Mathematical programming models of "representative" farms are commonly used
to evaluate policies such as input subsidies and output price supports. On
canals in India, upstream farmers routinely use more irrigation water than
allotted. In such circumstances, the programming model should encompass fa
rmers' locational heterogeneity. Were, a representative watercourse with th
irty farms is calibrated to the eight crops, fifteen irrigation turns, yiel
d responses to water, and seepage in Maharashtra. Not only does water "thef
t" increase the social cost of price policies, but the policies' increased
inducement to theft by upstream farmers leaves those downstream with less w
ater and lower incomes.