The realities of the COVID-19 pandemic mean that we need to rethink who is likely to show up at the remaining 2020 primaries and, perhaps, even the November general election.
Campaign professionals should be attuned to how polling, modeling, and voter targeting should be adjusted for the remainder of this cycle.
One of the primary challenges in public opinion polling is the art and science of determining who is likely to show up and vote on Election Day. There are several ways that pollsters determine who is likely to vote.
First, we study the voter file to understand who has voted in a recent or similar kind of election – such as a Presidential year Congressional primary, or which voters have registered to vote since the last election.
Both of these selection methods can be valuable predictors of who is likely to vote and thus are likely future voters who are included in who will be contacted by pollsters.
Second, of those selected, based on the criteria above, pollsters use a series of screening questions when initially speaking with voters, to understand how likely they are to vote on Election Day.
In some cases, voters are quizzed on “when Election Day is” or “where their polling site is located” to confirm that they are just not providing socially acceptable answers (“Of course, I always vote!”).
These two steps determine who gets administered the poll. When these steps are not followed, polling can be inaccurate.
Let’s look at some of the polling in the lead up to the November 2016 general election. A small but significant number of moderate and conservative white working class voters were not included in the voter selection criteria because they hadn’t voted in the 2012 presidential – either because they were at least somewhat satisfied with the job Barack Obama was doing and/or were unenthusiastic about Mitt Romney’s candidacy. Pollsters incorrectly assumed many in this cohort would not vote in November 2016 either, and thus these voters weren’t contacted as a result.
That meant many polls accurately showed who was going to win if the turnout looked similar to the 2012 turnout, namely Hillary Clinton would prevail. But on Election Day 2016, a slightly different electorate showed up and voters handed Donald Trump a small but winning margin in states including Pennsylvania, Michigan, and Wisconsin.
So which voters are selected by pollsters is vital to the accuracy of the poll and to the projection of the likely outcome.
Now, let’s consider two of the most likely COVID-19 scenarios and what polling needs to take into account:
Scenario One is what we saw recently in the Wisconsin Democratic primary: turnout is lower because voters fear going to the polls in-person and the mechanics of requesting and then voting by absentee ballot elude most voters (as well as some Board of Election officials), especially so-called low information voters.
If this scenario is repeated throughout the remainder of the 2020 cycle, the anticipated surge in turnout that experts have been predicting would not fully materialize.
Pollsters would need to adjust in two ways: this scenario will require a much tighter screen of who is included in the pool of voters to survey and require a tighter screen on the intent-to-vote question. Putting this in place makes polling more expensive to conduct, so campaigns need to be willing to pay slightly higher costs.
At the outset of a poll, many pollsters ask whether voters are “absolutely certain to vote,” are “very likely, ‘somewhat likely,” “not likely”, or “not very likely” to vote.
Usually, voters indicating any of the first three choices are included in a poll, but a lower expected turnout means pollsters may only be able to select voters answering “absolutely certain” or “very likely” to vote. A tightening of the selection criteria will be needed and, indeed, is necessary.
Scenario Two is that turnout is higher – states, or the federal government, allow voting by mail as states like Colorado, Hawaii, Oregon, Utah, and Washington already do.
Indeed, this proposal is being pushed for the whole country by Sens. Amy Klobuchar (D-Minn.) and Ron Wyden (D-Ore.).
Under this scenario, since it’s easier to cast a ballot by mail, turnout goes up. As an example, Colorado, where all registered voters received not an absentee ballot application, but an actual ballot to fill out, voting by mail “dramatically” increases voter turnout. In all age groups, but especially those who tend to vote less frequently.
In other states, as we saw recently in the Ohio primary, you need to request an absentee ballot, and then are sent a ballot to return by a specific date, which doesn’t boost turnout quite as much.
The challenge with either of these scenarios is that a larger pool of voters with no meaningful voter history including many non-habitual voters – offers fewer clues about whether someone will turn out or not. That means new turnout models will need to be built and refined more than a typical model, which is built around more frequent voters.
Chris Wilson, of WPA Intelligence, a leading GOP polling, and analytics firm told me recently that he’s not sampling differently, but he and his team are busy building models and scoring voters for their likelihood to vote by mail.
Increasing the number of voters who can vote by mail could lead to even higher voter turnout than what experts are already predicting. In this case, past election participation will be a far less important qualifier: the traditional polling screens will need to be relaxed.
Not including the white working-class “surge” voters in the 2016 election gave pollsters a Black Eye. It’s time for pollsters to update their methods and methodologies to account for a series of different COVID-19 turnout realities or risk another stain on our reputations. Pollsters, you are forewarned.
Bradley Honan is the CEO & President of Honan Strategy Group, a Democratic polling and data analytics firm.