I bet you really hate weather forecasts.
Shouldn’t the question be, after last time, why would anyone have listened to them THIS time?
And yet, it seemed everyone jumped on board, as soon as pollsters started spewing numbers again.
Sure, there was a lot of, ‘well they didn’t get it right last time, did they?’, but none of it changed what still became an endless news cycle of margins of this or that, how they’re shifting, what they MIGHT mean, etc,etc.
I’m not convinced it won’t be just more of the same next time, to be honest.
Here’s Berezow’s article on that for 2016:
Does the margin of error in the polls reflect the low response rate of people taking the polls? From stats, I think the margin of error is reflective of a perfect random sampling indicating the total population, but polling is not a perfect random sample. If 100% of people contacted answered the polls, then that might be the case, but that’s not what we see. If only 10% of people respond, then it seems like the margin of error should be much greater.
Weather forecasts are reasonably repeatable - you can cross reference the forecaster’s ‘chance of rain’ percentage with what days it did and didn’t rain in general or when conditions are similar to what he predicted. It’s quite possible to falsify weather predictions, but political predictions don’t work that way. You only get a particular set of voters, world events, issues, and candidates one time, and you can’t even gather a lot of the data you’d like to have. It’s possible to take something like ‘when there’s a high pressure front moving here and 40% cloud cover on radar here, there’s a 60% chance of rain’ and repeat it on 10 or 100 different days to see if you’re getting something around 6 or 60 days of rain. With the election, you only have the one shot, so it’s impossible to say if the percentage is right or approximately right (and certainly to discern 70.2% from 70.3%).
Weather forecasts have a practical purpose. Do I need to take an umbrella? Yes. Maybe. No. The negative effects of a weather forecast are small. It also makes sense. There are 360-plus days in a year, and many years in my lifetime, and it’s sensible to aggregate probabilities based on such extensive data. “Biden wins 60 percent of the time” is gibberish, because there only one damn election in question.
As Chief Haruspex at Ntrailz, where we bring a different and more hepatoscopy-driven perspective to elections than many news organizations, I’m no stranger to misguided outrage from the unenlightened masses. I understand why, to the untrained, ignorant, stupid, stupid, stupid eye of the layman, Ntrailz’s “2020 General Election Prophecy: Complete Results” might have looked like some kind of prophecy about the results of the 2020 general election. But when I wrote “Joe Biden will be declared President of the United States in a landslide before the sun sets over Mauna Loa on Election Day, which is to say before 5:54 P.M. Hawaii Standard Time, Nov. 3, 2020,” I was talking about a range of probabilities, not making a definite prediction. When I went on MSNBC to announce that Ntrailz’s Sacred Rooster had pecked the grains covering the letters “β,” “Ι,” and “Δ,” and excitedly told Rachel Maddow that this “absolutely guarantees a Biden blowout,” I assumed anyone with a grade school background in alectryomancy would know I was speaking figuratively. And when I tweeted, “JOE BIDEN CANNOT LOSE! 50 STATE SWEEP! ZERO VOTES FOR TRUMP NATIONWIDE! BET YOUR HOUSE ON IT!!!” I was doing cocaine. Once you filter out the noise and misinformation, the election results actually validate important features of our model. I want to explain in a little more detail, so that you can all see exactly why you owe me an apology.
Saying that polls are completely useless, as some people seem to be doing it, is vastly overstating things. Polls just have a much bigger margin for error than people assume. But once that’s appreciated, they’re very useful. If my candidate is running 5% ahead in the polls I’m a lot happier than if he’s running 5% behind in the polls, despite my fully appreciating that the polls could be off, and I suspect this is true of the vast majority of poll-denigrators.
And the problem is that they’re extremely useful to certain people. Not me, because I’m a lay person. But if you’re a big time contributor or a political organization, it’s of crucial importance that you donate your resources to races where they can have an impact. So you’re going to pay a lot of attention to polls, imperfect as they are. So polls are inevitable, and once they are going to exist, we may as well have the most reputable organizations as possible doing them.
We might not agree very often, but this reflects my thoughts exactly.
Because Trumpers seem to be the polar opposite of shy. Oversized TRUMP flags flying from monster trucks, wandering the streets revving their engines and wasting gas precisely so people can see them waving those flags. Multiple lawn signs and Trump flags on their homes. MAGA hats. Trump phone cases. And I’m seeing this in the northern suburbs of Chicago (not quite the uber-progressive North Shore but close) that are fairly moderate-to-progressive Democratic.
Sure, but that’s only the ones you can see.
That’s like someone saying “All black people are good at basketball; I see them playing on the courts all day long.” There are many blacks who aren’t or can’t play.
Maybe, but fraud of the magnitude would be easier to detect and you don’t think there are a few Democrats poking around? How could you pull it off?
Plus you have to consider that roughly the same thing happened last time. Getting away with it twice would be so much harder.
Ditto. I see all of that here in Connecticut as well. I would like to have put up a Biden sign or bumper sticker, but don’t want my house or car vandalized by some crazy Trumper.
I work with a few Trumpers, and they have Trump paraphernalia all over their cubicles. One has had a Trump 2020 sign in his cube since 2016. None of these folks are the least bit shy about their support.
(On the other hand, I do have one close friend who confessed that he supported Trump a couple of years ago only after getting a few drinks into him. I have to admit it changed my opinion of him.)
So you deliberately started this thread before all the results are in, when record levels of early voting and mail voting have caused multiple pundits, election officials, etc. to warn that early results may well be superseded as more votes are counted? 
Why? So you could say
Considering the more recent numbers from Wisconsin, Michigan, and Nevada, and Trump’s dwindling lead in Pennsylvania as more votes are counted, it seems a bit disingenuous to figuratively stick a pin in at such an early point and declare that “it looks like a Trump win.”
Wait, you’re going to chastise me for this after your post with like ten links? After I responded with a direct quote? I’m done with you.
This is an absurdly dumb take for a publication. Obviously, you can’t ban election forecasts.
The argument to ignore them is also incredibly weak. The fundamental problem is that making predictions about the future is really hard. The fact that predictions are imperfect doesn’t mean that they are without value (or, as some people seem to think, actually of negative value).
I had a thought about how to test the “shy Trump voter” hypothesis… which is that it should be much stronger in left-leaning areas than right-leaning. There’s presumably much more incentive for someone living in a big blue urban city to hide their Trump support than someone living in a red rural county. So if we had fairly fine-grained opinion polls, and could compare them to actual voting outcomes, and we see that the polling is more accurate in red areas than blue areas (specifically, more inaccurate-towards-blue in blue areas), then that would be evidence that shy-Trump-voter is a real thing.
Doing a little deeper inspection.
In Illinois the 538 polls indicated a 55% to 39% Biden advantage. I’m left to wonder where the extra 6% went, but whatever.
Looking at the results, the numbers are pretty close. Biden’s numbers are basically dead on and Trump over-performed by about 4 points, clearly grabbing almost all of that missing 6%.
So, what does this mean? Certainly Illinois is not a battleground and there weren’t a ton of polls conducted in the state, at least not that are in the 538 composite, so this isn’t a ton of data. The results for Biden tracked very very close to the actual result, so that’s good. I don’t fully understand the pollster’s methodology there, so I don’t know if that missing 6% are undecidedes or if that’s reflective of a margin of error or what. Also, I don’t see margin of error posted so nothing to comment on there.
There’s one of two assessments I can make. The polls were accurate in Illinois, or the polls were reasonably close but there is evidence of around a 4% discrepancy that could indicate a “shy” Trump voter.
Repeating the exercise in another non-battleground state.
In Tennessee, they have essentially the opposite polling result. Trump showing a 55% with Biden at 41%. Here there’s a smaller slice of missing voters, just about 3.5%. Compared to Illinois there seems to be even less polling data to go off of.
The results are quite a lot different. Trump reached close to 61% and Biden slipped to just 37%. This is about a 6% bump for Trump and about a 4% slide for Biden. Again, I think there’s a hint that of that missing 3-4% almost all of it swung to Trump, but that doesn’t account for all of it. I supposed these polls are so far and few between that it’s hard to say if this is an error or if it’s just incomplete information.
Let’s pick a random battleground.
Here in Wisconsin we have lots and lots of polls to work with. Biden 52% and Trump 44%. Once again there’s a missing 4% or so.
The results show a pretty different story. Biden at 49.4% and Trump at 48.8%. As the last few votes trickle in I think there’s a reasonable chance the widens ever so slightly, maybe reaching a difference closer to a full 1%. Certainly it will not end up close to that 4-8 point advantage. Once again the missing 4% almost entirely broke for Trump. Yes, there’s close to 2% tied up in 3rd party candidates, but this is pretty consistent across all states and the polls do not show them. This also shows Biden shedding between 2-3% of his lead to Trump. Even the most bearish of highly graded polls had Biden at 47% and Trump at 43% leaving ~10% unaccounted for.
Clearly this 538 composite poll was wrong by a lot. I absolute terms its off by at least 7 points. The non-composites ranged from 4% to 13% after dropping the outliers. No issue with sample size here, there were 4-5 new polls taken every day in the final weeks.
Spot checking a few other states: New York polled at close to a 30% spread, but the actuals are closer to a 18% spread…but there’s a mountain of mail in votes coming in so this may widen. Iowa was polling neck and neck with the latest composite showing Trump +1.3%. The actuals have Trump +8 but this will probably narrow a little with mail-ins. California polled with Biden +30 and the actuals so far are showing Biden doing even better with +33 and likely to widen. Kentucky is Trump +16 in the polls and Trump +26 in the actuals.
Based on these examples, I draw a few conclusions. The bluer the state, the more accurate the poll…maybe. There is definitely an under-representation of likely Trump voters in every poll, somewhere in the ballpark of 4%. The redder the state the more likely the polls are to be wildly under reporting the Trump vote. Maybe the polls, in general, tend to artificially moderate the numbers. Very red precincts poll less red. Very blue precincts polls less blue. This last point could explain the overall polling error towards Trump as there tend to be far more deep, deep red districts in every state.
Here’s a question: did the polls predict the very high turnout in this election? Because if they didn’t, then maybe the problem isn’t with their political evaluation, but rather their likely voter evaluation.
Intuitively, that seems like a much harder thing to measure in the first place. And in particular, it seems like you’d encounter a far larger number of undecideds when it comes to actually voting than ones that can’t pick D vs. R. It may also be that the very nature of enthusiasm is different between likely D vs. likely R voters (say, if fear is a better motivator).
I can’t fathom how that would have any meaningful impact based on the way polls are conducted. Similar to how people hanging up on pollsters has absolutely no impact on the projection. They call 5000 people and if only 1000 answer the phone, then the percentage is based on that 1000…not the original 5000. Same with likely voters. If a person is not a likely voter, they don’t get polled at all. “Non-likely voters” are no different than every other person they don’t call, and when they post a percentage that ratio gets applied to the entire state.
They certainly get polled on questions relating to whether they’re a likely voter or not. If those questions are somehow biased (in the way that just calling landlines can be biased, etc.), then that will bias the results unless correct for in some other way. So it may be that the political questions are fine, and giving a good result, but they are rejecting too many Rs as not-likely when in fact they are.