Introduction
Experimental Design
Preliminary results
Conclusion
Reference group neglect: an experimental test Noémi Berlin1 1 Université 2 Social
Marie-Pierre Dargnies2
Paris 1, Paris School of Economics
Science Research Center Berlin (WZB)
November 17 2011
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role We are interested in:
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role We are interested in: how people update their beliefs following the reception of a performance feedback
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role We are interested in: how people update their beliefs following the reception of a performance feedback how people adjust their competitive entry to their beliefs about their relative performance and the level of the competition
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role We are interested in: how people update their beliefs following the reception of a performance feedback how people adjust their competitive entry to their beliefs about their relative performance and the level of the competition whether men and women and low-performing and high-performing subjects are different in this respect
Introduction
Experimental Design
Preliminary results
Conclusion
Motivations Men are known to have a higher taste for competition than women (Niederle and Vesterlund (2007), replicated many times) Determinants of choice to enter a competition are still not fully understood Beliefs and the way they are updated are often suspected to play an important role We are interested in: how people update their beliefs following the reception of a performance feedback how people adjust their competitive entry to their beliefs about their relative performance and the level of the competition whether men and women and low-performing and high-performing subjects are different in this respect We build an experimental design where we carefully elicit beliefs both before and after we provide subjects with a feedback on how they did relative to others so that we are able to track how they are updated We also manipulate the level of the competition one evolves in
Introduction
Experimental Design
Preliminary results
Conclusion
Related literature
Camerer and Lovallo (1999) find evidence suggesting people are subject to "reference group neglect". Namely, in their experiment, excess entry in a competition is much larger when subjects self-selected themselves into the experiment knowing payoffs would depend on skill. Subjects seem to ignore the fact that all subjects they are competing against also think that they are skilled.
Introduction
Experimental Design
Preliminary results
Conclusion
Related literature (2) Wozniak (2009) shows that giving feedback about past relative performance removes the gender difference in tournament entry as: High ability women choose more competitive compensation schemes. Low ability men choose less competitive compensation schemes.
Möbius et al. (2011) provide the subject a noisy feedback as a simple binary signal whether her performance is among the top 50%. They find that: Conservatism: subjects update less than Bayesian agents would in response to both negative and positive information and women are more conservative than men. Asymmetry: Subjects adjust more to positive than negative information (no gender difference).
Introduction
Experimental Design
Preliminary results
Conclusion
Experimental design
Task= additions of five two-digits numbers (Niederle and Vesterlund, 2007). Step 1 piece rate (PR) remuneration : 5 minutes to solve as many additions as one can (0.5e/correct addition). Step 2 tournament (T): winner if performance is above the performance of a randomly chosen teammate (1e/correct addition). Incentivized belief-assessment questions (1): Elicit beliefs (in %) about belonging to each quartile.
Introduction
Experimental Design
Preliminary results
Conclusion
Incentivized belief-assessment questions (Möbius et al. 2011 )
Elicit beliefs about belonging to each quartile. What is your percentage chance of scoring in the top quartile? Let’s say your answer is x. Then the computer randomly picks a number y between 0 and 100. If y ≤ x, you earn 1e if your step 2 score belongs to the 1st quartile, otherwise you earn nothing. If y > x, you earn 1e with y % probability, with 100 − y % probability, you earn nothing.
Introduction
Experimental Design
Preliminary results
Conclusion
Experimental Design (2)
Step 3: Choice between Piece Rate and Tournament (Choice 1). If tournament is chosen: winner if performance is above the step 2 performance of a randomly chosen teammate.
Feedback: indicating to the subject if she is either below or above the median (based on step 2 performance). Incentivized belief-assessment questions (2): Elicit beliefs about belonging to each of the two possible quartiles.
Introduction
Experimental Design
Preliminary results
Experimental Design (3)
Step 4: Choice 2 (Control or Treatment) Control Choice between PR and Tournament
Treatment Choice between PR and ability grouping* Tournament.
⇒ *Ability grouping tournament: A subject competes against someone having the same performance level (below or above the median).
Conclusion
Introduction
Experimental Design
Preliminary results
Conclusion
Experimental Design (4): Control
The purpose of this control group is to get a benchmark of the participants’ decisions to enter a tournament when they already made a similar decision just before and have received a feedback on whether their previous performance was below or above the median.
Introduction
Experimental Design
Preliminary results
Conclusion
General information about the experiment
Run in Paris (LEEP). Same number of men and women in each session Control: 112 subjects (56 men, 56 women) Treatment: 116 subjects (58 women, 58 men) Remuneration: One step randomly chosen at the end of the experiment + belief-assessment questions + 7e show-up fee. Average payoff: 15.3e.
Introduction
Experimental Design
Preliminary results
Conclusion
Performance 46,5% of men and 41,2% of women are above the median (Fisher’s exact test: p=0.50) No gender differences in the distribution across quartiles (Chi-square test: p=0.37) No gender differences in step 2 performance (Mann Whitney test: p=0.49)
Introduction
Experimental Design
Preliminary results
Conclusion
Beliefs Men are more confident than women
Before receiving the feedback: Men and women think their performance is respectively 65,5% and 57,8% likely to be above the median (p=0.02) Among subjects who are below the median: 56,7% for men vs 54,6% for women (p=0.70) Among subjects who actually are above the median: 75,6% for men vs 62,4% for women (p Men are not so much affected by the information they get on their performance level but react more strongly to the level of their competitors.
Introduction
Experimental Design
Preliminary results
Conclusion
Concluding remarks Women update their beliefs to the performance feedback more strongly than a bayesian agent would but not men. Low-performing participants adjust to the level of the competition while high-performing participants do not. "Surprise effect": Women react more strongly to the feedback when they did not expect it. While women are especially sensitive to the information on their own performance level, men react more strongly to the level of their competitors. Many things left to do: see how men and women act on their beliefs, consequences of competitive choices on welfare: who loses money by making the wrong choice and how much is lost?...
Introduction
LP
HP
Experimental Design
Women Men Diff Women Men Diff
Control Choice1 Choice2 24,1 6,9 58,8 55,8 p=0.000 p=0.000 40,7 68,2 p = 0.007
63 72,7 p = 0.3
Preliminary results
Diff p = 0.01 p = 0.73 p=0.02 p=0.64
Traitement Choice1 Choice2 34,2 47,4 55,6 85,2 p=0.015 p=0.000 45 71 p = 0.009
50 67,7 p = 0.07
Conclusion
Diff p=0.098 p=0.000 p=0.65 p=0.7