Quantitative Editor, @FiveThirtyEight | PhD, @LSEGovernment | 
in 
, via 
| she/her
You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Last week, a CNN poll had Biden 15 points ahead of Warren, while a Quinnipiac poll had Warren leading by 7. Is this a sign that the polls are too far apart? In short: No, once we account for house effects, the spread is within a normal range.https://fivethirtyeight.com/features/biden-up-15-warren-up-7-are-primary-polls-too-far-apart/ …
I ran some simulations to see whether the spread in the polls in October is larger than we'd expect. And based on sampling variability alone, it is -- way larger. The standard deviation of Biden's polls, for example, should be around 2, but it's actually 3.5.pic.twitter.com/aZ6YK2bMi5
But sampling variability isn't the only thing driving the spread. Once we account for house effects, the observed standard deviation is within the range we'd expect. And house effects are healthy! We *want* different pollsters making independent methodological decisions.pic.twitter.com/4JKhJgjl3s
What about the CNN and Quinnipiac polls? Turns out, once you account for house effects, they're exactly the kinds of normal outliers you'd expect to see in this kind of distribution of polls.pic.twitter.com/1BLyW2TURs
Super excited for my first morning story at @FiveThirtyEight! With a ton of thanks to @ellawinthrop and @jazzmyth for the beautiful charts, and @sfrostenson, @CopyCurmudgeon, @DataDhrumil, @NateSilver538, and @micahcohen for everything else!
It would be interesting to see what assumptions go into top pollsters’ methodologies. I assume the “house effect” is due mostly to varying assumptions on turnout demographics but the specifics would be educational, if available.
In that case, it would be useful to have some analysis of the sampling and weighting methodologies to understand why there is such variation. My sense is that the biggest variables are assumptions about the turnout of younger voters, but it's hard to be sure.
From SSRS: they are explicit that the names were read in random order. In both polls, respondent has to wait thru the whole list before they answer.pic.twitter.com/y33pz8Swyb
By that time, most people would fall asleep!
next time someone asks me how these polling gaps are possible I’ll refer them to this piece! it’s the perfect level of detail, convincing but not overwhelming
Can you can share the house effects you calculated for each of the prominent 2020 Democratic primary pollsters?
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
