Intelligent thoughts from Nate Silver on the riddle wrapped inside a mystery inside an enigma of the 2024 presidential-election polls; Selzer vs. Siena vs. the herd: who is...
It's not a common form of visualization, but a graph showing the Harris win probability (and/or the width of our posteriors as of current information) as a function of different combinations of pollster informational characteristics would be interesting. At least in this sort of analysis our inference about voting outcomes "goes thru" (in the Bayesian network sense) assumptions about pollster informational characteristics, so that's the set of relevant priors to study sensibility to.
(I think this generalizes to the problem of getting information online on a topic you aren't well-informed on; rather than evaluating the probability of statements, you should focus on evaluating the reliability of experts and sources, although it's usually not as quantitative as in the mixture-of-mixtures-of-polls case.)
"Siena has Gallego leading Lake by 7%-points while Harris trails Trump by 5%-points in Arizona."
I really have trouble believing that Gallego will beat Harris' margin by _twelve points_. Five or six, sure. But something is wonky here. There aren't that many ticket splitters anymore.
I'm pretty much doing what Professor DeLong suggests, just following Nate's comments. He has been around the block too many times and does not make foolish mistakes on data interpretation. My gut feeling is that this election is decided by the female electorate. They make up a slight majority of voters and if they turnout is huge (they have a greater propensity to vote than males), Harris will win.
A Trump victory will lead to a 'Lysistrata' outcome with a resultant very high increase in incels.
It's not a common form of visualization, but a graph showing the Harris win probability (and/or the width of our posteriors as of current information) as a function of different combinations of pollster informational characteristics would be interesting. At least in this sort of analysis our inference about voting outcomes "goes thru" (in the Bayesian network sense) assumptions about pollster informational characteristics, so that's the set of relevant priors to study sensibility to.
(I think this generalizes to the problem of getting information online on a topic you aren't well-informed on; rather than evaluating the probability of statements, you should focus on evaluating the reliability of experts and sources, although it's usually not as quantitative as in the mixture-of-mixtures-of-polls case.)
"Siena has Gallego leading Lake by 7%-points while Harris trails Trump by 5%-points in Arizona."
I really have trouble believing that Gallego will beat Harris' margin by _twelve points_. Five or six, sure. But something is wonky here. There aren't that many ticket splitters anymore.
In other words we have no idea but it feels good to talk about it. [Or does it?]
I'm pretty much doing what Professor DeLong suggests, just following Nate's comments. He has been around the block too many times and does not make foolish mistakes on data interpretation. My gut feeling is that this election is decided by the female electorate. They make up a slight majority of voters and if they turnout is huge (they have a greater propensity to vote than males), Harris will win.
A Trump victory will lead to a 'Lysistrata' outcome with a resultant very high increase in incels.
Sadly, this is not VAR.