The data is collected in two locations: Birbhum in West Bengal and Udaipur in Rajasthan, where they got a sample of 166 GPs and 100 GPs, respectively.
First of all, they took data from the 1991 census of Birdhum (right before the policy was implemented) and check that there are no significant differences between reserved and unreserved GPs.
After that, they conducted a survey in 2000, from where they collected the data in two stages: First, GP level data, where they conducted an interview with the Pradhan where he or she is asked a set of personal questions (family background, education, political experience...) and a set of questions related with the activities they carried out while they were in charge. And second, village level data, where they conducted a survey of three villages in the GP: two randomly assigned and the one where the Pradhan resides. While the survey, they drew a map of the village with a group of 10 to 20 villagers with the main features of each area. The minutes of the village meetings and the number of complaints to the GP in the previous six months are also collected. They found the village level data more reliable than the GP level because it is not provided by the Pradhan and because it was easier for the villagers to provide information about the town where they reside. Therefore, using the GP level data they get ride of sample error because they get the information for all the GP. To solve this, they made the analysis separately for both levels and compare the results.
The same procedure of village level data collection was made in Udaipur in Rajasthan, and also there are no significant differences between reserved and unreserved GP as in West Bengal. They provide GP level and village level of data for Birfhum in West Bengal, but only village level data for Udaipur in Rajasthan.