Experimental Design Details
To test whether relative pay comparisons affect worker effort and utility, we construct a design to accomplish two goals:
(i) for each worker, define a clear reference group of co-workers for relative pay comparisons;
(ii) be able to compare outcomes for workers whose absolute pay levels are the same, but who vary in their co-workers' (reference group) wages.
Workers will be employed full-time at worksites in low-skilled manufacturing tasks. Within each worksite, workers will be randomly divided into up to "teams", with 3 workers in each team. Within a team, all 3 teammates will perform the same production task. Each team at the worksite will have a unique task - for example, one team will make rope, another team will produce incense sticks, another team will weave floor mats, etc. - with a total of up to 10 tasks per worksite. Production will be an individual activity - teammates will sit together but will not need to do any work jointly. Rather, the purpose is that each worker's 2 teammates will be the only other people at the worksite making the same product, and will therefore be the most salient reference group for wage comparisons.
Before assigning treatment status, workers will undergo a "training period." During this period, all workers will receive the same daily wage. At the start of training, workers will be told that their post-training wages may depend on their output during training. Once training ends, workers' baseline productivities will be assessed, given their relative productivity ranks in the team: low, medium, or high.
Teams will then be randomized into one of four wage treatments:
(i) Heterogeneous: Each team member is paid according to his baseline productivity. The wages for the lowest, middle, and highest productivity workers are wL, wM and wH respectively, where w_training < w_L < w_M < w_H
(ii) Compressed_L: All team members are paid the same daily wage of w_L (the "low" wage).
(iii) Compressed_M: All team members are paid the same daily wage of w_M (the "medium" wage).
(iv) Compressed_H: All team members are paid the same daily wage of w_H (the "high" wage).
After being randomized into these treatments at the end of training, each person will work under his assigned wage until his fixed employment contract ends. We are sure to emphasize throughout the contract period that there will be absolutely no offers of future employment.
The design incorporates three important sources of heterogeneity. First, the various tasks differ in how easy it is to observe the output of one's teammates. We will stratify wage treatment assignment by task, generating variation in the observability of co-worker output within each treatment cell. Second, since output is continuous while productivity rankings are discrete, there will be natural variation in how much a worker's productivity level differs from that of his teammates, creating variation in {wage difference}/{productivity difference} ratios within each cell. Lastly, given that workers at each worksite are recruited from multiple villages, there is variation in the strength of pre-existing relationships and social ties between workers across teams.
In our analysis, we aim to separately identify effects on the intensive (effort) and extensive (attendance) margins. Measuring the response on attendance captures the extensive margin. However, isolating the intensive margin is more difficult. If there are attendance differences, then regressions of output on treatment status conditional on attendance will be biased. This is the standard selection problem in Heckman (1979). We propose to solve the selection problem by constructing an instrument for attendance based on weather shocks and the difficulty of traveling to the worksite. When it rains, it becomes much harder for workers to travel to the worksites, thus increasing the likelihood of an absence. However, the exclusion restriction for rainfall alone might not be satisfied; on rainy days, workers who do manage come to the site might be less productive. To solve this problem, we plan to interact rainfall with measures of the distance and material of the roads between each worker's village and the worksite. For each worksite, we recruit workers from at least three villages, thus giving us within worksite-day variation in the difficulty in coming to work.