This is less about vaccines than about polling methodology. It was a rare opportunity to compare polls to definitive data: we knew for sure exactly how many vaccines were administered.
Traditional polling did fine. Other methods -- one from Facebook and one from the Census -- did poorly, despite more data, because of systematic selection bias. In fact the amount of data made it worse, because it magnified the effects of small systemic biases.
This was taken early in vaccine delivery, when the total number vaccinated was small. That, too, would magnify errors. I wonder if the same survey methods work any better over time.
It would be interesting to see - but that effective sample size of 10 for the Facebook survey of 4,500,000+ people should be terrifying to researchers in Big Data.
2 comments
[ 2.7 ms ] story [ 12.4 ms ] threadTraditional polling did fine. Other methods -- one from Facebook and one from the Census -- did poorly, despite more data, because of systematic selection bias. In fact the amount of data made it worse, because it magnified the effects of small systemic biases.
This was taken early in vaccine delivery, when the total number vaccinated was small. That, too, would magnify errors. I wonder if the same survey methods work any better over time.