Economic Theory (2005)DOI 10.1007/s00199-005-0037-7SPECIAL ISSUE ARTICLEYaw Nyarko · Andrew SchotterBarry SopherOn the informational content of advice:a theoretical and experimental studyReceived: 15 November 2004 / Accepted: 2 September 2005© Springer-Verlag 2005Abstract This paper examines the market for advice and the underlying perceptionthat advice is useful and informative. We do this by first providing a theoreticalexamination of the informational content of advice and then by setting up a seriesof experimental markets where this advice is sold. In these markets we providebidders with a demographic profile of the “experts” offering advice.The results of our experiment generate several interesting findings. The raw biddata suggest that subjects bid significantly more for data than they do for advice.Second, in the market for advice there appears to be no consensus as to who are thebest advisors although on average economists demand the highest mean price andwomen suffer a discount. In addition, we find that whether a subject suffers from arepresentativeness bias in the way he or she processes data has an impact on howhe or she bids for advice and on his or her willingness to follow it once offered.Finally, we find that on average people impute a low level of informativeness ontoadvice, consistent with their bidding behavior for data versus advice.Keywords Advice · Decision making · Risk aversion · ExperimentsJEL Classification Numbers D81 · G11 · C91This work was done under grant number SES-0425118 of the National Science Foundation. Theauthors would like to recognize the Center for Experimental Social Science at New York Uni-versity for its additional support. We also acknowledge the help of Elizabeth Potamites for herresearch assistance.A. Schotter (B) · Y. NyarkoDepartment of Economics, NewYork University, 269 Mercer Street,NewYork, NY 10003, USAE-mail: [email protected]. SopherDepartment of Economics, Rutgers University, 75 Hamilton St., NJ Hall Room 202,New Brunswick, NJ 08901-1248, USAY. Nyarko, A. Schotter, and B. Sopher1 IntroductionIt is commonly thought that a picture is worth a thousand words. If that is so, onemight ask how much data is a piece of advice worth. In other words, if advice isimportant then we should be able to measure it in two ways: how much data woulda rational decision maker be willing to give up in order to receive a piece of advicefrom a person who has just engaged in the decision problem he or she is about toengage in or, alternatively, how much would that person be willing to pay for suchadvice from a person with a given set of characteristics.The fact that we expect that people will bid different amounts for advice fromdifferent types of people implies that in the “market for advice” certain types ofpeople are likely to fare better than others. Such markets for advice and influencemight function in a number of ways. Under one scenario there may be a perceptionthat certain people or types of people are worth listening to. These perceptionsamount to broad stereotypes that may bestow huge rents on some of the agents inthe market.Such stereotypes, if theypersist, can lead to what we will call “perception rents”,i.e. amounts paid for the advice of agents in excess of the expected informationalcontent contained in their opinion. If such rents are substantial, they present us witha potentially large inefficiency. An alternative to perception rents is what we willcall the “chauvinistic bias”. Here people tend to believe that advice from peoplelike themselves is the best and hence tend to bid higher amounts for people withcharacteristics like theirs whether or not those types give the best advice.A related question deals with the impact of what are sometimes called “repre-sentativeness” and “conservative” biases, relative to Bayesian updating, and theirimplications for the process of advice giving and following. For example, whenupdating one’s beliefs, a rational Bayesian decision maker is expected to placea certain amount of weight on his previous prior (or the base rate) and a certainamount on new information (the sample) as it arrives. How much weight is placedon the new information depends on the strength of his or her prior. If a decisionmaker places more than the Bayes-optimal weight on the prior (or base rate) heor she is called “conservative” while if excessive weight is placed on the samplehe or she is considered to be subject to the “representative” bias, thinking, in thelimit, that the sample received is in some sense representative of the populationfrom which it was drawn. Such people fail to take base rates or priors sufficientlyinto account.These concerns have wide ranging implications for our research on advice giv-ing and following. For example, if we could measure the degree to which a decisionmaker is subject to one of the biases discussed above, could we correlate that char-acteristic to the decision maker’s willingness to pay for and follow advice. Moreprecisely, if conservatives are reluctant to update their priors on the basis of newinformation, are they therefore less inclined to pay for advice and also follow itonce it is given?Also who are more persuadable, conservatives or representatives?All of these concerns can be summarized under the name of “advice bias”.In this paper we study an experimental market for advice in an attempt to mea-sure both the informational content of advice and the market for it. To do this wecreate a set of “experts” by having some of our subjects get experience playinga simple 2×2 “game against nature” a large number of times. These experts areA theoretical and experimental studysurveyed to obtain information about their gender, GPA, major, and year in school.Advice is elicited from these experts and is then sold to a new set of subject “cli-ents” who play the game once and only once. The prices generated by this marketfor advice furnish us with an opportunity to measure potential perception rents inthe market. In addition, by observing the decisions made by the client subjects,we are able to measure whether these perception rents are wasteful or not. Byobserving the way subjects update their priors in the experiment, we are able tocategorize them along the conservative-representative spectrum according to howmuch weight they put on their prior and how much they place on the data sam-ple they observe. We are then able to