Is polling really dead? Let’s take a beat.

Some thoughts on what happened & why we need more information to make a full assessment.

A lot of post-election chatter has focused on polling misses in various states including some takes that polling is “dead.” Before we call time of death on public opinion research there are a few things to consider.

1. Were the polls off in some places? Yes. Do we know by how much at this point? No. Once all of the votes are counted, we should have a better idea of the magnitude of the error. It’s worth noting, the polls were correct in some places.
2. The pandemic created variables that did not exist in previous years, including the methods by which many Americans cast ballots. It’s difficult to precisely account for the impact the added variables had (a point here and a point there makes a difference).
3. There appear, at least at the moment, to be some regional differences in how the polling shook out. They are worth examining once the all of the votes have been tallied.
4. Declining response rates are probably a factor. Fewer people are answering pollsters, which requires that some pollsters make judgements about weighting their samples to reflect the broader electorate. Specifically, certain types of voters are harder to reach than others.
5. The concept of “shy Trump” voters has been batted around as a reason why the polls in 2016 and again in 2020 appear to underestimate his support (again for 2020 the final tally remains to be seen). My hunch is, it’s more a problem of defining the electorate. Pollsters have to make judgements, based partially on historical turnout data, about who is likely to vote in an election. 2016 brought more non-college white voters out to the polls than in prior years. That may have be the case again in 2020.

Polling is a useful tool to understand what a population is thinking. It helps inform the deployment of resources, the framing of issues and the construction of messaging. That being said, polling is far from a perfect instrument. Shorter surveys, new modes of data collection, and more in-depth qualitative research should help improve reliability. We still have a long way to go before we can make a comprehensive assessment of what happened in 2020.