This is an awesome breakdown. Thank you. To me it says that if we in the Harris camp keep the pedal to the metal all the way to election day then we will likely win, even if the numbers are close! Let’s keep pushing forward and reaching those voters folks. Write postcards, knock doors, phone bank, and bring three people with you to vote for Harris!
Every now and then one comes across a perspective that is such an original way of looking at things that it seems counterintuitive at first. But once someone explains it clearly, it seems so obviously valid that you wonder why anyone ever looked at it the old murky, unreliable way.
Thank you, Mike and Carl, for your crystal clear explanations of polling, which are making me feel like a genius (for a few hours, at least). 🙏
Another reason to include basic maths and statistics in high schools. Part of what fed MAGA's outrage in 2020 was the comparing 2016 raw numbers to 2020 numbers: "How could Trump loose if he out performed his previous outcome?"
Carl does a wonderful job of explaining for those who want to know.
An amazing amount of data and excellent explaining for those of us civilians. Bottom line, correct me if I'm wrong, is that I'd rather be Harris-Walz than Trump-Vance going into the home stretch. Many thanks for this great article.
Lots to nerd out on here, but one quick question. In your video a couple of posting ago, Mike, where you were talking about polls that included (but didn't split out) undecides. In these polls, do they assume undecides split evenly as shown in the bar charts? Do they even give you enough information to tell? (that's 2 questions... I know).
Polls (in the US, that is) traditionally report the numbers for the top two candidates: say, 49%-45%
The rest of those unstated numbers could be third-party or undecided.
Many analysts simplify that 49-45 poll to "up by 4" which creates some problems
Pollsters - in US cases - don't "assume" what undecideds will do - they just report how many their poll had.
Analysts, and forecasters, tend to assume (as a default assumption) that undecideds will split evenly. That can create some problems and it's a pretty bad assumption honestly.
As a forecaster, we can look at all the data from that sliver of undecided (demographics? Past vote? What issues matter? Will they even vote?) and use that to inform a forecast.
In July, that was easy. Undecideds leaned D. Now? It's much, much harder to detect.
This is a lot of words, and im not gonna read all of them so i am reacting to just the first few paragraphs
Yes, polls are a snapshot in time, but they are predictive of future results. Say you are a voter who says you will vote trump (or kamala) in september.sure you can change your vote jn the next 3 months, and many people do, but any reasonable human beeing would take fsr greater than even odds that you will in fact vote for whatever candidate you choose.
Ar even ods, with no other information I would always bet on the person ahead. As a matter of fact ill take even odds regardless of the answers to all but question 2. And we have precisely 0 information about the nature of the runners.
At a certain point, do you start to look at early voting data? Such as the number of mail in ballots requested and returned broken down by party / gender / ethnicity.
I definitely look at it but that information is very hard to incorporate into a forecast. The demographics of who is expected to vote early ..and where..is subject to so many errors that I'd rather not pile on more assumptions. It's true I might miss a small edge one way or the other, but I'd rather have that than a dramatic error of my own making
This is an awesome breakdown. Thank you. To me it says that if we in the Harris camp keep the pedal to the metal all the way to election day then we will likely win, even if the numbers are close! Let’s keep pushing forward and reaching those voters folks. Write postcards, knock doors, phone bank, and bring three people with you to vote for Harris!
Every now and then one comes across a perspective that is such an original way of looking at things that it seems counterintuitive at first. But once someone explains it clearly, it seems so obviously valid that you wonder why anyone ever looked at it the old murky, unreliable way.
Thank you, Mike and Carl, for your crystal clear explanations of polling, which are making me feel like a genius (for a few hours, at least). 🙏
Needless to say, I am very happy to see this breakdown. This is a very good and in depth and really brings the points home. Thanks Carl and Mike.
Another reason to include basic maths and statistics in high schools. Part of what fed MAGA's outrage in 2020 was the comparing 2016 raw numbers to 2020 numbers: "How could Trump loose if he out performed his previous outcome?"
Carl does a wonderful job of explaining for those who want to know.
An amazing amount of data and excellent explaining for those of us civilians. Bottom line, correct me if I'm wrong, is that I'd rather be Harris-Walz than Trump-Vance going into the home stretch. Many thanks for this great article.
Lots to nerd out on here, but one quick question. In your video a couple of posting ago, Mike, where you were talking about polls that included (but didn't split out) undecides. In these polls, do they assume undecides split evenly as shown in the bar charts? Do they even give you enough information to tell? (that's 2 questions... I know).
Not Mike, but hopefully I can answer -
Polls (in the US, that is) traditionally report the numbers for the top two candidates: say, 49%-45%
The rest of those unstated numbers could be third-party or undecided.
Many analysts simplify that 49-45 poll to "up by 4" which creates some problems
Pollsters - in US cases - don't "assume" what undecideds will do - they just report how many their poll had.
Analysts, and forecasters, tend to assume (as a default assumption) that undecideds will split evenly. That can create some problems and it's a pretty bad assumption honestly.
As a forecaster, we can look at all the data from that sliver of undecided (demographics? Past vote? What issues matter? Will they even vote?) and use that to inform a forecast.
In July, that was easy. Undecideds leaned D. Now? It's much, much harder to detect.
Thank you, Carl. I appreciate your taking the time to respond.
This is a lot of words, and im not gonna read all of them so i am reacting to just the first few paragraphs
Yes, polls are a snapshot in time, but they are predictive of future results. Say you are a voter who says you will vote trump (or kamala) in september.sure you can change your vote jn the next 3 months, and many people do, but any reasonable human beeing would take fsr greater than even odds that you will in fact vote for whatever candidate you choose.
That's a long comment, not reading all of it.
"Yes, polls are a snapshot in time, but they are predictive of future results"
So is a snapshot of a footrace
If the person who is ahead in the snapshot doesn't eventually win, does that mean the snapshot was wrong? Yes or no?
No, but it does mean that something happened. And if you cant explain what it is than you shoild assume your measurements are incorrect.
If I offered you a bet on a race, would you bet in who was jn front or behind?
"if I offered you a bet on a race, would you bet in who was in front or behind"
It depends on:
1) How far they are from the finish line
2) Who the racers are
3) How much closer to the finish line the leader is
4) The race environment
What about you? Would you bet on who was in front or behind?
Ar even ods, with no other information I would always bet on the person ahead. As a matter of fact ill take even odds regardless of the answers to all but question 2. And we have precisely 0 information about the nature of the runners.
"as a matter of fact ill take even odds regardless of the answers to all but question 2"
That's what's known as a sucker
In a race to 50 meters, you think a photo with me pictured ahead 49-47, isn't as good of information as a photo with me pictured ahead 5-3?
Be glad no one wants your money
No, but i would take both bets at even odds
"at even odds, with no other information"
That's the thing. We do have other information.
Read the article if you'd like to understand this better.
But the thing is having read your article, you don't
At a certain point, do you start to look at early voting data? Such as the number of mail in ballots requested and returned broken down by party / gender / ethnicity.
I definitely look at it but that information is very hard to incorporate into a forecast. The demographics of who is expected to vote early ..and where..is subject to so many errors that I'd rather not pile on more assumptions. It's true I might miss a small edge one way or the other, but I'd rather have that than a dramatic error of my own making
That makes total sense. Thank you!