The large population in the health plan allowed controls to be matched to PCR-positive children on many potential confounders, and matched controls were more similar to PCR-positive children than were PCR-negative controls on all measured potential confounders. However, matched controls were probably not as similar to PCR-positive children as PCR-negative controls were with respect to unmeasured potential confounders, such as the propensity to have undergone a PCR test to detect pertussis. Because we believe that such unmeasured confounders were probably a greater source of bias than the ones we were able to measure, we considered the analysis involving PCR-negative controls to be more informative.