I’m starting to question 538s predictions not because there is any problem with the models, but because there is a filter on what polls they choose to include. I don’t want to call it bias, it’s just a blind spot that their model isn’t getting all the input for.
Of course it’s not like anyone else is doing a better job either.
Likely voters are those that have voted before, that’s what makes them likely to vote again. For the most part they’re the more accurate people to be polling.
Because new voters are a rounding error smaller than the error bars of the sample size.
Polling is pretty much like the unemployment rate. Any individual reading is meaningless, it’s a multitude of readings over time that give any useful information.
I’m starting to question 538s predictions not because there is any problem with the models, but because there is a filter on what polls they choose to include. I don’t want to call it bias, it’s just a blind spot that their model isn’t getting all the input for.
Of course it’s not like anyone else is doing a better job either.
Polling is inherently problematic every time you see they polled “likely voters” as opposed to “registered voters”.
If they’re self selecting who they consider to be “likely”, it’s going to have a skewed result.
Likely voters are those that have voted before, that’s what makes them likely to vote again. For the most part they’re the more accurate people to be polling.
And discounts first time voters.
Some even only count people who voted in the last 2 elections.
Because new voters are a rounding error smaller than the error bars of the sample size.
Polling is pretty much like the unemployment rate. Any individual reading is meaningless, it’s a multitude of readings over time that give any useful information.