Field experiments methodology

The method of field experimentation  involves the separation of the target group into subgroups, each of which is subjected to controlled impact. Thus, it is possible to estimate the effect of the selected factors separately and to minimize the problem of endogeneity. Endogeneity problems inevitably arise when studying the impact of formal and informal institutions on economic development.For example, a researcher may observe that social trust is high, that courts work well, and that individuals comply with contracts, but in other settings, social trust is low, the courts work poorly and individuals do not comply with contracts.  The covariance of formal and informal institutions makes it difficult to identify the relative effects of each factor.

An illustrative example of the "field experiment" is given in Bannerjee et al. (2009), which evaluates the effectiveness of different approaches in reforming the police in the Indian state of Rajasthan.They began in 2005 by conducting a large survey of residents in Rajasthan on their perceptions of the performance of the police and the nature of their interactions with the police. They then randomly divided police stations within Rajasthan into 5 groups.  In the first group of stations, officers received training to improve their professional competence and skills in dealing with the public. In the second group of police stations, work assignments and schedules were made more regular and transparent. In a third group, community monitors were present in police stations for extended periods. In a fourth group, the transfer of officers was frozen for 18 months to minimize pressure from superiors. In a fifth group, no reforms were attempted.  Researchers then conducted a new survey of the residents of Rajasthan on the performance of the police and compared the average responses of residents in each of the five types of police stations. They found that in stations where all police officers had undergone training 16 percent fewer respondents reported fearing the police than in stations where there was no training. In addition, 31 percent more crime victims reported being fully or mostly satisfied with the police.  This field experiment not only demonstrated the success of the training program, but was also able to identify the magnitude of the impact of the reform.

This method of field experiments is widely used in development studies, since it provides an opportunity to assess the impact of various government policies. For example, Olken (2007) compares various anti-corruption strategies for road building in Indonesia and finds that the threat of inspection reduces corruption (increasing the probability of inspection from 4% to 100% reduces the unaccounted-for costs by 8%), while the possibility of public control has no discernable impact.
Normally, to perform a field experiment, the research team reaches an agreement with the agency that implements government policies. The agency will then identify the group which is the target for the selected government policy and the control groups. Specific measures of the selected governmental policy are chosen randomly for each group. Then these measures are implemented, the researchers collect data on the behavior of actors in different groups, and compare groups with each other (Humphreys, Weinstein 2009).

Further reading:

Banerjee, Abhijit, Raghabendra Chattopadhyay, Esther Duflo, and Daniel Keniston. Rajasthan Police Performance and Perception Intervention, May 2009 (draft), http://www.povertyactionlab.org/sites/default/files/publications/118_Duflo_Rajasthan_Police_Performance.pdf

Benjamin A. Olken. Monitoring Corruption: Evidence from a Field Experiment in Indonesia // Journal of Political Economy, vol. 115, no. 2, 2007

Macartan Humphreys, Jeremy M.Weinstein. Field Experiments and the Political Economy of Development // Annual Review of Political Science, Vol. 12, pp. 367-378, 2009