This paper models the way in which farmers use information about pre-existing varieties of a particular cereal crop to estimate the yield of new varieties on their farms. Generally, the yields of new and existing varieties are very highly correlated. Farmers are shown to gain more "effective information" than when there are no existing varieties for comparison, particularly when there are relatively few observations of the yield of the new variety. This paper builds on the approach of Fischer, Arnold and Gibbs (forthcoming) in which additional information about the innovation received in a particular growing season is of less benefit to the farmer than information from an additional growing season. Both papers are embedded in Bayesian framework. The approach yields plausible adoption curves, and has implications for policy as it enables estimates to be made of the gains in yield which would accrue from earlier adoption due to additional data being available about the yields of of existing varieties, and to the extent to which learning-by-doing can be encouraged.
