What Needs to be Done?
Some initial notes on the future of polling and organizing for progressive causes
Democrats have been talking about turning Texas blue for nearly a decade, on the assumption that a fast-growing Hispanic population and suburban growth would finally make it possible for Democrats to win statewide elections. But this year’s election saw Latino voters in South Texas support Trump in surprisingly large percentages, underscoring the danger of assuming that voting patterns among various groups of voters will remain constant indefinitely.
Dan Blaz, Washington Post, November 14, 2020
Polling may need to start thinking smaller – or perhaps smarter – in light of the partisan non-response and general non-response problems. We need to spend a lot more time understanding the electorate this year and where it did and did not conform to expectations about the election from the polling. But, here are some initial observations from the data we have so far and one potentially big idea to consider.
There certainly seems to have been a partisan non-response problem with the polling, something I’ve discussed in an earlier post. When coupled with the general non-response problem (live interviews reach only about one percent of voters called), pollsters must rely on weighting protocols. It is not as easy as it sounds. In the early days of polling, there was a Democratic non-response problem. Rather than intentionally avoiding polls (as we suspect is part of today’s problem), pollsters did not make enough effort to reach poor and working class folks who made up a significant portion of Roosevelt’s New Deal coalition. Over time, George Gallup figure out how to more accurately create representative samples in polling. From then until recently, partisan non-response was not a problem. There was no partisan divide between respondents and non-respondents, so weighting with demographic characteristics was usually good enough to get to representativeness in the sample. It was unclear in 2016 and 2018 why we saw the polling error we did, but after 2020 it seems like there is a partisan non-response problem and that has underestimated Republican support in general. However, in the 2018 midterms, we saw polling underestimating both Democratic and Republican support in different areas of the country.
This year (as in 2018) we see regional variation as well as demographic variation working in conjunction with the partisan non-response problem. For instance, we can explain why the Latino voters in South Florida behaved differently than the Latino voters in California. The Latino community in South Florida has a large portion of ex-pats and refugees from Cuba, Venezuela, and Colombia who we assume are more inclined to be persuaded by the GOP’s anti-socialist rhetoric. In California, a large portion of the Latino population is Mexican. This community would be expected to be more persuaded by opposition to the GOP’s border and racist policies. Perhaps an oversimplification, but even so it doesn’t help us understand why did Latinos in South Texas apparently voted more like South Florida than California.* Organizing among these populations with an objective to understanding their voting motivations would not only help improve the organizing efforts, but it would improve the polling.
I think one answer is for Democratic Party organizations – particularly those that can raise money – to partner with community-based organizing groups all around the country. These party organizations and PACs (“the party”) should raise money for polling that also helps fund the canvassing work of local progressive organizing groups. There are a number of good reasons to do this, from supporting the work of progressive organizing on an ongoing basis (rather than only at election time), identifying and supporting rising community leaders, making better connections between local issues and state and federal policy, and establishing closer ties between community organizing and electoral work. But, for the purpose of this post I want to suggest that this kind of a relationship would help strengthen the ability to more accurately poll voters.
Here is how it might work. The party would partner with progressive CBOs that canvass their membership and/or larger constituencies. This partnership would include subsidizing the canvass operation – not taking it over; this is a partnership and local control should be prioritized.** The canvass would continue to focus on the issues and needs the CBOs have already established through their own processes, but there would be some additional work done to understand local voters better for polling purposes. What that means will likely differ from locality to locality and on the problems pollsters have already identified for that area. In this sense the party would utilize the canvassing work of organizing groups to groundtruth*** polling. That is, the canvass can help pollsters understand if they are missing something important that will compromise the accuracy of their polling. At the same time, organizing groups can help educate voters in ways that can (1) organize and mobilize them to vote for progressive candidates and ballot questions and (2) make them more comfortable with participating in polling.
At this point it is worth raising the question again: Why do we care about polling?
Think of it this way. Polling is about understanding public opinion. Period. A lot of people (myself included) find following the polls fascinating – very much like following sports. But politics is not a sport, and polling should be able to tell us something meaningful about what people who will, or can be motivated to, vote think at a given point in time. It needs to also tell us when the opinions of voters change or may be persuadable. This, in turn, helps organizers to allocate resources, frame messages, and target canvassing campaigns, voter registration drives, and GOTV efforts.
I am not saying that folks on the ground did not see what was coming. In fact, we know folks in South Florida were concerned about the Latino and Black vote there (Black men gave less support to Biden than they did to Clinton). I bet the same is true in South Texas. However, this groundtruthing was never connected to polling, even though we had really important CBO, nonprofit, and union coalitions working to get out the vote in places like Arizona (Arizona Wins) and Georgia (ProGeorgia). We can safely say that groundtruthing was not connected to the internal party polling since we observed how campaigns on both sides behaved with the information they had.
The mid-terms in 2022 will be even more important than we thought going into this year’s election. The Senate map is more favorable to Democrats than it was this year and this should be the first year of new Congressional maps based on the 2020 Census. In this era of new challenges to the political conventional wisdom, we should be thinking of how we will more effectively campaign to win the Senate and hold and expand the House majority. We should ignore the historical trends of mid-term elections in a new president’s first term, except to understand them better in order to overcome them. Accurate internal polling is essential to this.
* We will need to look closely at the data and follow-up surveys in the weeks to come to really understand the relationships between voting, demographic characteristics, and regional variation.
** CBOs use different organizing models. Presumably, they know best what works for the population and community they are organizing. The needs of the party (whether polling or otherwise) can be incorporated into the partnership, but they should not be used to change the organizing model and mission of their CBO partners. In addition to this not being helpful or respectful, it would undermine the very reason for the party engaging in the partnership in the first place - which is to better understand and connect with local voters.
** Groundtruthing is a term used in various disciplines and fields to refer to information provided by direct observation (i.e. empirical evidence). It is sometimes used to mean a process that privileges direct observation over information provided by inference. But, more commonly – and in the sense I mean it here – it refers to using direct observation to validate or refute information provided by inference.