Confirmation bias is a pervasive phenomenon. Whether it’s a news article or an additional piece of scientific data, people tend to interpret new information as evidence confirming their prior beliefs—even if it actually doesn’t.
“People have been aware of this phenomenon for a very long time,” says Tobias Donner, a neuroscientist at University Medical Center Hamburg. “The problem is that it’s not understood why this phenomenon is there and what brings it about.”
In a new set of experiments, published in Current Biology, Donner and a team of researchers suggest that confirmation bias arises because people tend to pay more attention to information that’s consistent with their prior belief. “Attention seems to be the decisive mechanism that brings about this confirmation bias,” Donner says.
The results, he says, is a step toward pinpointing the neural mechanism behind confirmation bias. Ultimately, understanding its underpinnings could help mitigate this bias, whether in the realm of business, politics, or everyday decision-making.
Most of the previous research on confirmation bias has been based on experiments using higher-level reasoning. In this study, the researchers analyzed a lower-level task involving visual perception. “The fact that we observe this confirmation bias even in these low-level decisions tells us it comes from very deep in the decision-making machinery in our brain,” Donner says.
The researchers asked 16 volunteers to look at a cloud of dots on a computer screen. The dots would rapidly move in somewhat random directions, and the volunteers had to determine whether the dots were generally going in a clockwise or counter-clockwise direction. Then, the dots would move again and the volunteers had to move the cursor to align with the angle they thought the dots were moving on average in both cases, relative to a reference line.
The estimated angle, the researchers found, tended to be consistent with whichever direction the volunteers chose in the first part of the task. This was true regardless of which direction the dots were actually moving, demonstrating confirmation bias.
Next, by statistically analyzing the data across thousands of trials, the researchers measured how much the second viewing of moving dots influenced the volunteers’ final decisions. The influence, they found, was greatest when the estimated angle was consistent with the initial decision of whether the dots were moving clockwise or counter-clockwise.
Using a mathematical model of the data, the researchers showed that when the final and initial decisions were consistent, the brain effectively amplified the influence of the second viewing. When the final and initial decisions were not consistent, the brain seemed to suppress the influence of the second viewing.
Other research, including work in monkeys, has suggested this type of amplification and suppression when explaining how attention works in the brain, notes Jaime de la Rocha, a neuroscientist at IDIBAPS in Barcelona who wasn’t involved in this study. When asked to pay attention to a particular feature—say, a person with a red hat—certain neurons become more attuned to red hats, amplifying the perception of red hats, and suppressing the perception of, say, blue hats. Similarly, by first deciding that a bunch of dots are moving clockwise, the subject may be more attuned and pay more attention to clockwise-moving dots rather than other dots moving differently.
But while the new analysis is careful and compelling, it may be too early to conclude that attention is at the heart of confirmation bias. “It’s certainly a plausible mechanism,” says Alan Stocker, a computational psychologist at the University of Pennsylvania who also was not involved in this study. But to determine exactly what’s going on, he says, you would need a model that explains how such behavior arises, and not just one that describes the data, as is the case in this study.
The researchers found similar results involving a more complex task, in which they asked volunteers to determine the average value of a series of numbers. It’s less certain, however, how much selective attention can explain confirmation bias in higher-level processes like evaluating numbers, de la Rocha says. There hasn’t yet been much evidence linking attention to such tasks.
Still, he says this is one of the first studies linking confirmation bias with any mechanism, and could therefore offer insights into how to reduce the harm from broader instances of bias—even if we’re still a long way from fully understanding how a social media feed biases political opinions. “You could say this is a first step to understanding what’s going on at that level,” he says.