Can We Trust Midterm Polling Data?

Polling underestimated Trump's support in both 2016 and 2020, and questions have arisen over whether they're a reliable metric. So with the midterms coming up, we discuss the factors that influence the polls and whether we can trust them this time round.

TLDR is all about getting you up to date with the news of today, without bias and without filter. We aim to give you the information you need, quickly and simply so that you can make your own decision.

TLDR is a completely independent & privately owned media company that's not afraid to tackle the issues we think are most important. The channel is run by just a small group of young people, with us hoping to pass on our enthusiasm for politics to other young people. We are primarily fan sourced with most of our funding coming from donations and ad revenue. No shady corporations, no one telling us what to say. We can't wait to grow further and help more people get informed. Help support us by subscribing, following, and backing us on Patreon. Thanks!

////////////////////////

[ChildofWar]

As a polish person, I am confused as to why you wouldn't trust us.

Also drinking game: drink or take a shot every time he says the word "pole/poles/polling"

[Bigdog]

I think that one thing most Americans can agree however is that we're tired of the binary one party vs. the other.

[yt.personal.identification]

Polls are performed on "expected voters".
After Roe overturning, there will be many many "unexpected voters" not in the polls.

[blaarfengaar]

2016 polls were actually pretty accurate, the only real exceptions were in like 2 or 3 states like Michigan, Wisconsin, and Pennsylvania iirc, but the national level polls were well within the margin of error

[nickmacarius]

The United States has very inaccurate polling. Not only because as was mentioned some pollsters have a political agenda to get the results they want. But there is also another factor. I use to live in Iowa, and pollsters always "predict" that our elections are competitive, but in reality it's a very Republican state. I remember in 2014 it was predicted that race for governor was neck-&-neck, but Terry Brandstad (R) won with over 60%. Now in Iowa we are first in the nation during Presidential elections, which results in presidential campaigns being active in the state of Iowa for nearly a year & a half. Needless to say that all the campaigning, the mail advertisements, door-to-door activists, and the pollsters calling becomes very annoying. Most of us become apathetic about it after a few months, so when pollsters call we just ignore them - that goes especially for the small town/rural areas, which are mostly Republican.

[djp]

I've been voting for 20 years. Never received any poll questions. They must be polling old farts only with landlines.

[Rofl]

I don't know anyone who has ever been polled even once (despite me knowing many people who have been registered voters for decades), I'm not implying a conspiracy: I'm sure the pollsters are "trying" to reach as "random" sample as possible, but I'm not so sure they are doing a good job.

[DrustZapat]

Another question (and in my opinion a better one) is which polls historically are most often correct. Of course bias is an important factor, especially non-response bias, but if I learned anything from Superforecasters by Philip Tetlock and Hate, Inc. by Matt Taibbi it's this: we put way too much faith in people who get a lot of air time that are often wrong in their predictions.

So who predicts the best?

[docholiday]

Polls are no longer accurate in a close race..
Ignore them.
Very informative video

[JasonTaylor-poxc]

As we found out in 2016, polls are a tricky thing. Good honest people have no problems lying to those conducting a poll if they think they will be judged for their response. Not everyone who is polled will bother to vote, especially in mid-terms. Mid-terms are really about who is most upset since most folks won't bother - especially if they live in a "safe" state or district.

[frapino]

The problems is simple to solve:
1. You need larger pools of people to increase accuracy.
2. You also need a larger spread of the polling locations to catch diferences in different areas..
3. Pollsters sell more polls if they show a close race. Solution is to set fixed dates and amount of polls no matter the result.
4. Stop paying for inacurate polls.

But media wont do theese things because they get more viewers/readers when they report a close race or they they show different results from different pollsers. The real soloution to get this issue fixed is to first stop consuming polls on our side until they become more serious, as long as we keep consuming nonsense they will keep feeding us nonsense.

[XIIchiron]

You didn't mention the electoral college or other types of gerrymandering at all which I feel is a major oversight - even when polls are relatively accurate (like in 2016) they don't account for the fact that _some votes are worth much more than others._

[whoiamiamnot]

I'm American born and raised. Im retired Army. I'm a registered voter. I've never not voted. I'm not a registered member of any party.
I've never done any type of political poll. I've asked all of my immediate family. All of my friends including other military veterans. People that are registered party members. None of them have ever taken part in any political poll.
That's why I don't put much weight behind them. Who are the people doing these polls? How are they done? Over the phone, mailed in, online? It just seems like a flawed system to gather data.

[napoleonibonaparte]

Polling in general won’t give conclusive images in tight races.

[shja]

When a person or a group starts viewing the "other side" as demonic or completely irredeemable, the genuine mistrust among people is the next stage. Polling is just as useless, in predicting the winner, as asking ones neighbour who they are voting for. Polling has to take on a new role(imo this should've always been the case) which is to just show what the country's general mood is but not a definitive one.

[uddishbagri]

1:20 the easy part framing the question LOL, "Leading Questions - Yes Prime Minister" search this on youtube and see how important it is to ask the "Correct" questions

[reillygrant]

The videos on the US have come so far from the first few you guys produced. Very proud

[doctorspock]

Hi,

A correction when it comes to stratified sampling, which is the fondation behind this whole idea of having a representative panel instead of having a uniform measure over the entire population you put in your panel (you'd call that "randomly selecting people" even though, well, both strategies are random).

Consider the following toy problem. You have 2 coins, one is skewed with a rate of Heads that is 40%. The other one is skewed with a rate of Heads that is 60%. Your experiment is the following: you select one of the two coins (evenly, 50% proba each) and then you check out the proportion of Heads. In this case, you can think of those "Heads" as the number of people voting Trump and the first selection would be the college educated people or whatever.

So there are two Monte Carlo strategies to address this problem :
1 - you go through this experiment a number of time and you get a proportion of Heads and that's it, you're done. You can compute an estimate of your statistical error.
2 - you can select your sample so that you get the same number of skewed coins of each type, and only then compute the proportion of Heads. You can compute an estimate of your statistical error, and we can demonstrate it is a bit better than the former.

However, *both techniques work*.
So when you were claiming "College education was fudging the estimates and causing problems", well no not really. It is amongst the statistical error which is the only error publicly estimated in data poll (the other error types are taken into account somehow but don't participate to their error report). When you're saying that there could be another factor which would cause similar problems, again no. Say, if people start voting with "if they wear a belt" or whatever and we don't control for it with stratified sampling (sampling people with belt exactly at the proportion they exist in the voting population) we are just going to have a higher variance (which is anyway estimated) and therefore the error is gonna be showing at higher. But this is controlled.

Now, the part that explains that polls are wrong is more complicated, it has to do with model error and bias error. An example of bias error is the fact your sampling population is different from your panel population, something you explained very well with the example of trump voters being overly reluctant to answer polls. They correct for that, but don't estimate the error and frankly it's very rough how they do it. The model error has to do with either people lying in their response or changing their mind. Again, there's some correction but it's roughly done.

[CatherineKimport]

A big part of it is, like, the question they ask are terrible, and they take WAY too long to get through their questions. They always start out like, "oh, could you take this quick poll? It's only five minutes" and then like 30 minutes later they're still yacking away at you. And they phrase the questions so absurdly. I hate every pollster. I wish we could ban them.

[atlasb]

How do they ask people to participate in polls? Do they call citizens, door knock, or is it from those pop ups on websites?