That poll putting Trump ahead of Biden in all the major battleground states sure looks terrifying, but there’s never been an election more clouded by the unknown than this one.
A week after Halloween and the scary monsters are still abroad in the land.
Scary polls!
Scary plans!
Boogedy, boogedy!
It was a great weekend for intellectual doomscrolling, to say nothing of galloping paranoia. First, The New York Times comes out with a poll that shows the president is trailing Fulton County (Ga.) Inmate No. PO1135809 in all the major battleground states.
The polls didn’t predict the 2016 win by Trump. They didn’t predict the 2018 blue wave accurately. They didn’t predict highest voter turnout in 100 years in 2020. They failed to predict a red wave that never materialized in 2022. BUT, I’ve got a good feeling about polling in 2024!
"When looked at in historical context, what stands out isn’t that polling in 2016 was unusually poor, but that polling of the 2004, 2008 and 2012 presidential races was uncannily good — in a way that may have given people false expectations about how accurate polling has been all along.
The other factor is that the error was more consequential in 2016 than it was in past years, since Trump narrowly won a lot of states where Clinton was narrowly ahead in the polls."
https://fivethirtyeight.com/features/the-polls-are-all-right/
I wish my job allowed the level of reliability that Political Experts, ‘Economists’ and Meteorologists have.
Meteorology is really hard, but comparing it to politics and economics is false equivalence. Meteorology is governed by well proven mathematical models, and we can use them to make predictions. The problem is that the earth is really big, so we just don’t have computers powerful enough to simulate it finely enough. Add to that it’s a chaotic system and it becomes difficult to predict accurately very far into the future. Weather predictions have actually improved dramatically the last few decades, and I expect they will continue to do so along with advances in computing. Economics and politics may as well be random guessing, but is often worse than random guessing, because we have no reliable proven model for human behavior.
Anytime people get mad at the metrologist for being wrong, I remind them that they’re LITERALLY PREDICTING THE FUTURE.
As someone trained in statistics, I will say you have nailed this. The only ‘poll’ that has any hope of being accurate is counting the votes the day after. Even if we had good models for the complex intricacies of human behavior with respect to voting (I will correct that we do have good models of human behavior, Game Theory and Network Theory have gone a long way towards providing workable models for things) there is no way to guarantee clean input data. Garbage in, conservatives out. Pollster bias, population bias, selection bias, social pressures causing disingenuous responses. You can’t get away from any of it. Phone polling requires that people answer and actually participate, which eliminates swaths of personality types which skews your data. In-person polls have to be conducted in person, so the location choice skews the types of people who are likely to be present at the time you are polling. Even focus groups are inherently flawed as a polling methodology, but they are as close to clean as one can get. You still only obtain the opinions of people who have the time during the day to go somewhere for hours. So it is predominantly young college students between classes, stay-at-home parents while the kids are in school, people who are unemployed, or retirees looking for something to do. If it is not one of those, it is likely someone who is doing the focus groups to make ends meet and the repeated use of the same people for different polls often leads to other forms of data contamination.
Bottom line, don’t trust polls. If you aren’t doing them, likely the demographics that you represent are not doing them either.
Another issue that affects how accurately weather models are seen is a difference in what they output and what is reported. The output of the models is a series of 2D maps of various variables and how they change over time and space. The models will predict that conditions will form for cloud formation and will lead to precipitation at a certain point. They are pretty good at predicting that part. Where it starts to get less accurate is determining where those things will happen and when with specificity. They’ll be pretty sure that there will be rain from this particular system, but it might move north of city x, go right over it, or go south of it.
So that 40% chance of rain is actually “99% chance it rains, but 39.5% chance it rains here and 59.5% chance it doesn’t rain here but somewhere else nearby instead”.
My appreciation for what they do increased after I started using windy.com, which gives the map of predictions over time instead of “here’s what will happen in this city”.
Oh and weather patterns can be smaller than cities, too. That 40% chance of rain could even mean “40% of the city will be rained on”. On cloudy days, you can often look around and see rain in the distance in various directions around you, sometimes it passes over you sometimes it doesn’t.
Right, that’s exactly what it means, by my understanding. A weatherman can’t predict the chances that a particular individual will experience precipitation, but that’s what the average person immediately thinks when they look at a weather forecast.
What a weatherman can do though, is predict how much of a particular area may experience precipitation, based on measurable things like cloud pattern shape and size, wind speed and direction, geography, etc. Once you realize that, it’s actually kind of intuitive that precipitation chances are reported as “percent of an area that will experience precipitation”, and not “percent chance that I will experience precipitation”.