Using Field Experiments to Elicit Risk and Ambiguity Preferences: Behavioural Factors and the Adoption of New Agricultural Technologies in Rural India

Abstract

We conduct a series of experiments in rural India in order to measure preferences related to risk, loss, and ambiguity. By combining these results with a discrete choice experiment over new and familiar rice seeds, we demonstrate how these behavioural parameters affect decisions to adopt new agricultural technologies, especially when the new technologies are risk reducing. We find that risk averse and loss averse individuals are more likely to switch to new seeds demonstrating risk reducing characteristics, while, contrary to expectations, ambiguity averse individuals are no more willing to retain their status quo than switch to cultivating the new variety.

Acknowledgements

This paper was prepared as a contribution to the Cereal Systems Initiative for South Asia, with generous funding provided by the United States Agency for International Development and the Bill and Melinda Gates Foundation. Any opinions expressed herein are those of the authors and do not necessarily reflect those of the funding agencies. The authors wish to acknowledge David Spielman, Mark Rosegrant, and Claudia Ringler, who provided helpful comments on an earlier draft of this paper, as well as participants at the 2013 Agricultural and Applied Economics Association annual meeting in Washington, D.C., August 4–6, 2013 and two anonymous reviewers. The data and code used in this paper will be made available from the authors upon request. Any and all remaining errors are our own.

Notes

1. At the time these experiments were conducted, USD 1 was approximately equivalent to INR 54.

2. It should be noted, however, that respondents were not forced to switch. In conducting the experiments, several examples were given prior to eliciting responses. In one of these examples, the riskless option is chosen throughout, while in another example, the risky option is preferred in all rounds.

3. Indeed, one criticism that is often raised against Multiple Price List experiments is that they only allow the estimation of an interval value, rather than a point estimate. As a reviewer has suggested, experimental elicitation of behavioural parameters – especially when the estimate is an interval – necessarily implies that the behavioural parameters are measured with error. If indeed this is the case, then regressions incorporating these parameters as explanatory variables are prone to attenuation bias at least. While an exposition of these measurement issues is beyond the scope of the present paper, this could be an area for future research.

4. When switching rows take extreme values {0, N}, there are potentially an infinite number of parameter combinations for which some of the switching rows are consistent if the parameters’ possible value ranges are unconstrained. If we assume 01.5 and 01, then we can approximate parameter combinations for these as the mean of a truncated range of possible values for and .

5. In the case of our experiments, there are no potential losses, so knowledge of is not needed to estimate .

6. While the actual treatment of wins and losses resulting from the lottery is not symmetric (that is, cash payments for wins while losses are ignored), we do not feel that this impinges upon incentives for several reasons. First, respondents were told prior to the experiments that they would win or lose real money based on their responses to the lottery choices and the outcome of a randomly drawn lottery, so it was in their best interest to respond as honestly as possible. Second, enumerators were not responsible for any real transactions: payments were ultimately made by field supervisors on the teams’ last day in each village. Therefore there is little chance for respondents to undertake strategic behaviour in response to knowing they would not be asked to pay money for losses incurred in the third experiment.

7. A check variety is a variety (usually one commonly grown in a particular agroecology) that is used for comparisons in agronomic field trials.

8. Tanaka et al. (2010) report average values of  = (0.59, 0.74) and (0.63, 0.74) in the south and north of Vietnam, respectively, while Liu (2013) reports average values of  = (0.48, 0.69) in her sample of Chinese cotton farmers.

9. A significant number of respondents did not switch at all in both series and are reflected in the height of the cone corresponding to (N,N) in the figure. At the same time, a large number of respondents switched in the first row in both series, represented by the cone at (1,1). These frequencies were fairly large relative to others and have been truncated to retain the aesthetics of the image.

10. While the distinction between risk averse and risk seeking has to do with a very clear change in the curvature of the value function, the distinction between extreme risk aversion and moderate risk aversion is admittedly somewhat arbitrary. We have selected as demarcating extreme risk aversion based on analysis of kernel densities (not reported), which indicated a multi-modal distribution, with a nontrivial density roughly centred on .

11. An alternative approach, which would allow one to ascertain the effects of these parameters on attribute preferences, would be to interact these parameters with the attribute levels. Since this was not our objective in this study, we have opted for the simpler alternative indicated above.

12. The RPL estimator employed here exploits the longitudinal nature of the data (each survey participant responds to nine choice scenarios) by controlling for random effects. If the behavioural parameters are measured with error (see note 3 above), then under the assumption that the measurement error is uncorrelated with dependent variables and independent from random variation in farmers’ preferences or errors in utility maximisation, then the composite error term in this regressions is itself random, and we can consistently estimate the marginal utilities and marginal choice probabilities using a random effects estimator.

13. Note we cannot say anything about whether they would actually adopt the new seeds if they were available. That is, we lack a proper counterfactual to make such assertions. These new seeds may be prohibitively expensive, or farmers may not have adequate liquidity or access to credit, or there may be supply side constraints. We are limited to making statements about farmer’s preference of these hypothetical seeds relative to one they are familiar with.