110 likes | 285 Views
“ To Belong or to Be Different? Evidence from a Large-Scale Field Experiment in China ” by Monic Sun, Xiaoquan Zhang, Feng Zhu. Michael J. Dickstein Stanford University NET Institute Conference on Network Economics June 7, 2013. Outline. Experimental Design Simple model of behavior
E N D
“To Belong or to Be Different? Evidence from a Large-Scale Field Experiment in China ”by Monic Sun, Xiaoquan Zhang, FengZhu Michael J. Dickstein Stanford University NET Institute Conference on Network Economics June 7, 2013
Outline • Experimental Design • Simple model of behavior • Comments on empirical approach • Summary data • Implications of findings Michael J. Dickstein, Stanford University
Experimental Design • Key Idea 1: Escape the classic “reflection” problem described by Manski (1993) • “reflection” problem: difficulty of inferring whether the average behavior in a group influences the behavior of individuals that comprise that group • Authors’ solution: randomize information on peers Michael J. Dickstein, Stanford University
Experimental Design • Key Idea 2: Use data from choices on a social network platform; agent’s decisions are subject to evaluation by peers • Key Idea 3: Agents shown message “Most popular color is X with Y%”, where X = color not equal to initial choice • Key Idea 4: Randomize whether the experimental message includes a reminder that choice is public Michael J. Dickstein, Stanford University
Simple model of behavior • Utility • Current (implicit): “Best color” depends on preference for convergence/divergence • Value from color of virtual home • Bayesian updating • Prob0,c(“best color”) -> choice • Prob0,c(“best color”) + signal -> Prob1,c(“best color”) -> choice • Identity of “best color” transmitted in signal may matter Michael J. Dickstein, Stanford University
Simple model of behavior • Specifically: • If new information comes close to priors, leads to little adjustment (no switch). • If new information differs from priors, may lead to switch • In simple Bayesian setting, increasing the percentage of the majority color affects the updating process in a monotonic way Michael J. Dickstein, Stanford University
Simple model of behavior • But, we observe a non-monotonic probability of convergence as the rate of adoption increases • Recent work: “Ego” utility consequences, Eiland Rao (2011): • Subjects receiving negative feedback on their intelligence and beauty were far less predictable in their updating behavior and exhibited an aversion to new information • In response to good news, inference more closely followed to Bayes’ Rule (as in the neutral experimental condition) Michael J. Dickstein, Stanford University
Comments on Empirical Approach • Data • To external validity: who are the users of the Virtual Homes app and Kaixin more generally? Likely to be least conformist? Michael J. Dickstein, Stanford University
Comments on Empirical Approach • Data • To external validity: who are the users of the Virtual Homes app and Kaixin more generally? Likely to be least conformist? • Is there a natural taxonomy to the colors {yellow, green, pink, blue, red, gray}? • Unconditional choice probabilities at initial choice • Correlations between initial choice and individual characteristics? Michael J. Dickstein, Stanford University
Comments on Empirical Approach • Data • To external validity: who are the users of the Virtual Homes app and Kaixin more generally? Likely to be least conformist? • Is there a natural taxonomy to the colors {yellow, green, pink, blue, red, gray}? • Unconditional choice probabilities at initial choice • Correlations between initial choice and individual characteristics? • Balance in the demographic covariates by randomized majority share? Michael J. Dickstein, Stanford University
Comments on Empirical Approach • Implications? • Validity for payoff-relevant choices? • Ex. Physicians choice of treatment regime • Exploit decisions after receiving signal to identify “non-conformists”. Valuable for tailoring marketing campaigns. Michael J. Dickstein, Stanford University