Motivated reasoning

You drink four cups of coffee a day. A news story reports a study linking heavy coffee drinking to worse sleep. You look up the study, notice it relied on people’s own sleep diaries, and decide it doesn’t show much. A week later a friend sends a story about a study finding no link, and you share it without opening it.

Your criticism of the first study may be perfectly fair: sleep diaries are an imperfect measure. What marks this as motivated reasoning is the asymmetry. The unwelcome study had to survive a search for flaws, and the welcome one didn’t have to survive anything. The test is whether you’d have read the methods of the second study as closely if its result had been the one you didn’t want.

• A theory of how it works. Ziva Kunda’s 1990 review distinguished accuracy goals (wanting to be right, whatever the answer) from directional goals (wanting a particular answer). She argued that both work by influencing which beliefs and strategies people apply, and that directional goals bias the search of memory and the construction of arguments. But people “do not seem to be at liberty to conclude whatever they want to conclude merely because they want to”: they draw the desired conclusion only if they can muster the evidence for it, maintaining what she called an “illusion of objectivity”.
• Doubting unwelcome evidence. Kunda describes her own 1987 study in which people read an article claiming caffeine was risky for women. Women who consumed a lot of caffeine were less convinced than women who consumed little. Men showed no such difference, and the women showed it only when the health risk was described as serious.
• Demanding more of unwelcome evidence. In Peter Ditto and David Lopez’s 1992 studies, participants tested their saliva with a strip of paper (in fact plain yellow construction paper, which never changed color) for a made-up enzyme condition said to raise the risk of pancreatic disorders later in life. Some were told that no color change meant they had the condition; others, that it meant they didn’t. Those given the unwelcome reading took longer to decide the test was finished, were more likely to retest (13 of 25, against 4 of 22), and rated the test as less accurate. A third study held constant how unexpected the result was, and those given the unwelcome result again cited more irregularities in their lives that might have affected the test. The authors call this motivated skepticism: less information is needed to reach a preferred conclusion than a non-preferred one.
• Choosing what to read. A 2009 meta-analysis by William Hart and colleagues found a moderate preference for information that supports what people already think, believe or do over information that challenges it (d = 0.36). The preference was weaker when the available information was low in quality, and it reversed when the challenging information was useful for a goal people currently had.

Nicholas Epley and Thomas Gilovich sum up the picture: motivated beliefs come from “reasoning in the service of some self-interest, to be sure, but reasoning nonetheless”, and the result is “biased beliefs that feel objective”.

• Confirmation bias is favoring what you already believe, and much of it happens with nothing at stake: a hypothesis you are merely testing can steer your search. Motivated reasoning is driven by wanting an answer, which may or may not be what you already believe. Ditto and Lopez’s participants had no prior belief about a made-up enzyme.
• Belief bias is accepting weak arguments for believable conclusions. It appears even when nobody cares about the conclusion, which is what separates it from motivated reasoning.
• Self-serving bias, crediting yourself for success and circumstances for failure, is one of the patterns motivated reasoning is used to explain. Ditto and Lopez presented their results as evidence about a core part of it.
• Appeal to consequences is an argument: “it would be terrible if this were true, so it isn’t.” Motivated reasoning is a process that can reach the same place without anyone making that argument out loud.
• Backfire effect was proposed as an extreme form, in which counterarguing against a correction leaves a belief stronger. Large replications have rarely found that extreme form, and its failure doesn’t undo the milder findings above.
• Bias blind spot: people tend to rate themselves as less biased than others, which makes motivated reasoning easier to spot in someone else than in yourself.

The central question is the symmetry test: “Would I accept this evidence, or dismiss it, if it pointed the other way?” Deciding in advance what would change your mind, before seeing the result, takes the preference out of the standard.

Kunda reviewed studies in which people who expected to be evaluated or to justify their judgments showed smaller biases, which suggests that raising the stakes of being wrong can help. She also noted that those studies hadn’t tested what happens when people want a particular answer at the same time, so that shouldn’t be treated as a proven remedy.

• Unequal scrutiny isn’t automatically biased. If a claim contradicts a lot of what you know, it is reasonable to check it harder: an unexpected result is more likely than an expected one to come from an error. Kunda noted this problem in older studies, where people threatened by evidence may also have had different prior beliefs about it. Good studies in this area, like Ditto and Lopez’s third study, are designed to separate wanting an answer from expecting one.
• The political research is disputed. Much recent work on motivated reasoning concerns politics, and there the separation is hard. Ben Tappin, Gordon Pennycook and David Rand argue that the two most common experimental designs are often confounded, because political identity goes along with prior beliefs, and discounting information that conflicts with your beliefs, or a source you distrust, can be rational. They are explicit that this doesn’t mean reasoning is unaffected by motivation, only that these designs often can’t show that political motivation is the cause. James Druckman and Mary McGrath argue similarly that the evidence on one politically divided topic is equally consistent with people who try to be accurate but differ in what they consider credible evidence.
• Some specific findings haven’t held up. Kahan and colleagues reported that people who were better with numbers did better at interpreting data about a skin rash treatment, but polarized more, not less, when the same data concerned a divisive policy. A large preregistered replication by Emil Persson and colleagues didn’t find good evidence for that pattern: politically congenial answers appeared, but mostly weren’t amplified by numeracy the way the theory predicted.
• It can explain anything after the fact. Pointing out that someone wants a conclusion doesn’t show that their reasoning for it is bad, or that the conclusion is false. Dismissing an argument because of the motive behind it is the Genetic fallacy; the reasoning still has to be checked on its merits.