According to scoping surveys—conducted in the same setting but with non-experimental subjects—and field interviews, various caste groups that are present in this setting can be ranked in terms of ritualistic or social practices. Broadly, there are three main social categories—General Castes (GC), Other Backward Castes (OBC), and Scheduled Castes / Scheduled Tribes (SC/ST)—and they are ranked in this order. Furthermore, the castes within the Scheduled Castes category can also be rank ordered. In addition, some of the caste groups are strongly associated with specific historical occupations. Based these observations, I select three castes in the SC category, who are vertically differentiated in ranking (in the order below) and have strong occupational associations (in parentheses):
- Dhoba (washing clothes)
- Mochi (working with leather)
- Hadi (cleaning latrines)
I additionally select three castes—each perceived to be of similar rank to one of the above—and group them into the appropriate rank. I also select one OBC caste that is perceived to be higher in ranking than the rest, resulting in 4 distinct ranks. The sample is composed of the following groups, with the targeted number of observations written in parentheses.
- Sundi (120), Kaibarta (120), Dhoba (80), Pana (120), Mochi (80), Kela (120), Hadi (80)
(Edit: I additionally conducted a ranking survey, which helped concretely measure how these caste groups are ranked. The pre-assigned rankings based on the survey results inform my predictions for experimental outcomes.)
I recruit male daily wage laborers who are interested in participating in a low-skill manufacturing job for one day. Each participant is asked to answer a survey about whether he is interested in taking up different kinds of job offers. One of the offers he discusses will be randomly selected at the end of the survey as his actual job offer. If his answer during the survey indicates acceptance, the randomly selected offer is implemented.
While all job offers mainly involve producing paper bags—a task that has no caste associations—each offer also requires working on one extra task. Some of the extra tasks are associated with specific castes, while the rest are not. The extra tasks include the following:
- Caste-associated tasks:
o Washing clothes
o Mending leather shoes
o Sweeping latrines
- Linked Control tasks:
o Washing agricultural tools
o Mending grass mats
o Sweeping animal shed
- Control tasks:
o Making ropes
o Deshelling peanuts
The three linked-control tasks arguably require very similar skills as the three Caste-associated tasks that appear in the same order. In addition, all the tasks either do not require any prior training or only involve working as an assistant to an experienced trainer. Hence, no prior experience or skill is required for doing any of the tasks.
While the total working time is always set as 5 hours, the job offers require different amounts of time to be spent working on the extra tasks, namely 10 minutes, 30 minutes, 1 hour, or 1.5 hours. For example, if the job offer involves working on some extra task for 10 minutes, the worker would spend 4 hours and 50 minutes on producing paper bags.
The job offers are the same in all the other respects: paper bag production happens at a work site 30-50 km away, extra tasks are performed in a private space, and wage is set at the locally prevailing daily wage rate. In addition, all workers are invited to participate in a focus group discussion about agricultural practices that would happen later in the day. The description regarding the specific content of the focus group discussion is randomized across workers (more details below).
The experiment happens in the following steps. A surveyor sits down with a worker and describes all the potential job offers. The job offers are either for that day or the next day, depending on the time of the survey. The worker indicates whether he is willing to take up each of the offers. After going through the entire list of job offers, the worker rolls dice to determine which offer he actually receives. If his answer indicates acceptance for that offer, the randomly selected offer is implemented on that same day or the next day. If his answer indicates rejection, the worker does not get any work. This process is a modified Becker-DeGroot-Marschak (BDM) mechanism, which ensures (under weak theoretical assumptions) that it is in the worker’s best interest to answer truthfully regarding each job offer.
Those who get work answer the endline survey at the end of the work day and receive wages at the work site. Those who do not get work are offered a gift whose market value is equivalent to Rs. 50 to answer the endline survey. Everyone is invited to participate in focus group discussions at work sites or in their own villages.
The endline survey collects basic information on demographics (age, literacy, education, family, etc.), wealth (assets and work), why they decided to take up or decline certain job offers, their prior experiences with the tasks, opinions regarding caste-related vignettes, and if they received work from another source (if they declined their actual job offer).
The job offer survey scripts are individually randomized in the following ways:
- There are four different orderings of tasks
- The time requirements for the extra tasks are sorted either in ascending or descending order
- Either making ropes or deshelling peanuts appear in the job offer list
In addition, the description regarding the focus group discussion is randomized at the village level in the following way:
- Private: the discussion will be about agricultural tasks and practices. They will also discuss whether people were willing to wash agricultural tools during the job offer survey. However, answers regarding any other jobs will not be discussed.
- Public: the discussion will be about agricultural tasks and practices. They will also discuss whether people were willing to wash agricultural tools during the job offer survey. In addition, they will discuss whether people were willing to take up all the other job offers.
Hence, during the job offer survey, those in the Private condition expect that only their answers for the job offers involving washing tools will become known to other people in their villages. Those in the Public condition, however, expect all their answers during the job offer surveys to become known to other people in their villages.
My main hypothesis is that the fall in workers’ take-up rates for the job offers involving the caste-associated tasks—compared to the jobs offers involving the linked-control tasks—will depend on workers’ own caste rankings. That is, workers will be more likely to decline the job offers if the extra tasks are associated with castes ranked lower than their own. For example, for Kaibarta in Rank 2, two tasks—mending leather shoes and sweeping latrines—are associated with castes that are ranked lower than their own. In particular, my hypothesis is that this pattern will hold even when the job offers require working on the extra tasks only for 10 minutes.
(Edit: Supplementary experiment
In addition, I run a supplementary experiment to measure how much workers demand in additional wages in order for them to take up jobs that present identity concerns. Participants will come to the worksite to produce paper bags, but will be given offers to switch to working on the tasks that present identity concerns (as well as the control tasks). The total working time would still be fixed, so they will decide whether agree to switching to working on another task for some of the total time (i.e. for 10 minutes, 30 minutes, or 1 hour). Their willingness to accept the offers will be elicited using a price list that varies from 0 to 3000 Rupees, which is 1000% of their daily wage. After the workers indicate their choices, one offer will be randomly selected and implemented (a BDM-type method). This design will provide a precise way of quantifying the utility cost of violating caste identity concerns in monetary terms. Workers will be randomized into private and public conditions as in the main experiment.)