Read on Twitter
Here's what you should do with any horserace poll result for the next 8 weeks:
-See that it says "Biden X%, Trump Y%"
-Automatically assume those are fuzzy estimates, not precise % of support each has.
Why? Because surveys have error. (1/5)
-See that it says "Biden X%, Trump Y%"
-Automatically assume those are fuzzy estimates, not precise % of support each has.
Why? Because surveys have error. (1/5)
We talk about "margin of error" but that's just the start.
Who did the poll use to get to that estimate? They say "likely voters," but we really have no idea who will vote - that population doesn't exist yet!
That's error we can't estimate. (2/5)
Who did the poll use to get to that estimate? They say "likely voters," but we really have no idea who will vote - that population doesn't exist yet!
That's error we can't estimate. (2/5)
When we discuss subgroups, especially, say, Black and Hispanic voters: Those sample sizes get small in a typical poll of ~1,000. You get 100-200 max in each group. Those margins of error shoot up! So add more fuzziness to those estimates in your head. (3/5)
If you want to know more about overall poll error, google "Total Survey Error," or read the first half of this:
(4/5)
http://centerforpolitics.org/crystalball/articles/poll-based-election-forecasts-will-always-struggle-with-uncertainty/
(4/5)
http://centerforpolitics.org/crystalball/articles/poll-based-election-forecasts-will-always-struggle-with-uncertainty/
But for the love of all that is important to you, DO NOT compare polls to actual results from 2016 without acknowledging MOEs at minimum. And assume ALL polls have some fuzziness around their numbers. Train your brain to not see exact numbers, but to see approximations. (5/5)
