We thank Eliane Ralison for data entry assistance and Bart

We introduce a method for properly attributing observed productivity and risk changes among new production methods, farmers, and plots by controlling for farmer and plot heterogeneity. Results from Madagascar show that the new system of rice intensification (SRI) is indeed a superior technology. Although about half of the observed productivity gains appear due to farmer characteristics rather than SRI itself, the technology generates the estimated average output gains of more than 84%. The increased estimated yield risk associated with SRI would nonetheless make it unattractive to many farmers within the standard range of relative risk aversion. Key words: Madagascar, rice, SRI, technology adoption, yields. Economic growth theory holds that techno-logical change is the primary driver of long-run economic growth and improvements in hu-man nutrition and well-being. However, the mere development of a technology is insuffi-cient because its adoption by producers may be slow, partial, reversible, or absent all to-gether. Cross-sectional variation in the extent and rate of new technology adoption is of-ten attributable to characteristics of the farmer or the farm, especially to differences in ed-ucation, access to extension or financial ser-vices, risk preferences, plot biophysical condi-tions, and farm size (Sunding and Zilberman). Due to the fact that farmers ’ adoption choices are typically limited by the availability of reli-able information, finance, labor, or other com-plementary inputs—perhaps especially among poor smallholders in the tropics—observe