CMU Researchers Find Amplification Of Bias In Advice To The Unidentified And Many
Professionals often give advice to many anonymous people. For example, financial analysts give public recommendations to buy, hold or sell stock, and medical experts formulate clinical guidelines that affect many patients.
New research from Carnegie Mellon University’s George Loewenstein and Duke University’s Sunita Sah demonstrates that advisers confronting a conflict of interest give more biased advice when there are multiple advice recipients as opposed to just one recipient, and in the case of just one recipient the advice is more biased when the adviser does not know the name of the recipient.
The findings, published in the journal Social Psychological and Personality Science, also show that an increased intensity of feelings toward single, identified recipients appears to drive the bias; advisers experience more empathy, and appear to have greater awareness and motivation to reduce bias in their advice, when the recipient is single and identified.
“Logically people should be more concerned about the advice they give to multiple recipients than to single recipients since it will affect the welfare of more people,” noted Sah, a post-doctoral associate at Duke’s Fuqua School of Business who worked on this research while completing her Ph.D. at CMU’s Tepper School of Business. “But, people feel more empathetic toward a single, identified, advice recipient, so they tend to put more care into the advice and behave less selfishly than they do if there are many recipients.”
Loewenstein, the Herbert A. Simon University Professor of Economics and Psychology within CMU’s Dietrich College of Humanities and Social Sciences, added, “It is a perfect example of how emotional reactions to situations can often drive us to do exactly the opposite of what logic would prescribe.”
Sah and Loewenstein conducted two experiments in which subjects, acting as advisers, gave advice to other subjects – “estimators.” Those playing the role of advisers viewed a 30 x 30 grid of dots, some filled and some clear, and gave advice to the estimator or estimators on the number of filled dots. Estimators had to estimate the number of filled dots in the large grid, but only viewed a 3 x 3 subset of the grid. The researchers created a “conflict of interest” between the two parties by paying the estimators more if their estimates were accurate but paying advisers more based on how much the estimators overestimated the number of filled dots.
In the first experiment, advisers were told the name and age of the single estimator for the “identified” condition, whereas no such information was provided for the “unidentified” condition, and the adviser only knew the estimator as “the estimator.” Advisers gave more inflated advice when they were not given the identifying information about the estimators, a result that is consistent with prior research showing that identification leads to greater sympathy toward a potential victim.
The second experiment repeated the identification manipulation of the first, and also compared advisers who gave advice to a single advice recipient (identified or not identified) or to a group of advice recipients (also identified or not identified). They replicated the results from the first experiment, and also found that advisers gave more biased advice to groups than to individuals, even though in the former case more people would be adversely affected by the biased advice.
This study is one of many collaborations between Loewenstein and Sah, including a 2010 paper published in the Journal of the American Medical Association that helps to explain how physicians rationalize accepting industry gifts.
On the Net: