Do Performance Agreements Help Improve Service Delivery? The Experience of Brazilian States

Produced by: 
The World Bank
Available from: 
July 2015
Paper author(s): 
Lorena Viñuela
Laura Zoratto
Education - Health
Microeconomics - Competition - Productivity
Fiscal Policy - Public and Welfare Economics

A growing number of states and municipalities in Brazil rely on results-based management, and many other local and state governments are considering adopting the practice.  This paper examines the experiences of the Brazilian states that have implemented results agreements linked to variable pay. The analysis compares current with pre-intervention outcomes in the education, health, and security sectors. The changes are examined in relation to regional trends to determine whether the improvements depart in meaningful ways from the overall trend. In addition, a truncated time-series cross-section model is used to control for several additional factors influencing service delivery outcomes. The results suggest that, at least in the short and medium term, the implementation of results agreements is associated with significant and positive changes in outcomes in the security and education sectors. On average, states using team-level targets and performance-related pay have 15 fewer homicides per 100,000 inhabitants than those that do not, all else equal. Similarly, states that have introduced performance agreements and a bonus for teachers and school staff have improved their Basic Education Development Index score for public secondary schools by 0.3 additional points compared with the scores of states with similar characteristics. The conclusions are in line with the findings of in-depth impact evaluations and case study work in the education and security sectors (Bruns, Evans and Luque 2011, Milagres de Assis 2012). The paper does not analyze unit or team level data, which would be necessary to draw more rigorous conclusions about how results based interventions affect the behavior of civil servants and outcomes over time. Therefore, the results should be interpreted with caution, as some of the assumptions behind the models cannot be examined with the available data.


Research section: 
Latest Research
Share this