A census was conducted of all hair salons in Lusaka and a grid was imposed on the GPS-mapped locations of the salons to divide the city into equal geographical areas of 650m by 650m each. A buffer zone of 75 m on all sides of the grid cell was excluded, resulting in at least 150 m between salons in adjacent areas. The resulting areas, each measuring 250,000 m2, served as the unit of randomization. Salons located in buffer areas were not invited to join the program. The final sample for randomization consists of 205 distinct neighborhoods, containing 1222 hair salons, all of whom were invited to join the training. To prevent spillover effects, agents in the same neighborhood were assigned to the same treatment and salons neighbors are either in the same treatment or not part of the program.
Invitation to attend the training was extended to all 1222 stylist of whom 747 ultimately became agents. Attendees were allocated to one of four intervention groups - no reward, 10% margin, 90% margin, and non-financial (stars and thermometer display). Thereafter, dispensers or single packs could be purchased either during a monthly restocking visit by SFH representatives or by calling a toll-free number dedicated to the female condom program.
To elicit an incentive-compatible measure of pro-social motivation toward HIV causes, a dictator game was implemented during the training program. Agents were told that, in addition to the training show-up fee (40,000 ZMK), each of them would receive 12,500 ZMK, which they could keep for themselves or donate, in part or in full, to a well known charity in Lusaka that provides care to HIV/AIDS patients. The amount donated is taken as a proxy for the agents' motivation for the cause. Since this is likely to be correlated with the agents' wealth, it is always used together with asset- and socio-economic-status measures in the analysis that follows.
Finally, given the non-financial treatment with its public display of thermometer and stars differs from the other three treatment conditions in that the customers do not that the stylists are being rewarded or their level of condom sales performance, a placebo star treatment is given to the control group in a later round. The thermometer looks the same as that of the actual treatment group but the number of stars reflect average sales by all salons, rather than the individual salon sales.