Comparing health risks that include ratios of integers (e.g., 12 in 1,000) is challenging. We tested whether a worked-example intervention with number-line visual displays improved adults' risk-comparison accuracy, whether pretest confidence moderated learning, and which individual differences related to accuracy. Replicating prior work, U.S. adults randomly assigned to the intervention (n = 883) were more accurate than control participants (n = 949) at solving health-related math problems accompanied by number-line visual displays and a transfer problem without a visual display. One day later, most participants were accurate and there were no differences between conditions. However, there was a small effect on accuracy one day later among those who learned from the intervention. Adults were more likely to learn from the intervention if they made a low- relative to high-confidence pretest error. Identifying as male, more accurately estimating numbers on number lines, lower math anxiety, higher educational attainment, and being older were associated with greater accuracy. [This paper will be published in "Journal of Applied Research in Memory and Cognition."]