The backfire effect is when correcting misinformation hardens, rather than corrects, someone’s mistaken belief. It’s a relative of so called ‘attitude polarisation’ whereby people’s views on politically controversial topics can get more, not less, extreme when they are exposed to counter-arguments.
The finding that misperception are hard to correct is not new – it fits with research on the tenacity of beliefs and the difficulty of debunking.
The backfire effect appears to give an extra spin on this. If backfire effects hold, then correcting fake news can be worse than useless – the correction could reinforce the misinformation in people’s minds. This is what Brendan Nyhan and Jason Reifler warned about in a 2010 paper ‘When Corrections Fail: The Persistence of Political Misperceptions’.
Now, work by Tom Wood and Ethan Porter suggests that backfire effects may not be common or reliable. Reporting in their ‘The Elusive Backfire Effect: Mass Attitudes’ Steadfast Factual Adherence’ they exposed over 10,000 mechanical turk participants, over 5 experiments and 52 different topics, to misleading statements from American politicians from both of the two main parties. Across all statements, and all experiments, they found that showing people corrections moved their beliefs away from the false information. There was an effect of the match between the ideology of the participant and of the politician, but it wasn’t large:
Among liberals, 85% of issues saw a significant factual response to correction, among moderates, 96% of issues, and among conservatives, 83% of issues. No backfire was observed for any issue, among any ideological cohort
All in all, this suggests, in their words, that ‘The backfire effect is far less prevalent than existing research would indicate’. Far from being counter-productive, corrections work. Part of the power of this new study is that it uses the same materials and participants as the 2010 paper reporting backfire effects – statements about US politics and US citizens. Although the numbers mean the new study in convincing, it doesn’t show the backfire effect will never occur, especially for different attitudes in different contexts or nations.
So, don’t give up on fact checking just yet – people are more more reasonable about their beliefs than the backfire suggests.
Original paper: Nyhan, B., & Reifler, J. (2010). When corrections fail: The persistence of political misperceptions. Political Behavior, 32(2), 303-330.
New studies: Wood, T., & Porter, E. (in press). The elusive backfire effect: Mass attitudes’ steadfast factual adherence. Political Behaviour.
The news is also good in a related experiment on fake news by the same team: Sex Trafficking, Russian Infiltration, Birth Certificates, and Pedophilia: A Survey Experiment Correcting Fake News. Regardless of ideology or content of fake news, people were responsive to corrections.
Read more about the psychology of responsiveness to argument in my ‘For argument’s sake: evidence that reason can change minds’.
Mind Hacks 
I wonder if conclusions from studies of Mechanical Turk participants can be generalized to the population as a whole. Reading through the study (well, skimming) there was no discussion of the limitations of Mechanical Turk: to develop a demographically meaningful panel, how the mechanics of price and timing impact participation, how the context of “doing work” changes results (Except for A.14.3 which doesn’t look at MT as a unique platform).
Ultimately disappointing.
That is a limitation, but no more of a limitation than using University participant pools. What reasons are there for suspecting the results wouldn’t generalise? Obviously no one study is without limitations, but this results fits with other results which suggest the backfire effect is rarer than has been claimed
I’ve always found generalizing off Uni participant pools limiting, too. Anyway, my question led me down a fun rabbit hole: http://www.behind-the-enemy-lines.com/2015/04/demographics-of-mechanical-turk-now.html
Isn’t it easier to verify that university students are not bots? It’s hard to believe someone hasn’t coded truly mechanical Turks to answer such surveys.
You can put in catch questions and find that levels of satisfycing (ie where participants just try and click through questions) are shockingly high in University samples. M Turk samples aren’t higher in this behaviour, but maybe they have more experience with the catch questions
Turk samples are probably more representative of the population than most other studies that use purposive sampling. Regardless, generalization is speculative!
Woods and Porter mention that some polarized topics are prone to the backfire effect. So, contrary to a line in this post, backfire effects may be “common” and “reliable” after all. They are only uncommon or unreliable in factual domains — according to Woods and Porter. Non-factual domains are supposed to be prone contractual effect, according to Woods and Porter.
But which non-factual domains? All of them? Only some? Which ones? Is it that only, sthe only non-factual domains that are crucial for identity preservation are prone to the backfire effect? (E.g., domains in which arguments and evidence challenge the non-factual beliefs that one holds as a self-proclaimed Democrat/Republican, Atheist/Theist, etc.) Are other non-factual domains prone to the backfire effect?
(I find myself wanting clarification about the supposed meaning of ‘factual’ here, but perhaps that takes us too far afield.)
I take your point that we don’t know from a demonstration of lack of backfire effects in these studies how common backfire effects may be but I think it is fair to call them unreliable/elusive – in that the papers linked here test for them in circumstances for which we might reasonably expect them (politised statements bearing on the ideological position of the participants) and no backfire effects were found. That’s new information, even if it doesn’t prove the effect never holds
Thanks! Please do keep posting on this subject. But the link to the Wood & Porter paper is wrong: it should be https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2819073
Thanks! Corrected!
As a computer scientist, I’m definitely not an expert in the field. However I’m surprised that Wood and Porter do not cite the results of Skurnik et al, 2005 or Peter and Koch, 2016: they show that the backfire effect is more robust after several days.
It is a pity Wood and Porter didn’t investigate the effect of a delay in their experiments. Maybe we should say that the *immediate* backfire effect is elusive. But what about the longer term?
Or am I missing something?