Confirmation Bias

What it is

The label covers a family of related habits. In his 1998 review in the Review of General Psychology, Ray Nickerson grouped them into several guises. People seek evidence that favours a hypothesis they already hold. They give extra weight to supportive evidence once they have it. They interpret ambiguous material in the favoured direction. They remember the past as more consistent with their view than it was. The review also follows the idea into practical settings, among them medical diagnosis, jury verdicts and the way scientific communities hold on to established theories.

Nickerson draws a line between two readings of the bias. In the motivated reading, people defend beliefs they value, so the pull comes from what they want to be true. In the unmotivated reading, people are partial even on questions they do not care about, so the pull comes from how a working hypothesis shapes the search and the reading. His review found evidence for both, and he argued that an adequate account has to cover both. He also noted that the term usually points to unwitting selectivity, which is why it is treated as a feature of ordinary thinking rather than a fault of character.

The idea has an older root in studies of perception. In 1954 Albert Hastorf and Hadley Cantril reported what students at Princeton and Dartmouth made of a football game played between the two teams in 1951. Students at each university counted more rule infractions by the opposing team when they watched a film of the game, and the authors concluded that the two groups had in effect seen different games. The study concerned loyalty and perception, and it gave later researchers a clear picture of two people holding the same material and keeping different records of it.

In effect

The most cited laboratory demonstration of uneven reading is the 1979 experiment by Charles Lord, Lee Ross and Mark Lepper. Participants with strong views for or against the death penalty read descriptions of two studies on whether it deters crime. One appeared to support their side and the other to cut against it. Each group rated the study that fitted its view as more convincing and better conducted, and each moved further in the direction of that study. The authors reported the result as attitude polarisation: after reading the same balanced material, the two sides ended further apart than they began. They explained it as a difference in scrutiny, with supportive findings accepted lightly and unwelcome ones taken apart.

Matthew Syed gives the idea an institutional setting in Black Box Thinking. He describes aviation as a system in which every accident is investigated independently and the findings are shared, which he calls an open loop. Where failures are explained away and the information never travels, he calls the loop closed. For the reason closed loops form he turns to Leon Festinger's work on cognitive dissonance: people tend to reframe evidence that threatens a belief about themselves rather than revise the belief, and the more senior and publicly committed the person, the harder the reframing. Read this way, confirmation bias is not only a feature of individual readers. It can be built into how an organisation handles its own results.

The practical implication in Syed's account is structural. An independent reader, and a habit of sharing what was found, separate the interpretation of evidence from the person who has the most at stake in it. The football and death-penalty studies point the same way, since their participants had no shortage of evidence and no reader without a side.

What it does not say

The term does not mean that people never change their minds, and it does not describe a flaw that disappears with intelligence or good intentions. The studies above show uneven treatment of evidence under particular conditions: strong prior views, material that bears directly on them, and readers who are asked to judge it.

It also does not establish that opposing evidence makes opinions harder. Andrew Guess and Alexander Coppock defined backlash as people who meet evidence against their view holding it more strongly afterwards, and tested for it in three survey experiments, two on online worker samples and one on a nationally representative sample. They found no sign of backlash, even in conditions set up to make it most likely, and concluded that it is much less common than the information-processing literature often implies. The 1979 experiment measured how convincing people found each report and where their attitudes ended up. The survey experiments asked whether counter-attitudinal information pushed views the wrong way. Uneven reading of evidence is well documented, and the stronger claim that opposing evidence hardens opinion has found less support.

Finally, the label should not be stretched over every case of disagreement. Two people can reach different conclusions from the same facts for reasons of values, access or definition, and the research on confirmation bias is about a specific pattern in how evidence is sought, read and recalled.

---