Experimental Design
we conducted a census of private clinics in two low-income neighbourhoods of Pakistan’s second largest city, Lahore: Walton and Dharmapura. Listing was done in two phases: (1) clinics were identified, photographed, and geo-located by local informants. Our research team confirmed these locations, and collected other preliminary information, such as fees and specialties. This process generated a set of 52 clinics, with consultancy fees ranging from PKR 50 to 1,000 (about USD 0.42 to 8.40 at the time of the study), which were then refined to a universe of 36 feasible clinics, after excluding irrelevant clinics (i.e., clinics for mental health, pediatrics, and gynecology). A vast majority of clinics were single provider clinics and while our unit of analysis is the clinic, the study was designed such that all visits were at the same time and day for each clinic, to maximize the likelihood of meeting the same healthcare provider.
We conduct an audit study, using professional actors hired through a local acting troupe. we selected 4 actors for the study: 2 male and 2 transgender actors. Our actors were given a consistent background, symptoms and opening script, adapted from Das et al. (2016). Our standardized patients reported symptoms that were consistent with asthma.
Each of our 36 clinics were visited twice, at least two weeks apart, once by a male actor and once by a transgender actor (order randomly determined). We kept the day and time across each wave constant for every clinic to maximize the probability of facing the same healthcare provider. We randomized the gender of the patient visiting each clinic in the first wave. The minimum two-week gap was implemented to ensure there was limited contamination across the two visits. Recall that patient load at these clinics is large, and given this and the gap in time, we do not anticipate healthcare providers noticing any similarities across visits and found no evidence that they did.
Before each visit, actor health was checked to ensure they did not carry any contaminating physical symptoms. Across all visits, actors were accompanied by a research assistant who would remain outside the clinic. The research assistant formed part of our safety protocol, in addition to collecting data immediately after a completed visit.
Upon completion of all visits, medicines dispensed (typically unmarked) or prescribed were identified and categorised by a team of three pharmacists. Overall, the study took a year to run, with background work, such as focus groups with our panel of doctors, and identifying an appropriate acting troupe beginning in the summer of 2017 and continuing into the fall. Preparations for the visits (auditions, training, piloting and two rounds of listing) took place in the Winter of 2017, with our rounds of data collection taking place in February and March 2018. Finally, medication categorization and data cleaning were completed over the summer of 2018.