Do observed contracts have the properties predicted by the principal-agent model with moral hazard in contract theory? This paper tests the predictions of such an agency model in the context of sharecropping in North India. The most well-known explanation for sharecropping is based on the principal-agent model with a trade-off between risk and incentives. Even though sharecropping contracts are quite prevalent in rural areas of developing countries, there is very little evidence on factors determining the instruments used to provide incentives in such contracts. Using information in the data giving rise to exogenous variation in technology across regions, this paper tests for the effect of cultivation technology on the incentive structure of share contracts as predicted by the agency model. Existing empirical evidence on sharecropping seems to indicate that yield on sharecropped plots is lower than on owner-operated plots, i.e., there is an incentive problem or moral hazard (Radwan A. Shaban, 1987; Jean-Jacques Laffont and Mohamed S. Matoussi, 1995). There is some informal evidence that landowners use various mechanisms to improve efficiency in sharecropping by participating in cost sharing and by repeating contracts. Robert M. Townsend and Rolf A. Mueller (1994) examine the nature of these mechanisms in detail but their data do not permit econometric tests. More generally, though the principal-agent model has been widely studied there is little existing empirical evidence for it. Michael C. Jensen and Kevin J. Murphy (1990) find that executive compensation is only weakly sensitive to firm performance. In recent work Rajesh K. Aggarwal and Andrew A. Samwick (1999) find that executive’s pay-performance sensitivity is decreasing in the volatility of firm’s performance. However, their results are sensitive to the inclusion of other characteristics of the firms in the regressions. As Pierre A. Chiappori and Bernard Salanie (2003) note in a recent survey, empirical work on contract theory using nonexperimental data needs to be careful in adequately correcting for underlying heterogeneity across agents. Otherwise the parameters of interest would be hard to interpret if such heterogeneity affects contract choice. For example, Douglas W. Allen and Dean Lueck (1995) find no role for risk in the choice between share contract and fixed rent contracts but they do not take the heterogeneity across agents into account. In a recent paper addressing the issue of heterogeneity across agents, Ackerberg and Botticini (2002) find a significant role for risk in the choice between share contract and fixed rent contract. After correcting for endogenous matching between landowners and tenants, they find that wealthier tenants are more likely to be in fixed rent contracts. In this paper, we are able to address such estimation issues and check for the robustness of our results regarding the relationship between technology and the design of share contracts * Department of Economics, Pennsylvania State University, University Park, PA 16801 (e-mail: [email protected]). This paper derives from related work done earlier in my dissertation. I am grateful to two anonymous referees for very helpful suggestions. I thank James Heckman, Lars Stole, Robert Townsend, Kala Krishna, and seminar participants for useful comments. Financial support from the Andrew Mellon Foundation for both rounds of fieldwork is gratefully acknowledged. Any errors remain my own. 1 See Nirvikar Singh (1991) for a survey of various theories of sharecropping including Steven N. Cheung (1969), C. H. Hanumantha Rao (1971), Joseph E. Stiglitz (1974), David Newbery and Stiglitz (1979), Avishay Braverman and Stiglitz (1982), Mukesh Eswaran and Ashok Kotwal (1985), and Sudhir Shetty (1988). A common feature of the different theories is an emphasis on uncertainty and asymmetric information. 2 An exception is Daniel A. Ackerberg and Maristella Botticini (2002). Ackerberg and Botticini differ from our paper in that they examine the role of tenant’s risk aversion in the choice between fixed rental contract and share contract—they do not examine share contracts per se. 3 See John E. Core and Wayne Guay (2000).

What are the mechanisms by which societal discrimination affects individual achievement, and why do the effects of past discrimination endure once legal barriers are removed? We report the findings of two experiments in village India that suggest that the mechanisms of discrimination operate, in part, within the individuals who are members of the groups who have been discriminated against. We demonstrate that publicly revealing an individual’s membership in such a group alters his behavior in ways that make the effects of past discrimination persist over time. A growing literature in social psychology on stereotype threat finds that stereotyped-based expectations affect individual performance in the domain of the stereotype. A study by Jeff Stone et al. (1999) is illustrative. When college students were asked to perform a task described as diagnostic of “natural athletic ability,” blacks—stereotyped as better athletes, but worse students than whites—performed better than whites. When the same test was presented as diagnostic of “sports intelligence,” the performance of blacks declined, that of whites improved, and the racial gap was reversed. Evidence suggests that a mediating factor in stereotype threat is a change in self confidence (Mara Cadinu et al., 2005) In our studies, we investigated whether the public revelation of social identity (caste) affects cognitive task performance and responses to economic opportunities by young boys in village India. Subjects were sixth and seventh graders drawn from the two ends of the caste hierarchy. We asked subjects to learn and then perform a task under incentives, and we manipulated whether their peers in the experimental session knew their caste. Caste is well-suited to this manipulation because, unlike race, gender, and ethnicity, there are no unambiguous outward markers of caste among young boys. Six subjects, generally from six different villages, participated in each experimental session. In the control condition, the subjects were anonymous within the six-person group. In the experimental conditions, the experimenter publicly revealed subjects’ names and caste. In the task—solving mazes—in which performance was studied here, the low-caste subjects in the anonymous condition did not perform significantly differently from high-caste subjects; but when caste identity was publicly revealed in a mixed caste group, a significant caste gap emerged. The caste gap was due to a 20 percent decline in the average number of mazes solved by the low caste. The study shows that publicly revealing the social identity of an individual can change his behavior even when that information is irrelevant to payoffs. Our results are a generalization of the literature on stereotype threat. Like that literature, we find that individuals’ performance is more in accordance with the stereotype of the group when group membership is made salient in some way. Unlike that literature, salience in our experiments depends on the public revelation of social identity and more importantly, we do not argue that the domain of the tasks undertaken by † Discussants: Rachel Croson, University of Pennsylvania; Iris Bohnet, Harvard University; Stefano DellaVigna, University of California-Berkeley.