"Does Better Information to Civil Servants Improve Public-Service Delivery Outcomes?"
Theoretically, the effect of absolute performance feedback on effort is ambiguous.  This paper sheds light on this relationship using new data on the introduction of a performance-monitoring technology in the Ethiopian bureaucracy.  The technology provides local civil servants with accurate information on the returns to effort. The theoretical framework suggests that this should lead to an increase in subsequent effort under certain conditions; and that the magnitudes of the effects increase monotonically with the improvement in information.  The empirical evidence relates directly to the theory: over a three-year period, the technology leads to a 2.22 percentage-point increase in the enrolment of 7-14 year-old children in primary school, with no evidence of worse schooling conditions; and the effects are larger for districts with greater information constraints prior to the technology.  The results suggests that gains in enrolment are immediate, while it takes 2 years after the programme for local governments to improve schooling conditions, by hiring teachers and creating additional teaching space. Suggestive evidence on mechanisms indicates that formal incentives and changes in budgets are not driving the effects, but rather performance evaluations, non-monetary incentives, and career concerns. The programme costs around 12 USD per additional pupil enrolled, requiring minimal returns to education to pass any reasonable cost-benefit test.

"Efficiency Losses from Subjective Performance Evaluations and Supervisor Bias: Evidence from the Ethiopian Civil Service"
In contrast to standard agency theory, most compensation arrangements rely on subjective evaluations, conducted by middle-tier supervisors who are not residual claimants of output. This has the potential for efficiency losses through supervisor bias. This paper provides direct empirical evidence in this domain, showing that: supervisor bias correlates with joint tenure and gender, even conditional on actual performance; bias increases when there is more money on the line for the subordinate; a one-standard-deviation ‘over- valuation’ in the appraisal score is associated with approximately a 3% increase in wages; the wage effect increases by a further 4.5% if the manager is granted a one-standard-deviation increase in autonomy. At the organisation level, a one-standard-deviation increase in bias is associated with a reduction in service delivery of 0.04 standard deviations, equal to half of the gains from a one-standard deviation in management practices. Bias is more costly when the service delivery environment is more diverse, when there is less efficient monitoring in place, and when there is greater uncertainty in the service delivery environment. 

"When Does Decentralisation Improve Information?" (with Dan Rogger)