coefficients-of-variation

library(urbnindicators)

The Census Bureau API exposes an extensive set of estimates derived from responses to the American Community Survey. Each estimate has an accompanying margin of error (MOE). These MOEs are critical for accurately interpreting and responsibly using ACS estimates.

Margins of error account for the imprecision associated with extrapolating from a survey sample to the full population of interest. If a point estimate is 500, and the associated (90%) margin of error is 50, then we would expect that if we were to use the same survey design to calculate that same estimate 100 times, 90 of those 100 repeated survey efforts would produce point estimates in the range from 450 to 550. The remaining 10 iterations would produce point estimates outside of that range.

Because the ACS 5-year estimates (the only ACS data used in urbnindicators) accumulate sample over five years, they have smaller MOEs for a given geography and statistic than do the corresponding ACS 1-year estimates. Nonetheless, 5-year estimates rely on roughly a 3% sample of the national population. Thus can lead to estimates with small MOEs for large geographies and/or broadly-captured characteristics (e.g., the counts of different race groups in New York City). Conversely, this can lead to very large MOEs when looking at smaller geographies and/or less broadly-captured characteristics (e.g., the number of renter households that walk to work in a rural census tract in North Dakota).

When MOEs associated with a given point estimate become large relative to the magnitude of that point estimate, analysts should take pause and assess whether they can draw meaningful and valid inferences about the geography and characteristic in question. Yet the process of integrating MOEs into analytic workflows can be unfamiliar and time-intensive. This is especially true when analysts are working with derived estimates (e.g., the share of the population that is Black alone, non-Hispanic) rather than raw count estimates reported directly by the ACS (e.g., the number of people who are Black alone, non-Hispanic), because an analyst must create a derived margin of error to accompany their derived estimate.

urbnindicators helps to facilitate the incorporation of MOEs into analysts’ workflows through a few inter-related functions.