Prickly Neighbors: Religion, Risk and Human-Wildlife Coexistence

2026-07-06

2026-07-27

Ambiguity Aversion Index (b-index), Ambiguity Insensitivity Index (a-index) elicited across three within-subject treatments. These indices are borrowed from established literature on ambiguity preferences including Baillon et al. (2018).

References
Baillon, A., Huang, Z., Selim, A., & Wakker, P. P. (2018). Measuring Ambiguity Attitudes for All (Natural) Events. Econometrica, 86(5), 1839–1858. https://doi.org/10.3982/ECTA14370

We use a lab-in-field experiment conducted among a sample of Rabari and Non-Rabari herders in villages including Bera situated in proximity of the Jawai Leopard Conservation Reserve. The proposed experiment would rely upon established methods proposed in Baillon et al. (2018) and Lotito et al. (2024). We use multiple price list ambiguity elicitation exercises to identify uncertainty preferences in terms of a constructed “natural” event, predation, and a hypothetical general event. The elicited ambiguity preference measure will be compared across the social groups and sources of uncertainty to identify the effect of religion on ambiguity preferences toward predation events. To validate the findings of the experiment, individual ambiguity preference measures will be used to predict the likelihood of herders to retaliate against wildlife and take precautionary measures.

References


References:
Baillon, A., Huang, Z., Selim, A., & Wakker, P. P. (2018). Measuring Ambiguity Attitudes for All (Natural) Events. Econometrica, 86(5), 1839–1858. https://doi.org/10.3982/ECTA14370
Lotito, G., Maffioletti, A., & Santoni, M. (2024). Testing source influence on ambiguity reaction: Preference and insensitivity. Journal of Risk and Uncertainty, 69(3), 349–411. https://doi.org/10.1007/s11166-024-09444-4

We use a mixed design relying on both within-subject variation across treatments and between subject variation in social identity. Every treatment is assigned to each individual. The three treatments are:

Predation-Gains Treatment: The source of uncertainty in this treatment is the number of goats in a herd killed by leopards in a year. The ambiguous events are created using data on annual goat predation in 2021-2025 collected from the sample population during an earlier survey. Following Baillon et al. (2018) six events are constructed. An event is defined as: "A randomly selected herder lost x goats due to predation by leopards in a randomly selected year between 2021-25." The six events are 0-1 goats (Event: E1), 2 goats (Event: E2), 3 or more (Event E3), 0-2 goats (Event: E12), 2 or more goats (Event E23) and 0-1 or 3 or more goats (Event E31).

For each event, the subject's matching probability is determined using a multiple price list method. This involves having the subject make a series of binary choices between an ambiguous event and increasingly likelier risky events. The level of risk at which the subject switches from choosing the ambiguous to the risky event, is used to calculate the revealed range of the matching probability for that ambiguous event. If the predicted event, either ambiguous or risky, comes to pass the subject will win 300 INR, given that the event is selected for incentivization.

Ellsberg-Gains Treatment: This treatment is a modified Ellsberg Urns game adapted from Lotito et al. (2024). The source of uncertainty is the distribution of candies colored red, blue and yellow in a bag containing 9 candies. An event is defined as: "A bag contains 3 red candies and 6 yellow and blue candies. The exact mix of yellow and blue candies is unknown. The color of a randomly drawn candy will be x". The six events are red (Event: E1), yellow (Event: E2), blue (Event E3), red or yellow (Event: E12), yellow or blue (Event E23) and red or blue (Event E31). A correct prediction wins the subject 300 INR if an event in this game is selected for incentivization.

Predation-Loss Treatment: This is identical to Predation-Gains treatment except that this is framed as entailing a loss of 300 INR if the predicted event comes to pass. However, this game is not incentivized and the losses are hypothetical.

Each subject faces all three of these games; first the Predation-Gains and Ellsberg-Gains games then the Predation-Loss game. It was explained to players that the Predation-Loss game would not be incentivized and was treated as separate from the first two games. To account for potential order effects, the order of the Predation-Gains and Ellsberg-Gains games were randomized between subjects. Within each game, the order in which a subject faced the six events was also randomized. At the end of the game and a short survey, one of the events of Predation-Gains or the Ellsberg-Gains game was randomly selected for incentivization.

The Primary hypotheses are:

H01) Rabaris are less ambiguity-averse than Non-Rabaris in Predation-Gains as compared to the Artificial source (Ellsberg-Gains).
b_pred_rab - b_ells_rab < b_pred_nonrab - b_ells_nonrab

H02) Rabaris are more ambiguity-insensitive than Non-Rabaris in Predation-Gains as compared to the Artificial source (Ellsberg-Gains).
a_pred_rab - a_ells_rab > a_pred_nonrab - a_ells_nonrab

The Secondary hypotheses are:

H02) Rabaris are less ambiguity averse than Non-Rabaris in Predation-Gains.
b_pred_rab < b_pred_nonrab
H03) Rabaris are less ambiguity averse than Non-Rabaris in Predation-Loss.
b_pred_loss_rab < b_pred_loss_nonrab
H04) Rabaris are similarly ambiguity averse to Non-Rabaris in the Artificial source (Ellsberg-Gains).

H05) Rabaris are more ambiguity insensitive than Non-Rabaris in Predation-Gains.
a_pred_rab > a_pred_nonrab
H06) Rabaris are more ambiguity insensitive than Non-Rabaris in Predation-Loss.
a_pred_loss_rab > a_pred_loss_nonrab
H07) Rabaris are Non-Rabaris are similarly ambiguity insensitive in the Artificial Source (Ellsberg-Gains).

References:
Baillon, A., Huang, Z., Selim, A., & Wakker, P. P. (2018). Measuring Ambiguity Attitudes for All (Natural) Events. Econometrica, 86(5), 1839–1858. https://doi.org/10.3982/ECTA14370
Lotito, G., Maffioletti, A., & Santoni, M. (2024). Testing source influence on ambiguity reaction: Preference and insensitivity. Journal of Risk and Uncertainty, 69(3), 349–411. https://doi.org/10.1007/s11166-024-09444-4

Randomization into sub-sample with Predation-Gains first then Ellsberg-Gains/Ellsberg-Gains first then Predation-Gains:
Carried out in computer prior to field deployment

Randomization into sub-samples based on order of events within each game:
Carried out in computer prior to field deployment

Randomization to select which decision to incentivize:
Carried out using survey software during game play

Random draw to determine the outcome of selected ambiguous/risky event:
Carried out using survey software during game play

Individual-level

No

430 individuals/herders.

430 individuals/herders

Minimum sample size: 210 Rabari and 210 Non-Rabari herders.
Target sample size: 215 Rabari and 215 Non-Rabari herders.

Ambiguity Aversion Index (b-index) Power analysis for the b-index was conducted via Monte Carlo simulation (1,000 repetitions, α = 0.05, two-sided) in Stata. The primary test (H01) is a difference-in-differences comparing the within-subject change from Ellsberg-Gains to Predation-Gains across Rabari and Non-Rabari. The target effect size is 0.14 (28.6% of SD) b-index points, anchored to the difference between Natural and Artificial Gains conditions in Watanabe et al. (2024). Using a standard deviation of 0.490 (the maximum SD across gain-domain treatments in Watanabe et al., 2024) and a within-subject correlation of 0.50 (based on the intra cluster correlation (ICC) reported in Anantanasuwong et al., 2024), a sample of 210 subjects per group achieves a simulated power of 81.2%. Ambiguity Insensitivity Index (a-index) For the a-index, the primary test (H02) follows the same difference-in-differences structure. The target effect size is 0.20 (41.7% of SD) a-index points, again anchored to Watanabe et al. (2024). Using a standard deviation of 0.480 and a within-subject correlation of 0.30 (Anantanasuwong et al., 2024), a sample of 210 subjects per group achieves a simulated power of 96.2%. In both cases the standard deviations are chosen conservatively as the maximum observed within their respective domains in Watanabe et al. (2024), so the reported power figures are lower bounds on true power. The design also achieves at least 80% power for each secondary hypothesis. References: Watanabe, M., & Fujimi, T. (2024). Ambiguity attitudes toward natural and artificial sources in gain and loss domains. Journal of Risk and Uncertainty, 68(1), 51-75. Anantanasuwong, K., Kouwenberg, R., Mitchell, O. S., & Peijnenburg, K. (2024). Ambiguity attitudes for real-world sources: Field evidence from a large sample of investors. Experimental Economics, 27(3), 548-581.