This presumably causes problems for pollsters that weight on recalled vote and don't have the earlier panel data. I wonder if this phenomenon helps explain some of the swings in pollster accuracy by the newer polling firms that seemed to favor Trump heavily.
I wonder if there's any information on this from the end of Bush's second term. He was so unpopular he pretty much did not campaign for McCain. After the Afghanistan and Iraq wars and then the financial crisis, I don't think I can recall a point when a party was more set on distancing itself from a president. The post-Biden period has been dramatic but it isn't the same.
It’s interesting that there’s a noticeable drift even among people who in theory know they’re on the record on this question from the first survey. I wonder if there’s more change in polls where you can’t check against the baseline, and the respondent knows this.
It probably depends on whether this is intentional dishonesty or misremembering. There have been some primary elections where I was debating a couple candidates for a long time and can’t remember which I actually came down in favor of on the day I cast my ballot. I could imagine that my current feeling about those candidates would shift my memory more in those cases.
If I get a survey once in a while, I may not have any more memory of what I said in the survey than of what I actually did.
It's hard to imagine many people are unsure of who voted for in a presidential race though. It's not a local judgeship. You'd have to be REALLY low information to forget that one.
You don't have to be low information - you just have to have been conflicted in the weeks leading up to the vote. If there were some days you were leaning towards Trump and some days you were leaning towards Harris, it's very easy to think that you might also have changed your mind a few times in the days after the vote, such that a few weeks later you don't really remember which you were leaning towards on the day you actually cast the vote. If the issues that were tilting you in one of the directions are extremely salient today, this might make it really easy to remember the times you were leaning that direction, and easy to forget if you were thinking something different on the day you actually cast the vote.
In the particular case of having gone back and forth between Trump and Harris, it's hard for either of us to really understand what precisely the person could have been thinking, but the basic structure of their thinking seems the same as a case we can comprehend (like myself seesawing between Buttigieg and Warren in the days leading up to Super Tuesday in March 2020, and almost forgetting that at the last minute I actually voted for Sanders because I was so unhappy with Biden as a candidate).
That's what I'm saying, it's hard to imagine anyone flipping back and forth between Trump and Harris and not remembering which way they went. It's just a choice between VERY different options.
Huh, I would think that anyone flipping back and forth in *any* decision is moderately likely to mis-remember which way they actually went on a particular date. Even if the options are very different. And especially if they didn't think that their particular decision was causally efficacious in determining the outcome.
Is it possible these people are just unreliable from the start? Could the falsehood be that they voted? It says at the beginning a poll of registered voters, but that itself isn't proof they actually did.
I think we can at least hope that the vote recalled within a week or two of casting it is at last _more_ accurate than how it's recalled years later. I do agree with you that this adds a layer of uncertainty. You need gaussian priors, or something...
I don't think that people forgot which presidential candidate they voted for a year later. I also doubt that a bunch of people are so ashamed that they can't bring themselves to admit who they voted for to an anonymous online survey. So, I don't think that we can and should assume that their answer a year ago was more accurate. Polling can be helpful especially for big swings in opinion. But I think it's a mistake to think that anomalies like this indicates anything other than the fact the people filling them out probably aren't taking these things that seriously. Some people just make stuff up and they will probably forget what they made up when you survey them again.
I'm sure there's some of that too, but the "preference reconciliation" effect being discussed here seems consistent with other things you'd expect about human psychology, and I expect there is some truth to it. The question is, how much is signal, and how much is noise? Having a large sample that you can go back to for interviews, where you can weight on their votes as reported at the time of the election, seems like it will probably create marginally-more-accurate results than just weighting reports from today back to how the demographics looked at the time of the election.
This should call into question how sure people were that getting more people out to vote would not have changed the outcome, no? You can't use "who would you have voted for" as a strong indication that voters that stayed home were more likely to have voted for trump, if we know that it is heavily swayed by winners bias.
Granted, we can't claim otherwise, either. I am just pushing against the narrative that people know how people that stayed home didn't actually impact the outcome.
This presumably causes problems for pollsters that weight on recalled vote and don't have the earlier panel data. I wonder if this phenomenon helps explain some of the swings in pollster accuracy by the newer polling firms that seemed to favor Trump heavily.
I've always wondered about what Trump voters will have to say in 10-20 years. I guess we know now.
I wonder if there's any information on this from the end of Bush's second term. He was so unpopular he pretty much did not campaign for McCain. After the Afghanistan and Iraq wars and then the financial crisis, I don't think I can recall a point when a party was more set on distancing itself from a president. The post-Biden period has been dramatic but it isn't the same.
It’s interesting that there’s a noticeable drift even among people who in theory know they’re on the record on this question from the first survey. I wonder if there’s more change in polls where you can’t check against the baseline, and the respondent knows this.
It probably depends on whether this is intentional dishonesty or misremembering. There have been some primary elections where I was debating a couple candidates for a long time and can’t remember which I actually came down in favor of on the day I cast my ballot. I could imagine that my current feeling about those candidates would shift my memory more in those cases.
If I get a survey once in a while, I may not have any more memory of what I said in the survey than of what I actually did.
It's hard to imagine many people are unsure of who voted for in a presidential race though. It's not a local judgeship. You'd have to be REALLY low information to forget that one.
You don't have to be low information - you just have to have been conflicted in the weeks leading up to the vote. If there were some days you were leaning towards Trump and some days you were leaning towards Harris, it's very easy to think that you might also have changed your mind a few times in the days after the vote, such that a few weeks later you don't really remember which you were leaning towards on the day you actually cast the vote. If the issues that were tilting you in one of the directions are extremely salient today, this might make it really easy to remember the times you were leaning that direction, and easy to forget if you were thinking something different on the day you actually cast the vote.
In the particular case of having gone back and forth between Trump and Harris, it's hard for either of us to really understand what precisely the person could have been thinking, but the basic structure of their thinking seems the same as a case we can comprehend (like myself seesawing between Buttigieg and Warren in the days leading up to Super Tuesday in March 2020, and almost forgetting that at the last minute I actually voted for Sanders because I was so unhappy with Biden as a candidate).
That's what I'm saying, it's hard to imagine anyone flipping back and forth between Trump and Harris and not remembering which way they went. It's just a choice between VERY different options.
Huh, I would think that anyone flipping back and forth in *any* decision is moderately likely to mis-remember which way they actually went on a particular date. Even if the options are very different. And especially if they didn't think that their particular decision was causally efficacious in determining the outcome.
Is it possible these people are just unreliable from the start? Could the falsehood be that they voted? It says at the beginning a poll of registered voters, but that itself isn't proof they actually did.
I think we can at least hope that the vote recalled within a week or two of casting it is at last _more_ accurate than how it's recalled years later. I do agree with you that this adds a layer of uncertainty. You need gaussian priors, or something...
I don't think that people forgot which presidential candidate they voted for a year later. I also doubt that a bunch of people are so ashamed that they can't bring themselves to admit who they voted for to an anonymous online survey. So, I don't think that we can and should assume that their answer a year ago was more accurate. Polling can be helpful especially for big swings in opinion. But I think it's a mistake to think that anomalies like this indicates anything other than the fact the people filling them out probably aren't taking these things that seriously. Some people just make stuff up and they will probably forget what they made up when you survey them again.
I'm sure there's some of that too, but the "preference reconciliation" effect being discussed here seems consistent with other things you'd expect about human psychology, and I expect there is some truth to it. The question is, how much is signal, and how much is noise? Having a large sample that you can go back to for interviews, where you can weight on their votes as reported at the time of the election, seems like it will probably create marginally-more-accurate results than just weighting reports from today back to how the demographics looked at the time of the election.
This should call into question how sure people were that getting more people out to vote would not have changed the outcome, no? You can't use "who would you have voted for" as a strong indication that voters that stayed home were more likely to have voted for trump, if we know that it is heavily swayed by winners bias.
Granted, we can't claim otherwise, either. I am just pushing against the narrative that people know how people that stayed home didn't actually impact the outcome.