Experimental Design
\textbf{Elicitation for Structural Estimation:} On Day 0, we call in the CHWs to collect demographic data and elicit preferences after carefully explaining the experimental protocols. The CHWs divided $m$={15...57} tasks (household health surveys) between two dates that were one week apart, Day 7 (allocation $v_1$) and Day 14 (allocation $v_2$). The CHWs make these decisions at multiple task rates, essentially interest rates between the present and future, which were experimentally varied: $R$ = {0.4, 0.5...1.7, 1.8}. For each task allocated to Day 14, the number of tasks allocated to Day 7 was reduced by $R$. These were advance choices --- made one week before the task was to be attempted. On Day 7, all CHWs make the same choices again, but for that very day (before they had to complete the tasks for that day) and the next week. These were the immediate choices --- made on the same day the task had to be attempted. As multiple choices were made, in advance and immediately, and at multiple task rates, we chose one choice probabilistically for each CHW to implement. This is the decision-that-counts and was implemented on Day 7 and then again on Day 14. %This design was based on \cite{andreoni2012estimating}.\footnote{It has been used by and Chaudhry and Hussain (2022).} %In Figure \ref{decisionset}, we show one example of the decision sets we used to elicit these preferences. This exercise took place in September 2022.
To avoid corner solutions at 0 or $m$ households in allocation decisions, we set a minimum of 5 and maximum of 27 households in the decision sets we offered. The goal of setting a minimum was to ensure that CHWs worked on both dates and made a choice about how to allocate tasks between them. Further, when CHWs made decisions on Day 7, we did not remind them of the Day 0 allocations. Importantly, on Day 0, CHWs were making decisions involving two future work dates (one and two weeks later), whereas on Day 7, they were making decisions for the same day and the week after. From every CHW we elicit 15 advance and 15 immediate decisions over two weeks, and 1 of their 30 decisions was assigned to them as the decision-that-counts.
The purpose of this exercise was to elicit the effort allocations and structurally estimate every CHW's discount factor $\delta$, present bias $\beta$, and intertemporal effort cost $\gamma$.