Ethnic and social heterogeneity. The boundary between economic and noneconomic heterogeneity is a fuzzy one. Differential exit options appear to be economic because they are like unequal assets; nevertheless, the effect of different exit options, we suspect, operates through the weakening of social norms and sanctions. Less explicitly economic forms of heterogeneity have significant effects in the studies considered here. Bardhan (2000) controls for whether at least three-quarters of surveyed farmers in an ayacut are members of the same caste. This kind of caste homogeneity is strongly associated with the absence of intravillage conflict, but it is not significantly associated with rule conformance. (Bardhan did not include caste homogeneity in his statistical models of maintenance.) Khwaja (2000) computes a “fragmentation index” that is the average of ethnic, political, and religious fragmentation indices for the communities in his Pakistan study. Each fragmentation index is the probability, in a given community, that two randomly selected individuals belong to different groups. Khwaja’s fragmentation index bears a negative relationship with project maintenance.
A measure of social heterogeneity is whether or not irrigators come from more than one community. Dayton-Johnson (2000b) includes in his statistical models the number of ejidos (Mexican agrarian reform communities) from which unidad members are drawn: It is consistently and negatively associated with infrastructure maintenance. This provides strong support that when enforcement crosses ejido boundaries, it is less effective. Nevertheless, the Philippine study of Fujita et al. (2000) includes a similar variable—the number of villages represented in the irrigation system—and it is not significant. Baker’s (1997) study of 39 kuhl irrigation systems in Himachal Pradesh considers the effect of differentiation, which is “high when a kuhl irrigates more than one village, the irrigators of the kuhl are composed of multiple castes, and land distribution is relatively unequal” (1997:204). Baker proposes that, in the presence of high differentiation, increased opportunities for nonfarm employment can place intolerable stress on traditional kuhl management regimes.22 Tang’s (1992:68-72) evidence on the effect of “social and cultural divisions” is ambiguous and marked by small numbers of observations. The effect is mediated by the institutional nature of the irrigation system. Where it is community managed, sociocultural heterogeneity does not preclude good performance; where the system is agency managed, this heterogeneity is associated uniformly with poor performance. Possibly overlooked is the selection bias in community-managed systems: If a group is not able to organize itself, or if conflicts become too severe, the system will not exist when the researcher goes into the field. The same cannot be said of agency-managed systems. Thus, we are left with a sample of community-managed systems that have survived a process that requires high levels of cooperation.
Choosing rules. Heterogeneities, finally, can affect performance indirectly by means of their effect on the choice of rules. Dayton-Johnson (2000a) finds that wealth inequality increases the probability of observing proportional water allocation (as opposed to equal shares for all farmers). This is consistent with richer