Do incentives matter beyond the information conveyed by pay-for-performance contracts? Does loss framing matter? And do incomplete contracts generate spillovers on unincentivized tasks? This study reports on a framed field experiment with 1,363 maternity care workers in 691 primary health facilities in Nigeria to answer these questions. Participants were randomized into three study arms-(1) information with a flat participation fee, (2) performance-based rewards, and (3) performance-based penalties. In each arm, participants had to identify correct clinical actions based on the records of hypothetical patients receiving maternity care. Five of fifteen possible actions were incentivized but performance was measured on all fifteen. Compared to information alone, both rewards and penalties increase time on task by 11 percent, correct overall performance by 6 to 8 percent, and directly incentivized performance by 20 percent. Incentives also generate positive spillovers of 14 percent on unincentivized tasks. Loss framing does not affect performance. Results suggest that improving health worker effort by 8 percent would have an impact on neonatal mortality at par with the short run effect of adding a physician to a health facility. Finally, findings show that a small incentive captures most of the impact, implying that incentives work by making information more effective and that pay-for-performance contracts can be made significantly more cost-effective.