Dynamic use of agriculture incentives to promote conservation

A common problem in resource management concerns the offer of incentives to individuals and organizations to encourage their participation in conservation practices. For underlying acceptance rates that are not known initially, a structured process of annually offering different incentives and recording responses can guide decisions about the number of incentives to offer through time. We examined a dynamic model that includes combinations of different incentives that are offered to farmers each year, with the objective of strengthening participation in agricultural conservation. We considered 3 approaches for identifying incentive acceptance rates that measure participation, including an arbitrary rate assignment independent of incentive data; iterative averaging based on annual offers and acceptances; and Bayesian updating of expected acceptance rates with beta distributed rates and a binomial data distribution. We performed a proof-of-concept simulation that evaluated different strategies that utilize the estimated rates to offer incentives through time. The strategies we evaluated included matching offers to prior year acceptance rate estimates, possibly weighted by their precision; offering only the incentive with the largest estimated acceptance rate each year; and equally allocating offers among the incentives. We evaluated scenarios for non-stationarity in the underlying acceptance rates and considered alternatives for using offer data to deal with non-stationarity. Results indicated that the trajectory of future offers is marginally affected by variance weighting, and there is a strong influence of non-stationarity. Truncating data with a negative rate change accentuates the decline in an offer trajectory, whereas truncation with a positive rate change dampens the trajectory increase. We found that the way one handles variation in estimating acceptance rates, in particular non-stationarity, can have a substantial effect on the management of incentives, especially in the short term. The results can be generalized to allow for multiple changes in underlying rates over a project time horizon. We highlight the strong similarities between our framing of the incentives problem and the more general context of sequential experimentation, with its opportunity to improve conservation through adaptive learning.