Tweet
On polls: There was a very good article in the Economist talking about the differences in polling and modeling between 2016 and 2020. The modeling in 2016 failed to account for a bias:
In 2016 many pollsters in the northern battlegrounds made a methodological error that underestimated support for Mr Trump. Pollsters usually have to adjust their data to be demographically representative of the population as a whole. Imagine that in a pollster’s sample of likely voters, 50% do not have a bachelor’s degree. But according to the Census Bureau, in 2016 the share of voters without a four-year degree was 60%. So to get a representative sample, more weight must be given to this group. In 2016 many pollsters simply did not adjust their data for this bias, causing them to undersample white voters without college degrees who favoured Mr Trump but were less likely to take phone surveys than the typically better-educated supporters of Mrs Clinton.
See images in subsequent post.
As far as I can tell, The Economist has built an impressive model that has been tracking the US presidential election closely, updating the model daily. This is their final. As with any poll, I'm not taking this one at face; it's showing a resounding Biden victory that's not even close. Just taking the pulse in FL, reading the press, it's a lot closer than this model thinks it is. Of course, FL is one of those states where it is close so I'd expect some reporting bias that extrapolates FL's race to all races nationally. That's not accurate.
The other thing is that we're not going to see whether this model is correct or not for a while because of the ballot counting/recounting (assume some will be challenged) delays. I'm somewhat uncomfortable with that as there is going to be shit-ton of misrepresentations and lies coming from many sources. It's the lack of certainty in an election that is going to be prolonged that might give rise to social unrest.
In 2016 many pollsters in the northern battlegrounds made a methodological error that underestimated support for Mr Trump. Pollsters usually have to adjust their data to be demographically representative of the population as a whole. Imagine that in a pollster’s sample of likely voters, 50% do not have a bachelor’s degree. But according to the Census Bureau, in 2016 the share of voters without a four-year degree was 60%. So to get a representative sample, more weight must be given to this group. In 2016 many pollsters simply did not adjust their data for this bias, causing them to undersample white voters without college degrees who favoured Mr Trump but were less likely to take phone surveys than the typically better-educated supporters of Mrs Clinton.
See images in subsequent post.
As far as I can tell, The Economist has built an impressive model that has been tracking the US presidential election closely, updating the model daily. This is their final. As with any poll, I'm not taking this one at face; it's showing a resounding Biden victory that's not even close. Just taking the pulse in FL, reading the press, it's a lot closer than this model thinks it is. Of course, FL is one of those states where it is close so I'd expect some reporting bias that extrapolates FL's race to all races nationally. That's not accurate.
The other thing is that we're not going to see whether this model is correct or not for a while because of the ballot counting/recounting (assume some will be challenged) delays. I'm somewhat uncomfortable with that as there is going to be shit-ton of misrepresentations and lies coming from many sources. It's the lack of certainty in an election that is going to be prolonged that might give rise to social unrest.
Comment