So I got to watch the rest of it today, I didn't actually miss much, but she mentioned being a polling skeptic, so I wanted to explain polling once again because I find "polling skepticism" to be cringe and ignorant.
The purpose of public polling is to get an idea of what the public is thinking. Polliing isnt perfect. It relies on getting a representative sample of people. The smaller the sample, the less accurate it's going to be. Ideally polls are around 1000 people, and even then they can be off by 3 points in either direction.
For my own predictions, understanding many polls arent even that accurate, I go by a 4 point margin of error. And this is the thing people don't understand. Biden is down 4.5% in pennsylvania, which is the tipping point state in my current forecast. But, because 4 point margin of error applies to both Biden and trump, results can be off by up to 8 points and still be within the 95% confidence interval. Basically with 95% confidence, we can say thhat 95% of the time, the polls will be within 3-4 points of their stated values. Theres still a 5% chance of greater deviation, it happens, but its supposed to be relatively rare given methods are used.
In a sense, my own forecasts just show the median prediction. Like, let's bring up my most recent forecast from yesterday again.
Again, this is the median. PA is at a tipping point of +4.5% Trump, but that could, in actuality, range from being +3.5% Biden to +12.5% trump. 95% of all expected scenarios should occur within that envelope. This is why I gave Biden a 13% chance. He still can win 13% of scenarios here. Trump isnt guaranteed to win. He's just heavily favored to win. Again, keep in mind past discussions we've had on bell curves. 68% of all results should be within 4 points, 95% within 8 points. 99.7% within 12. Biden's chances aren't impossible, they're just low. Once you get beyond 4 points youre out on the tail end of a bell curve where your chances quickly dwindle. And that's kind of where Biden is. 1-3 point shift? Not particularly hard. 4+? Can be hard, possible, but hard.
"Poll skeptics" too often look at the results being off by 2-3 points like 2016 or 4 points like in 2020 and be like SEE? POLLS DON'T TELL YOU ANYTHING THEY GOT IT WRONG. Not really. The result was within the expected range of errors. It happens more often than you think. When people start talking like this, I start feeling like they dont understand polling and probability.
Now, AOC mentioned in her own races polls had her underestimated by double digits, what would cause that?
Uh, have you ever looked into polling for congressional races? It's not polled very often, most districts have zero polling, or at most 1-2 polls. And yeah the room for error is much higher. I think i actually tried to look into how accurate polling is and i think i found some sort of new york times article or something that basically said, 4 point MOE is about right for presidential races, senate races are more like a 6% MOE, and congressional house races are probably even worse as they're polled even more sparsely. So error might be higher than MOEs state. We knoticed this in my own track record, where anything within 4 points has a 72% chance give or take, and anything in the 4-8 range has a 4% chance. It was weird, there wasnt much of a bell curve as much as it was a top hat. That could be due to a relatively low sample size, but either way, yeah. Maybe in some instances the stated probabilities dont line up with the actual probabilities. In which case we should be even more concerned because Biden is on "hat rim" territory, and not the bowl part at this point.
But yeah. I just wanted to have his discussion on polling again because "polling skeptics" irritate the crap out of me and seriously need to take a stats and research methods class. I mean, yes, polling isnt always 100% dead on, we know that, it never claims to be. It claims to be right within a certain margin of error, and functionally predicts a range of outcomes with the official prediction being the functional median. Just because actual outcome may vary doesn't mean it's wrong. As I said, 1 in 8 outcomes even now will have Joe Biden winning. So he has a chance. Trump winning isnt "statistically significant" (outside of the margin of error). It's just the favored outcome.
And yeah. I don't want to beat a dead horse more than I already have.
No comments:
Post a Comment