The reliability of the American Community Survey for injury surveillance
To evaluate the reliability and predictability of 49 socioeconomic indicators constructed from the annual and multiyear American Community Survey (ACS) data cycles for monitoring injury inequalities across the USA.Methods
Cross-sectional analysis of the 2006–2013 annual and multiyear county-level ACS data cycles. Indicator reliability was assessed using the margin of error and coefficient of variation (CV). Overlapping multiyear data cycles were assessed for statistical dependence in the estimates. Negative binomial regression models were constructed from a selection of the most reliable indicators over time and across all data cycles using all-cause unintentional and homicide-related mortality records from the National Center for Health Statistics (NCHS).Results
Fewer than half of all indicators for each data cycle generated ‘high reliability’ CV estimates for at least 95% of all census counties. Indicator reliability did not linearly improve with increasing sample size afforded from the multiyear surveys. On average, changes in socioeconomic conditions for the same geographic areas were statistically significantly different (p<0.05) in 14% (rage 0–99%) to 16% (rage 0–93%) of all overlapping multiyear data cycles. ACS indicators that were among the most reliable across data cycles corroborated variable relationships derived using estimates from the 2000 decennial census and corresponding NCHS records for that year.Conclusions
Few of the socioeconomic indicators previously used to measure injury disparities are consistently reliable across all ACS data cycles. Researchers should be judicious when selecting consecutive multiyear data cycles to approximate changes in annual socioeconomic conditions. Among the indicators that are reliable, it is advisable to use estimates from the annual ACS data cycle as a crude barometer of injury inequalities and the multiyear files to confirm and add precedence to national trends every three and five years.