Suppose 57 percent of a company’s employees drive to work. Someone making the case for more parking might call that a clear majority; someone arguing against it might say it’s barely more than half. Two experiments found that people who had a reason to make a quantity sound large or small were more likely to describe it in words than they were when accuracy was the goal. Audiences found those descriptions convincing even when they knew the sender had an agenda.

How incentives changed the way people reported a result

In the first experiment, the researchers used the Prolific platform to group 1,000 members of the public into 500 pairs. One person in each pair told the other about the results of simple games and observations. In a parallel experiment through the University of Warwick’s Policymakers Lab, 146 academics were paired with policymakers from five countries, most of them central government officials.

When accuracy was rewarded, senders chose words over figures only 13 to 14 percent of the time. An incentive to persuade raised the likelihood of choosing words by 25 to 29 percentage points. Experts trying to persuade also put more slant into their descriptions than into the numbers they reported. The authors suggest that a sender may be more comfortable choosing a phrase that favors one interpretation than giving a figure that is demonstrably wrong.

Warning recipients about the sender’s motive did little to blunt the persuasive effect, according to Victor de Chaisemartin, a doctoral candidate at MIT Sloan and a co-author of the study. The American Economic Review is publishing the paper, “Numbers Tell, Words Sell,” by de Chaisemartin, University College London economics lecturer Michael Thaler and University of Warwick associate professor Mattie Toma.

When words help readers interpret a number

De Chaisemartin traces the research to something he noticed in news coverage and corporate reports: figures often reach readers in the form of words. An earlier media study found that 22 percent of news items included statistics, but 24 percent of those items conveyed the quantities entirely in words. A separate study of academic writing found numbers in the abstracts of just 37 percent of papers.

A description need not be spin. If a policy raises seatbelt use by two percentage points, the gain might sound modest. But because most drivers already buckle up, it could represent a sizable share of the improvement still possible. In another setting, a change of the same size might be an ordinary fluctuation. De Chaisemartin sees a case for giving readers both the figure and the explanation.

Climate scientists use standard terms for likelihoods in some forecasts. De Chaisemartin suggests that doctors, researchers and security officials could adopt similar conventions. He also sees a lesson for readers weighing statistical claims in the news: a description can clarify what a number means, but the choice to describe it rather than report it may reflect an intent to persuade.

One question the experiments leave open is what happens when a sender addresses the same audience repeatedly. De Chaisemartin would like to test whether senders become more concerned about their reputation for accuracy or their audiences get better at judging the claims over time.