Decoding the polls – What polling samples currently should be & what they are
Bottom Line: As I’m inclined to say... If the premise of anything is false, anything built on top of that false premise will be too. If the sample of political polls isn’t objective, the polling results can’t be taken at face value. Recently I illustrated how Gallup’s party ID shows the current landscape to be Democrat +1. Meaning that among those who ID with a political party, 1% more identify as a Democrat as opposed to a Republican. That means a proper sample for pollsters would be to sample 1% more Democrats than Republicans. But that’s generally not happening.
Near daily new polls are rolling in showing Joe Biden well ahead of President Trump in head to head polls. But here’s the thing. The samples are skewing the perspective. I took the two most recent head to head polls which came in Monday to illustrate the point. First up the CNN Poll, finding methodology – which in the interest of transparency should be at the top of any polling report – is regularly buried deep inside of these surveys. In CNN’s poll you’ll find the methodology on page 23. Why? Only they know but here’s what we know. They sampled 4% more Democrats than Republicans. The poll showed Biden up 4% - outside of the margin of error. But with the pollsters having oversampled Democrats by 3%, it’s safe to say any lead for Biden is negligible and certainly within the margin of error. Next up the ABC/Washington Post poll. You’ll find the sample data on page 6. Still buried a bit but easier to find than in CNN’s. They sampled 5% more Democrats than Republicans in a poll that show Biden above 50% nationally and with a ten-point lead over Trump. However, their poll oversampled Democrats by 4% and would imply a six-point race with Biden around 50% as opposed to over it.
This becomes one of the more challenging aspects of wading through polling data. It’s rare that polls can be taken at face value as the sample must first be accounted for prior to drawing deductions about the results. In both instances the polls still show Biden with at least a slight edge nationally once adjusted for sample, but nowhere near the large/outside of the margin of error types of leads implied without proper context. This is the issue with taking polling data at face value and it’s why it’s necessary to decode the polls. To be continued...