Experiments to inform the design of payments for ecosystem services in Uganda

The experimental features of this study include a modified referendum question regarding willingness to accept a hypothetical PES payment, and a series of incentivized games, one focused on harvesting of trees and the other on trust. Non-incentivized measures of risk and time will also be collected, as well as information on basic household demographics and assets.

2022-10-17

2023-06-01

Median willingness to accept a hypothetical PES payment
Harvest behavior in incentivized tree game
Money returned in trust game

Design WTA referendum question:

The referendum question design follows standard practice: it begins with a brief description of the contract for tree conservation on one hectare of land and then presents a randomized price, to which the individual responds that they would accept or not accept the payment. There are two versions of this question: one for households identified as having access to communal forest and one for households with mailo or private forest access. Within each of these categories, there are contracts offering the option of one payment per year or dividing the same payment equally across four quarters. There are therefore two levels of randomization: the price and the timing. Households will only respond to one version of the timing.

Follow up questions asking if they would like to enroll more land, and whether or not they would accept a higher or lower price for the same piece of land (higher if they responded no the first time, lower if they responded yes) are meant for descriptive purposes and also to test respondent understanding of the contracts.

Design tree game:

Participants will be asked to make decisions about harvesting trees in an experimental “forest” over several periods. Participants are not told how many periods over which they will make decisions, and so may think of the task as an infinite, discrete choice game.
Participants will be randomized at the individual level, in equal proportions, into 1 of 3 groups:
• Group 1: Insecure property rights (mailo treatment). Participants experience a 20% probability of being kicked off the land in each round. To determine whether the participant is kicked off the land, the participant will draw a marble from a bag containing 2 red and 8 blue marbles. If they draw a red marble, the game ends and no more harvesting decisions are made. The participant keeps their accumulated earnings from each period up to that point.
• Group 2: Insecure property rights with option to secure. Participants experience a 20% probability of being kicked off the land in each round, as in Group 1. However, they may secure their property rights by purchasing a certificate of occupancy for a cost, which will be deducted from their earnings. The cost of the certificate has been calibrated from pilot data on the true cost of certificates relative to the average profit per tree harvested.
Participants will be offered the option to buy the certificate at the beginning of each round. If they purchase a certificate of occupancy, they move to full property rights in which there is a zero probability of being kicked off the land. The game finishes until no trees remain or the periods expire. If they do not purchase a certificate, the game proceeds as in Group 1.
• Group 3: Full property rights. Participants make harvest decisions until no trees remain or the periods expire.

Design trust game:

The trust game follows the design of Berg et al. (Games and Economic Behavior, 10, pp. 122–142, 1995). Participants will be randomly paired with an unknown person who they are told is from somewhere else in Uganda. Each pair is made up of a Player 1 and a Player 2. To start, Player 1 receives 4,000 UGX. Player 1 then decides how much to give to Player 2. The amounts that Player 1 can give are: 4,000, 3,000, 2,000, 1,000, or nothing. Any amount that Player 1 decides to give to Player 2 will be tripled before it is given to Player 2. Next, Player 2 decides how much to return to Player 1. Player 2 can return any portion (including zero) of the amount they received back to Player 1.

Each participant will play the trust game twice: first as Player 1 and again as Player 2. We will follow the strategy method in which the participants state the amount they would return as Player 2 for every possible behavior from Player 1. After Player 1 and Player 2 choices are elicited, we will randomly select one strategy (Player 1 or Player 2) to be implemented for real payment.

Randomization done by a computer

Randomization occurs at the individual level.

No

28 villages across three regions for a total of 84 villages.

We are targeting a sample of 560 households per region, for a total of 1,680 households.

For the WTA experiment, we will randomize equal parts of the sample into either quarterly or annual payments. Within each of these, we will sort 1/7 of the sample into each of the 7 payment categories. Randomization is at the individual level but we will make sure there are balanced numbers of treatments and controls in each surveyed district. The specific numbers are unknown, since this randomization will occur within the two types of land tenure categories -- communal or individual, and these values will be collected within the survey.

For the tree game, randomization will occur at the individual level and the sample will be split equally across the three treatments.

WTA: The sample size was determined to satisfy the requirements of a WTA experiment with 5-7 levels of payment. There are 3 regions with potentially 2 types of payment (annual and quarterly), which makes 6 arms. We calculated power using a dose-response function common in the health literature and also simulated the possible sample size using data from a similar WTA experiment in Mexico. The most conservative estimate of sample size was 280 per payment type per region. Tree game: Our empirical objective is to compare average harvest behavior under full vs. insecure property rights. The optimal harvest rate under full property rights is 2 trees per round. Assuming a standard deviation equal to the mean, with 0.80 power and alpha = 0.05, adjusted for a 3-arm design (alpha = 0.05 / 2 = 0.0025), we could detect at minimum a 25% difference in average harvest rates across groups with our sample size (N=560 in each arm). Comparing only full rights with insecure rights (Group 1 v. Group 3), we can detect at minimum a 14.7% difference with this sample size. Trust game: Our empirical objective is to compare the impact of different property rights structures in the tree game on behavior in the trust game. We have calibrated power calculations using information on stated trust preferences from the Afrobarometer survey for Uganda and information on behavior in the trust game among Ugandan participants from Johnson and Mislin (Journal of Economic Psychology, 32, pp. 865-889, 2011). Matching the Afrobarometer survey with information on land tenure security suggests that insecure land tenure is associated with 22% lower trust in Uganda. Assuming 0.80 power and alpha = 0.05, adjusted for a 3-arm design (alpha = 0.05 / 2 = 0.0025), we could detect at minimum a 22% reduction in the amount sent by Player 1 in the trust game across groups with our sample size (N=560 in each arm).