Peer appraisals

Manish Gupte, Purdue University


My dissertation consists of three essays and studies different aspects of peer appraisals. The abstracts of the three essays are as follows. Essay 1: Limits to peer appraisals. The first essay studies situations when agents work in a team imperfectly 0observing their peer’s effort. The principal bases compensation on her signals on agents’ efforts as well as announcements by agents about their peer’s effort. Existing literature, Ma (RES 1988), Brusco (JET 1997) and Fisher and Hughes (JITE 1997), considers collusion amongst agents and describes mechanisms to achieve truth-telling at very small costs. It is shown that first best can be achieved if agents have perfect information. Management and labor literature as well as anecdotal evidence suggest peer appraisals have limits. We contrast the economics literature to show peer appraisals have limits which arise when agents’ have imperfect information and agents’ can exert peer pressure. Only peer pressure or imperfect information does not result in a cost. This setup is then expanded to include role of agents’ beliefs on their peer’s working. It is shown that when agents believe their peer is likely to shirk the cost to ensure truth-telling is high. Also, we show the principal can ensure agents announce what they observe even when they believe their peer always works or always shirks. Unlike existing literature where agents always work, we also study a setup in which agents have draws of cost of working and hence depending on this cost they shirk sometimes. Essay 2: Developmental peer appraisals. We develop a principal-agent analysis of developmental peer appraisals. In our analysis a worker can acquire imperfect information about her peer by monitoring her, incurring a cost in the process. If peer appraisals are implemented, the worker provides to the principal feedback on the peer’s specific training deficiencies, and the principal uses this feedback along with her own imperfect information to decide the type of training. Only the appropriate type of training improves a worker’s output. If peer appraisals are not implemented the assignment of type of the training is based only on the principal’s information. There is a fixed cost of implementing peer appraisals at the workplace. Compensation is based on imperfect signals on the two agents’ outputs. By analyzing the different tradeoffs which emerge while considering implementing such a mechanism we identify five factors that drive its usage, namely, the value of tournament-type compensation, the work ethic of workers, the level of training involved, the level of teamwork, and the quality of the principal's signal on worker type. Based on this analysis, we then predict conditions when peer appraisals are more likely. Using a unique dataset of HR practices of over 1500 UK firms, we present empirical evidence largely consistent with our theory. Essay 3: Peer appraisals-characterizing incentives “to take credit.” This essay should be viewed as exploratory in nature. In particular, the third essay considers a setup like Itoh’s (1991) in which agents produce output helping each other. Unlike Itoh’s analysis in our setup the usefulness of help is uncertain. This variation introduces an uncertainty in production as an agent could put in help effort but it may not be useful. If a signal on usefulness of help was available then this risk can be reduced by paying an agent an amount when help is not useful. Also, a signal on help provides better incentives to helping efforts. We introduce peer appraisals and show such a signal can be collected by the principal by asking agents to announce the help they received. However, an agent is better off if she announces the peer did not help her even if she received help. Management and labor literature as well as anecdotal evidence suggest agents “take credit”. Such behavior by agents can make the signal uninformative. Given this incentive we devise mechanisms to still use the agents’ signal. We also assume the principal has an imprecise signal on helping efforts and show this signal can be used to limit lying. We consider two setups. In the first setup agents announce the amount of help they received. We show a downward bias exists and the principal can limit the bias by punishing agents based on the square of the difference between the principal’s signal and the agent’s announcement. This punishment introduces a risk in agent’s compensation and hence a cost to the principal. Using numerical simulations we find that when the principal has a relatively good signal, agents signal on help is more valuable in terms of filtering risk due to uncertainty in help than providing the other agent incentives to work. (Abstract shortened by UMI.)




Barron, Purdue University.

Subject Area


Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server