RESOLVING THE LEDGER
This is one outlet's own report from Newsweek — the article as it was filed. Other outlets are covering the same event; open the full story to compare every source side by side.
See the full story · 1 sources
0 Share Newsweek is a Trust Project member See more of our trusted coverage when you search. Prefer Newsweek on Google to see more of our trusted coverage when you search. Pollsters spent years refining the boxes into which Americans fit but this month, voters in two Midwest Democratic primaries demonstrated how quickly the people inside those boxes can shift.
In Wisconsin, David Crowley defeated Francesca Hong by roughly half a percentage point in Tuesday's highly watched Democratic gubernatorial primary after two polls had shown Hong way ahead by about 20 points. A week earlier in Michigan, polls also missed the margin of Abdul El-Sayed's Senate-primary victory by double digits.
Has polling become bad at counting people? Maybe. But a more interesting possibility is that politics has become unusually good at exposing the limits of human categorization.
And polling is hardly the only institution responding to complexity by creating finer categories. California lawmakers are considering SB 1387, which would require state agencies already collecting ancestry or ethnic-origin information to offer a separate category for Jewish ancestry or ethnicity. The proposal is voluntary for respondents, does not concern voter registration and requires personally identifying information to remain confidential.
The bill and the polling misses serve very different purposes but they simultaneously expose the same tension. Institutions are getting more precise about how they classify Americans, just as elections are showing how little those classifications can tell us about what Americans will actually do.
Modern polling cannot work without categories. Pollsters weight respondents by characteristics such as age, race, education and political affiliation because 800 randomly assembled people do not automatically resemble the electorate.
Wisconsin clearly showed the distinction.
Even Crowley’s own pollster had him nearly 20 points behind shortly before the election. The state’s open primary made the electorate harder to model, and late changes in the race gave voters additional reasons to move.
Political strategist and CNN pundit David Axelrod reduced the problem to one sentence, “It’s hard to know who’s going to show up.”
That is the limit no demographic refinement can eliminate because pollsters can know a respondent’s age, race, education, income, party preference and voting history. None of those categories, however, answers the question that ultimately decides an election: whether that person will vote and what choice they will make when they do.
California's SB 1387 offers an unusually timely example of the impulse toward greater demographic precision. The bill would require state agencies that already collect ancestry or ethnic-origin data to offer a separate Jewish category, after supporters argued that existing classifications fail to capture an identity that can encompass religion, ancestry, ethnicity and culture.
There are practical reasons for wanting better data. For example, if a government wants to measure discrimination or health disparities affecting a particular population, a category that fails to identify that population limits what the data can reveal.
That being said, there is a distinction between making people more visible in data, which is what SB 1387 seeks to accomplish, and making them more legible as individuals, which is where polling repeatedly gets into trouble.
Demographic categories can tell institutions something important about a population. The Midwest primaries are a reminder of what they cannot tell them: how the individuals inside that population will behave.
The more significant problem for pollsters is that demographic categories imply a degree of political stability that voters do not always possess. Race, age and education change slowly while candidate preferences can change in days.
Wisconsin offered an unusually sharp example: Crowley reentered a disrupted race, won Governor Tony Evers’ backing and benefited from a late push by Democrats concerned about Hong’s general-election prospects. Yet even Crowley’s own pollster had him nearly 20 points behind shortly before Election Day. The survey may have accurately described the electorate at one moment and still failed to anticipate what those same voters would do days later.
That is obviously a harder problem than demographic misclassification. Hong polled close to the share she eventually received; Crowley was the candidate whose support moved sharply in the closing stretch. The polls may therefore have missed something no weighting adjustment can fully solve: voters changing their minds.
Pollsters can improve sampling, turnout models and the way they weigh demographic groups but they can’t turn a fluid electorate into a fixed one. The more precisely institutions describe Americans, the easier it becomes to forget that the people inside those categories are still capable of moving.
AIPROPX is an independent multi-source news index — we track, compare, and connect coverage from across the web into one place you won't find anywhere else.