Key message 3 – Performance incentives can backfire – particularly where desired outcomes are broad or difficult to measure
Performance pay is more appropriate and effective for jobs where the desired outcomes are easily defined and measurable. It is important for incentives to be linked to behaviours or outcomes within the scope of what workers can affect, and incentive schemes should be simple enough to be widely understood. When outcomes are harder to quantify or longer term in nature, performance incentives can lead public servants to focus too much on performing well on measurable outcomes at the expense of a broader range of important but more difficult-to-track goals.
Recent experiments illustrate some of these tradeoffs. One example is a randomised trial that rewarded some Kenyan teachers with in-kind prizes based on school-level performance on government exams (Glewwe, Illias and Kremer, 2010). The incentivised schools achieved higher scores on the government exam but did not score higher on a separately-administered independent exam. The researchers concluded that teachers may have increased emphasis on test-taking skills for the national exam rather than improving general instruction. Similarly, a policy change in the United States that allowed police agencies to keep revenues obtained from assets seized during drug arrests resulted in the police performing more drug arrests but reduced enforcement of other types of crimes (Baicker and Jacobson, 2007).
Research on the Nigerian civil service shows that incentives that reward ‘box ticking’ can reduce bureaucrats’ performance (Rasul and Rogger, 2013). However, in some cases, intermediate input measures can function as a proxy for outcomes – for instance, incentives for teachers may be based on teacher attendance rather than student learning because it is easier to monitor. One study in India found that teacher incentives based on attendance also increased learning (Duflo, Hanna and Ryan, 2012). In contrast, a similar study on financial incentives for nurses’ attendance in India found that incentives were not effective because the attendance reporting system was undermined over time (Banerjee, Glennerster and Duflo, 2008). In general, incentives can be based on inputs or outcomes depending on the nature of the task and how well the inputs or outcomes can be monitored.
An overall lesson is that incentives need to be pilot-tested and designed for each specific context, with care to avoid inadvertently incentivising perverse behaviours or ‘gaming the system’. A related challenge is how management practices can give greater autonomy and discretion to public sector workers in ways that motivate better performance rather than resulting in corruption or worse outcomes.