‘Which Corrections Work?’: Sources, Causes and Negations

Brendan Nyhan and Jason Reifler’s recent paper, Which Corrections Work?, sums up some of the major psychological forces that impede news outlets as they try to correct misperceptions and false beliefs.

The title perhaps undersells this paper, which doesn’t just pertain to the corrections that news outlets issue about their own stories (although that is certainly part of the authors’ concern). Instead, Nyhan and Reifler are concerned with media attempts to counteract false notions, more generally. (They say the paper is a summary of new research, as well as recommendations, for “journalists, educators and civil society groups who hope to counter the influence of false or misleading claims.”)

The paper addresses three phenomena that further false beliefs: unpersuasive sources, lack of corrective causal explanations, and negations’ tendency to reinforce the claims they are trying to negate. Nyhan and Reifler previously discussed these forces in Misinformation and Fact-Checking: Research Findings from Social Science, the guide for media professionals that they published in 2012. In the new paper the authors add a summary of results from survey experiments conducted between fall 2012 and spring 2013. In most cases these experiments strengthen the prevailing hypotheses.

In one case, the authors sought to extend the hypothesis that using a negation, such as “John is not a criminal,” can actually reinforce the claim that the negation was meant to debunk (ie, the claim “John is a criminal.”) Nyhan and Reifler point out that previous studies examined the formation of new impressions, not the effectiveness of corrections. The pair’s study, however, found no statistically significant variation between participants told that a former congressman was “not guilty” versus those told he was “exonerated.” Nyhan and Reifler don’t offer any reflections on why this study failed to replicate previous findings, though they still advocate that it is safer to use the affirmative language (such as “exonerated”).

It is notable that the study design does not obviously simulate an instance of correction. At no point were the participants told that the ex-congressman was guilty. But (and this is my hypothesis, not Nyhan and Reifler’s) it is arguable that the very appearance of articles describing accusations tends to paint parties in a “guilty” light, and therefore journalists writing acquittal stories may need to choose language that minimizes the possible reinforcement of miconceptions.

Here are the author’s three main recommendations, and a few thoughts on each:

Avoid citing sources who may be seen as partisan or ideological: This seems sound advice, where applicable. In some (many?) situations it will be difficult. Often the “partisan” think tank is the only one researching a given issue, because only those who share its political persuasion see the topic as worth pursuing. Indeed, by devoting resources to the topic, the think tank may then reinforce belief in its partisan nature. (“Oh, they research climate change, they must be in bed with the Democrats.”)

Offering alternative causal explanations: In some situations, this rule will be applicable and should be followed. But it should come with a heavy caveat: reporters should only offer an alternative causal explanation where one is known to be true, beyond a shadow of a doubt. Otherwise, they risk planting still more misinformation in people’s minds, and this further misinformation will prove just as difficult (or more so?) to dislodge. The point may seem obvious, but our media has shown itself to be a sort of addict, unable at times to resist creating those little morsels of causal speculation (was x to blame for y? the cable news personality asks. Probably not, but the very fact they’ve asked it could cause people to remember it as so.)

Using affirmative language, such as “John was exonerated,” in place of negating language, such as “John was found not guilty.” One issue with this strategy is that a positive word is not always available. An example that Nyhan and Reifler themselves use is “John is not a criminal” – it’s hard to think of a positive word or phrase that could be used instead, without risking editorializing. (“John is a law-abiding citizen?”) Another issue is that this approach could require news outlets to use less accessible language, as in the case of saying “John was exonerated.” Most editors probably won’t see “exonerated” as a particularly difficult work, but considering that 30 million Americans read at or below a 3rd-grade reading level (according to ProLiteracy), it’s a word whose use immediately puts the sentence beyond the comprehension of a big slice of the country.

Many editors won’t see this problem as particularly their responsibility – surely their word choices shouldn’t be held hostage to the country’s social and institutional failings? I would argue (and plan to argue more comprehensively, in future posts) that by making it their business, media outlets can make themselves relevant again. And when news outlets make comprehensive style and policy changes to incorporate the latest findings on the psychology of misinformation, they should not do so at the expense of readability.

In short: Much work remains to be done – this area of research is still new, and discussion about how to apply it practically is nearly non-existent. Nyhan and Reifler are to be commended for their work, not just to increase the base of empirical evidence, but to communicate this to media professionals and seek to advance the conversation.


One Reply to “‘Which Corrections Work?’: Sources, Causes and Negations”

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s