Biases and decision-making

Recently, I wrote a blog post (www.fcw.com/ blogs/thelectern) about a class on decision-making biases that I added last fall to my introductory public management course in the master’s program at Harvard University’s Kennedy School of Government. The example I gave in the blog was overconfidence. Studies show that people tend to assume their judgment is better than it is, which reduces the need they feel to gather more information before making decisions.

Intrigued by the post, a reader asked if I could write a column expanding on the academic research on decision-making biases. Research in that area is a growth field. The basic approach is to give people decision problems in a lab simulation or survey and see how they deal with them.

One bias we discussed in the class was the confirmation bias — the tendency of people to seek evidence confirming their existing beliefs rather than evidence that might contradict those beliefs.

That bias was first demonstrated in an experiment in which subjects were selected based on their opinions on the death penalty. All subjects were then presented with the same two research studies, one concluding that the death penalty deterred homicide and the other that it didn’t. Researchers found that subjects already in favor of the death penalty judged that the research methodology and analysis of the study concluding the death penalty had a deterrent effect were superior to the one that didn’t, while those already against the death penalty concluded the opposite.

Another fascinating example is referred to as the anchoring bias, in which some starting point for a person’s deliberation, even if arbitrary, influences the judgment he or she finally makes.

We illustrated that bias in class by having students complete a survey that asked their judgment about how great the incidence of serious accounting fraud was among large U.S. corporations. However, before being asked that question, half the students were asked whether they thought the incidence was more or fewer than 200 firms in 1,000, and the other half were asked whether they thought it was more or fewer than 10 in 1,000.

That pre-question served as an anchor for the judgment students later made. Those initially asked whether the incidence was more or fewer than 200 firms in 1,000 estimated that fraud was more than twice as high as the students given the other anchor.

Biases arise for good reasons. Sometimes they allow decision-making shortcuts that generally serve us well, and at other times, they evolve from valuable human traits — for example, overconfidence is a problem arising out of the valuable trait of self-confidence. Therefore, debiasing is hard. The next research frontier is to determine whether there are ways to eliminate biases and improve decisions.