Newsletters

To Build Trust in Polling, We Must Consider Institutional Constraints

08/29/2024

To Build Trust in Polling, We Must Consider Institutional Constraints

Travis N. Taylor, Center for Excellence in Polling

In his recent book, Poll-arized, market researcher John Geraci reported that only 17 percent of voters trust pollsters.

This lack of trust is alarming for pollsters, many of whom see our job and our research as instrumental to democracy.

Americans, though, pay attention mostly to the polling covered in media reports. These public horserace polls supposedly predict an election outcome. But when those predictions are off—when the pollsters get it wrong—trust in our work is further diminished.

That makes me wonder why pollsters, even the good ones, insist on running the national presidential horserace and the generic congressional ballot and why the media champ at the bit to cover these polls.

These two types of polls may be entertaining for some, but they provide little in the way of predicting election outcomes because they ignore the institutional arrangements of America’s federal government.

Presidents, of course, are elected through the electoral college, not through a national popular vote. This American innovation leaves national presidential horserace polling unable to predict the outcome of the presidential election on its own.

One alternative is to focus on polling in swing states—those in which the outcome is less certain than it is in places like Oklahoma or California. The other is to take the national data a step further, using MRP to estimate state-level vote share in each state, the way Morris did at The Economist.

The generic congressional ballot possesses even less forecasting power. American members of Congress are not elected in a national vote the way legislators in parliamentary systems are. Candidates are elected from districts, most of which comprise only a portion of a state. By failing to account for both the district system and the redistricting process, the generic congressional ballot is unable to accurately predict election outcomes or eventual control of the House.

Unlike the presidential horserace, few remedies exist to correct the generic congressional ballot. Congressional district polling is, of course, possible, but it must focus on the precious few competitive districts around the country—a prospect that is quite costly. Advanced analytics are also possible but exponentially less useful than in the presidential horserace because of the granularity needed to estimate district-level vote share.

Because they ignore institutional arrangements and are not useful for predicting election outcomes and House control, presidential horserace polling and the generic congressional ballot may be contributing to the distrust of polling and pollsters.

After all, if we as pollsters cannot live up to our promises and our polls cannot fulfill their purposes, why should Americans trust us or our work?

So, what should we do?

We should stop running these polls and focus on the more meaningful alternatives discussed above. These polls can’t live up to the expectations people have for them, so all they can do is disappoint polling consumers. Disappointment does not breed trust.

Alternatively, if we insist on running these polls, we should caveat them heavily. As pollsters, we already caveat our work: “This is just a snapshot in time.” “When you consider the margin of error…” “Who knows what October surprises are in store?” To these polls, we should offer readers an additional warning: “This poll may not be predictive because it doesn’t account for institutional arrangements.”

The bottom line is this: Despite how accurate a national presidential poll or a generic congressional ballot may be, neither can reflect reality because they don’t incorporate the reality of institutional constraints. As public trust in polling wanes, we as pollsters owe it to our fellow Americans to bolster their trust in our product.