I read you from Europe and I am fascinated by your work ( I bought your book few days ago - still not read). Here most of the electoral systems are proportional with (many) more than 2 parties: how does the science of polls change? Is it still a game mostly played by the undecided?
Brilliant observation. The math is much, much harder when you're dealing with 3, 4, 5 viable parties
The undecided impact is often more magnified - and uncertain.
The book talks about 2016 in the US (higher 3rd party) and specifically the state Utah (3 competitive parties) as a way to illustrate the issues in the assumptions of proportional allocation.
Both spread and proportional have issues, but only proportional allocates undecideds - when they are 10% or 20% (not uncommon) that introduces a huge source of error not otherwise present in the poll data
You have Trump leading 48.4 - 47.7 in NC, with 2.7% undecided, and a 54% Trump chance. Then you have Harris leading 48.1 - 47.4 in Michigan, also with 2.7% undecided, but your win probability for Harris is 66%.
Would be nice to know a bit more about your assumptions with undecideds, because unless I'm very much mistaken, your forecast predicts undecideds to break Harris.
Really enjoying your insightful analysis Carl. What % chance do you have Harris of winning the popular vote? Seen you've been debating her national polling on Twittter.
Interesting read. I think this point raises interesting questions, and I'd love to read a full length piece from you on your rationale behind this question (I did click the hyperlink, just want more):
"all else equal, a poll average of 49-46 has a far higher win probability than one of 46-40."
If your answer is, "buy the book", fair enough, I haven't yet, but I likely will.
I assume the main relevant parts of "All else equal" are sample size, demographics, methods e.g. calling landlines, proximity to election day, etc. It's an interesting thought experiment.
In a presidential election, I tend to think the true "undecided" vote share is low. While I do view 49-46 differently from say 46-40, the notion of two polls with those different results truly being otherwise "equal" is difficult to imagine, because in a vacuum I would assume the latter poll is methodologically worse in some way e.g. "14% undecided / 3rd party" is unlikely and could indicate that they didn't get as good of a sampling of likely voters. In other words, I agree that it's dubious to consider "46-40" as "+6" if "49-46" is "+3", and I suppose I do put more weight into the 49-46 poll as favoring the leading candidate more, though it's very close and again very difficult for me to imagine "all else being equal" in this hypothetical.
One interesting question to me is what methods could be inflating the undecided count?
Does listing extra third party candidates who won't actually appear on the ballot inflate the undecided count? It seems plausible.
What about the way that pollsters phrase the question? The type of poll e.g. voice call vs web form? There is no "undecided" option on the ballot. Is there an argument for removing "undecided" as an option from certain polls (specifically presidential ones, close to election day, no upcoming debate)?
Do other factors contribute to voters selecting "undecided" e.g. general dissatisfaction with the choices, even when they really would never vote for one of the two candidates? Or perhaps voters who are less likely to vote are more likely to select "undecided" on a poll, and their choice reflects a general apathy towards elections?
If you want the in depth "why" then of course the book is the answer.
The "all else equal" calculation has to do with third parties. The concept of the "finish line" (details in various posts and the book) is the percentage of votes that will guarantee victory. 50% is the only number we know with certainty guarantees victory (i.e. win probability = 100%)
But when there is substantive third party turnout (eg 2016) that number can be much lower.
You're right that Presidential Elections rarely have 10%+ undecided (though major non-US elections often do) but Senate, Gov races are quite common
But how many undecideds there are in a poll is directly related to how the question is asked - and how the data is reported. So much to get into, but these are all good questions. If you're interested in this level of depth, I'd definitely recommend the book
Hello, I wanted to ask about the chance that either candidate sweeps all 7 swing states. Why would the chance of that be less than Silver's 40% estimate? I would've thought that you would find the chance to be higher than him, given what you believe about the importance of undecideds.
Your averages are ludicrous; Trump is the favorite. Your favored candidate has lost her momentum and you have yet to update your stupid percentages. Seek help dude. It’s like a 68 percent Trump wins. You just think Trump is going to end democracy or whatever lol seek help Carl Beta Male Allen
I read you from Europe and I am fascinated by your work ( I bought your book few days ago - still not read). Here most of the electoral systems are proportional with (many) more than 2 parties: how does the science of polls change? Is it still a game mostly played by the undecided?
Brilliant observation. The math is much, much harder when you're dealing with 3, 4, 5 viable parties
The undecided impact is often more magnified - and uncertain.
The book talks about 2016 in the US (higher 3rd party) and specifically the state Utah (3 competitive parties) as a way to illustrate the issues in the assumptions of proportional allocation.
Both spread and proportional have issues, but only proportional allocates undecideds - when they are 10% or 20% (not uncommon) that introduces a huge source of error not otherwise present in the poll data
You have Trump leading 48.4 - 47.7 in NC, with 2.7% undecided, and a 54% Trump chance. Then you have Harris leading 48.1 - 47.4 in Michigan, also with 2.7% undecided, but your win probability for Harris is 66%.
Would be nice to know a bit more about your assumptions with undecideds, because unless I'm very much mistaken, your forecast predicts undecideds to break Harris.
Poll average to forecast isn't a simple translation. I made that clear in the first paragraph of the post I believe.
The probability comes from the range of outcomes - I'm guessing the NC band is narrower than MI.
Really enjoying your insightful analysis Carl. What % chance do you have Harris of winning the popular vote? Seen you've been debating her national polling on Twittter.
90%. Highly unlikely she doesn't win the popular vote. That number might be higher closer to the election
Great analysis
Interesting read. I think this point raises interesting questions, and I'd love to read a full length piece from you on your rationale behind this question (I did click the hyperlink, just want more):
"all else equal, a poll average of 49-46 has a far higher win probability than one of 46-40."
If your answer is, "buy the book", fair enough, I haven't yet, but I likely will.
I assume the main relevant parts of "All else equal" are sample size, demographics, methods e.g. calling landlines, proximity to election day, etc. It's an interesting thought experiment.
In a presidential election, I tend to think the true "undecided" vote share is low. While I do view 49-46 differently from say 46-40, the notion of two polls with those different results truly being otherwise "equal" is difficult to imagine, because in a vacuum I would assume the latter poll is methodologically worse in some way e.g. "14% undecided / 3rd party" is unlikely and could indicate that they didn't get as good of a sampling of likely voters. In other words, I agree that it's dubious to consider "46-40" as "+6" if "49-46" is "+3", and I suppose I do put more weight into the 49-46 poll as favoring the leading candidate more, though it's very close and again very difficult for me to imagine "all else being equal" in this hypothetical.
One interesting question to me is what methods could be inflating the undecided count?
Does listing extra third party candidates who won't actually appear on the ballot inflate the undecided count? It seems plausible.
What about the way that pollsters phrase the question? The type of poll e.g. voice call vs web form? There is no "undecided" option on the ballot. Is there an argument for removing "undecided" as an option from certain polls (specifically presidential ones, close to election day, no upcoming debate)?
Do other factors contribute to voters selecting "undecided" e.g. general dissatisfaction with the choices, even when they really would never vote for one of the two candidates? Or perhaps voters who are less likely to vote are more likely to select "undecided" on a poll, and their choice reflects a general apathy towards elections?
If you want the in depth "why" then of course the book is the answer.
The "all else equal" calculation has to do with third parties. The concept of the "finish line" (details in various posts and the book) is the percentage of votes that will guarantee victory. 50% is the only number we know with certainty guarantees victory (i.e. win probability = 100%)
But when there is substantive third party turnout (eg 2016) that number can be much lower.
You're right that Presidential Elections rarely have 10%+ undecided (though major non-US elections often do) but Senate, Gov races are quite common
But how many undecideds there are in a poll is directly related to how the question is asked - and how the data is reported. So much to get into, but these are all good questions. If you're interested in this level of depth, I'd definitely recommend the book
Hello, I wanted to ask about the chance that either candidate sweeps all 7 swing states. Why would the chance of that be less than Silver's 40% estimate? I would've thought that you would find the chance to be higher than him, given what you believe about the importance of undecideds.
While all 7 swing states are correlated it's not as strong in AZ, GA, and NC
Undecideds are also correlated, but not a monolith across states either
In a close election, the chances one or a few of the swing states go opposite the others is high
That makes sense. How likely is it, according to your model, that it will be a close election?
Depends on your definition of close!
Popular votes in the decisive swing states being within 2-3 points, high
Your averages are ludicrous; Trump is the favorite. Your favored candidate has lost her momentum and you have yet to update your stupid percentages. Seek help dude. It’s like a 68 percent Trump wins. You just think Trump is going to end democracy or whatever lol seek help Carl Beta Male Allen