This study is based on a collection of lending actions from the Kiva.org database. These data have been retrieved through the API (Application Programming Interface). Through this API, data can be requested from Kiva.org’s public database. With this purpose, scripts were written for datacollection. Further scripts were written to operationalize the variables in the research design.
The longitudinal data have been collected between May 10th 2012 and September 4th 2012. These data consist of all lending actions in this period. In total, 1.217.627 lending actions were saved. For each lending action, an identification number, the date and time, the loan (identification number for the loan) and the lender (identification number for the lender) were saved. 120.704 Lending actions were attributed to anonymous lenders. Based on the lending action data, 47.790 unique loans and 263.122 unique lenders have been saved. Subsequently additional data were collected. This was done retroactively in three days following the initial data collection period. In this period data for the loans, memberships and lending teams were collected.
The available data for loans is limited by Kiva.org due to privacy concerns. The exact lending amount for a lending action is not available. Instead, data is saved for the target loan amount, the amount lent and the amount of lenders that contributed. With these data per loan, the average amount lent is calculated. This is done by dividing the total amount lent by the amount of lenders that contributed.
The available data for memberships is also limited. Terminated memberships are not registered in Kiva.org’s database. Only continuous memberships have been retrieved, for each saving the date and time. Lenders have together joined 120.620 unique teams. In the first period of data collection, 14.783 lenders joined 20.070 (extra) teams. Each team joined was saved too, along with the amount of team members. Finally every team captain was was marked as such in the database. With this last data, regular team members can be distinguished from captains.
All data were saved in a MySQL database. Some other researchers have already requested and received these data (a PhD student HCI (Human–computer interaction) en ICTD (Information and Communication Technology for Development) from the University of Michigan and a student of Economics from the University of Nottingham). Given the popularity of the data, MySQL exports are available on request. Use the link to request your copy.
The data are compressed to 19.3MB. In the .rar file (extract with WinRar) you will find the following SQL files:
- lending_actions.sql (65.5MB) – lending actions
- lender.sql (10.5MB) – lenders
- loan.sql (2.36MB) – loans
- lending_team.sql (769KB) – teams
- membership.sql (5.48MB) – memberships
- setmaineffect.sql (37.5MB) – Dataset for testing the main effect of team membership on lending behavior
- setmodels.sql (9MB) – Dataset for testing the effect of lending behavior or team captains on that of team members (modelling effect). MySQL rand() was used to create a random selection from this dataset.