TY - JOUR AU - Bundorf, Kate AU - Polyakova, Maria AU - Tai-Seale, Ming TI - How do Humans Interact with Algorithms? Experimental Evidence from Health Insurance JF - National Bureau of Economic Research Working Paper Series VL - No. 25976 PY - 2019 Y2 - June 2019 DO - 10.3386/w25976 UR - http://www.nber.org/papers/w25976 L1 - http://www.nber.org/papers/w25976.pdf N1 - Author contact info: Kate Bundorf Sanford School of Public Policy Duke University Durham, NC 27708 E-Mail: kate.bundorf@duke.edu Maria Polyakova Center for Health Policy Stanford School of Medicine Encina Commons, Room 182 615 Crothers Way Stanford, CA 94305 Tel: 650/498-7528 E-Mail: maria.polyakova@stanford.edu Ming Tai-Seale School of Medicine, Family Medicine and Public Hea University of California San Diego 9500 Gilman Dr La Jolla, CA 92093 E-Mail: mtaiseale@ucsd.edu M3 - presented at "Economics of Artificial Intelligence", September 24-25, 2020 AB - Algorithms are increasingly available to help consumers make purchasing decisions. How does algorithmic advice affect human decisions and what types of consumers are likely to use such advice? We conducted a randomized, controlled trial comparing the effects of offering personalized information, either with or without algorithmic expert recommendations, relative to offering no personalized information for consumers choosing prescription drug insurance plans. Treated consumers were more likely to switch plans and to choose a plan that lowered their total spending on drugs. The behavioral response was more pronounced when information was combined with an algorithmic expert recommendation. We develop an empirical model of consumer choice to examine the mechanisms by which expert recommendations affect choices. Our experimental data are consistent with a model in which consumers have noisy beliefs not only about product features, but also about the parameters of their utility function. Expert advice, in turn, changes how consumers value product features by changing their beliefs about their utility function parameters. We further document substantial selection into who demands expert advice. Consumers who we predict would have responded more to algorithmic advice were less likely to demand it. ER -