canvassing, if anyone knows how polls work, i have a bunch of questions, so thx in advance.
Everyone arguing about which poll is correct, have any of you ever been asked in a GE poll for your opinion? I certainly never have nor any of my close family or network, so wondered how data is gathered.
If we assume it's not general public in the street, do we know if one group of people from any side are more or less honest with sharing their intentions?
If it's always the same people we gauge from who send in data, how likely are they to change at all and flip-flop more than once in six weeks and are they more or less likely to vote a particular bias on the basis they are already engaged and interested in politics versus the large swathes who either aren't fussed, vote because they should or don't bother at all.
Do we know that they represent a particular demographic fairly by age, race, sex, region or ant other factor.
Imo would need to be min 10k people on each to be diverse enough to draw conclusions at which level you'd assume by now I'd come across someone's who had been asked.
These are fairly common misconceptions.
How data is gathered varies by pollster. Many used to use phone (missed out young people) or internet (misses older people) only, many use a mix.
Overall though the aim is to get a representative sample of people. Some make some female. Some young some old. Some rich some poor whatever variables analysis shows matter.
Your thought about needing 10k is a common one but actually the maths works out that at around 1k people the sample doesn’t get better results.
You can play with sample sizes here
This free sample size calculator determines the sample size required to meet a given set of constraints. Also, learn more about population standard deviation.
www.calculator.net
and see what sort of confidence level you get but 1k will get you 95% confident basically. Now that’s not any 1k people. Thats a representative sample. So if there’s say 51% female voters then you need 510 females in the sample. Same for location, age, previous voting, etc.
Now obviously we could go “well they’ve got fewer left handed people” or “they’ve got fewer people called Rob” but we only care about variables that have been shown to correlate with voting intention.
Various pollsters will apply various weighting based on what their data says. For example “Labour voters stay home more often” or “Tory voters tend to come back late in the game” or whatever. There’s lots of well known issues with just asking people their opinions. People lie. People forget (it’s a well known phenomenon that people think they voted for the winner previously even if there’s evidence they didn’t). So there’s always some level of reweighting for this stuff.
Sometimes though not very often there might be a mistake in the samplingnor reweighting that leads to changes going forward. But generally we know they’re doing it right because polling is consistently very accurate.
As for your other questions. Generally it’s not the same people no. Though some like YouGov use a fixed group that they sample from (you can be part of a YouGov poll simply by signing up to their site).
The polls mostly say the same thing so it’s not really a matter of who is correct. What’s different as I posted in that graphic is that some ask slightly different questions or work with don’t knows differently.
So most will exclude don’t knows or assume they’ll vote with the rest. Some will ask a “squeeze question” which is basically “gun to your head you have to pick who do you pick”, JLL the one being mentioned right now not only do that but also try and account for known effects like the incumbents vote getting better the closer you get to election.
On the whole though the statistical basis for polling is very very sound. Lots of money is riding on the outcomes so there’s a big incentive to get it right. And by and large they do. Even famous “polling errors” (Trump, Brexit, etc) actually weren’t if you look at the polls. They were either margin of error, or actually right towards the end. There’s been some polling issues in the past but as an industry they mostly have fixed those.
Obviously if something happens thatsnwildly out of whack with what’s happened previously then some of the assumptions will be wrong. But generally these are quite small impacts.
Tl;dr: 1k people, if properly selected to represent the population, is enough to get an accurate answer with 95% certainty. And this is one of the most well tested pieces of science out there.
