In this dissertation, we use laboratory experiments to study otherwise unobservable interactions in the labor market. The key advantage of laboratory experiments is the ability to control conditions more tightly than in any other context, so it plays a distinctive role in serving as the first link in a longer chain running from standard theory to actual outcome in the real world. Our experiments can even provide valuable information on how people behave in situations where existing theory provides little or no guide to what should happen.

In our first experiment, we study the impact of specific incentive schemes on people's behavior by systematically varying them. We create an experimental labor market where "workers" can join "companies" that pay them according to different compensation schemes: (a) piece rate, (b) revenue sharing, (c) individual tournament, and (d) team tournament. In order to disentangle incentive and self-sorting effects, our experiment forces all workers to initially complete real-effort tasks paid by four given incentives respectively, and then use compensating differentials to elicit their preferences for different incentives. Therefore, based on the lab data from our sample of Chinese university students, we are able to study their productivity response to various incentives as well as their preferences for different types of compensation.

When we analyze individual productivity under four incentives, we find that: (1) Compared to the baseline performance paid by piece rate, under three team-based incentives, more competitive incentive generates higher performance improvement. (2) Feedback about relative performance reduces the performance differences between three team-based incentives. (3) Regardless of incentives and feedback information, an additional compensation in terms of sign-up bonus brings a positive and significant effect on individual performance. In addition, by eliciting subjects' preferences for different compensation schemes, we build a mapping from individual characteristics to their self-sorting outcome as follows: (1) Subjects with high relative performance always prefer individual tournament to other two team-based incentives. (2) Risk-averse subjects are less likely to choose individual tournament if knowing the information about their relative performance. (3) Cooperative incentives attract more women than men, which is partially explained by gender-specific social preferences. (4) Compared to children with siblings, only children are less cooperative but more competitive. (5) In the absence of feedback, overconfident subjects are more likely to enter into individual tournament than those under-confident subjects with the same ability. Interestingly, the provision of information about their relative performance eliminates the impact of biased self-assessment. As a result, the feedback helps reduce the gender gap in competition as well as the difference between only child and child with siblings.

In a different study, we design a new laboratory experiment to investigate the ways that trust between strangers evolves in a setting where noisy feedback regarding mutual trustworthiness is present. We use a two-player sequential trust game where each trustor receives a sequence of noisy binary signals that reveal the trustworthiness type of the trustee. As a result, we track the evolution of trustors' individual beliefs about trustworthiness types of trustees to document that subjects process information in an asymmetric way compared to a perfect Bayesian: they react more to negative feedback rather than positive. We show that our empirical results arise naturally in a theoretical model where there exists a complementary relationship between initial trust and optimally biased Bayesian information processing. Hence, we theoretically predict that greater initial trust must be counter-balanced by more asymmetric belief updating. We then use a novel method to demonstrate this hypothesis in the following-up experiment. We match participants from two different universities (in Hong Kong and Beijing, respectively) and prime them on the social identity of their counterparts. Consequently, by the introduction of social identity, we find that both initial trust level and asymmetry of belief updating are stronger for in-group matches than out-group matches, which is consistent with our theoretical prediction.