Benchmarking sex ratio at age zero as a data quality indicator for small populations: Using census 2000 as a case study

Sex ratio at birth is one of the simplest tools for evaluating the quality of population data on sex, but its usefulness as a quality-control indicator depends on knowing how much deviation from the expected value is normal statistical noise versus a genuine data problem. This paper develops that benchmark: using Census 2000 county-level age-zero population data, it models expected sampling variability in the sex ratio as a function of county birth counts, and compares it to what is actually observed, with particular attention to the 22% of U.S. counties with populations under 10,000. Results show that observed dispersion is closely, though not perfectly, explained by binomial sampling variability: the overdispersion factor is near 1 (0.91 to 1.10) for counties with populations of 50,000 and above, but rises to 1.77 for the smallest counties. Applying exact binomial control limits, appropriate given the small birth counts involved, 0.45% of counties fall outside the 99.73% control limits (the exact analogue of ±3 standard deviations), closely matching a separate 2010 replication (0.32%). The result is a practical decision rule—a funnel-plot-style control boundary—that flags counties whose sex ratios warrant closer review, rather than relying on intuition about what “looks extreme.”