Confidence Intervals
The U.S. Census Bureau calculates 90% confidence intervals for its American Community Survey (ACS) estimates. Use caution when interpreting values with wide (or illogical) confidence intervals. Confidence intervals that are farther away from estimates in either direction or go beyond logical limits (for example, <0 or >100%) indicate uncertainty due to small survey sample sizes. Smaller geographic areas or areas with small populations are likely to have wide confidence intervals. The U.S. Census Bureau does not explicitly identify which values are unreliable or statistically unstable, but publishes such values due to the critical need for the ACS data as the only source of data for small geographic areas. The margins of error published with each estimate allow data users to make their own informed decisions regarding the reliability of the estimate.
Trend Comparisons for Overlapping Periods of Measurement
For smaller geographies, including zip codes and census tracts, the U.S. Census Bureau only releases ACS estimates for 5-year time periods. Aggregation of data across multiple years increases the reliability of the estimates and narrows the confidence intervals. Because each 5-year period overlaps with the prior time period, users should be cautious when comparing 5-year estimates over time. The U.S. Census Bureau does not recommend comparing overlapping 5-year periods (e.g., 2016-2020 and 2017-2021) since much of the data in each estimate are the same. Please visit Period Estimates in the American Community Survey for more information.
In consideration of the U.S. Census Bureau guidance, ACS 5-year estimates do not display the “Prior Value” comparison on the HCI Platform. This comparison, which considers two adjacent overlapping 5-year periods of measurement, is effectively only comparing the two “end years” of the time periods that do not overlap. Because the margins of error that are published by the U.S. Census Bureau represent the entire 5-year period for each data value, it is not appropriate to apply the statistical test to the comparison of these two values.
The HCI Platform utilizes the Mann-Kendall Test for Statistical Significance for the “Trend” comparison. This test does not utilize the margins of error published by the U.S. Census for determining significance. The statistical significance for this trend test is determined by the number of comparisons that are possible across multiple values over time. Although this test is not ideal as each data value is not completely independent of other values to which it is being compared, the results of the test can still provide value to a user who is trying to understand if an indicator’s values are trending in any particular direction over time. By applying the Mann-Kendall test to overlapping time periods, the results may be biased conservatively towards a lack of a trend either upwards or downwards.
Census 2020 Geographic Boundary Changes
Use caution when comparing periods of measure that include the year 2020 and later with earlier periods. Due to Census 2020, census tracts may have split due to population growth or merged as a result of substantial population decline. In general, data users should be aware of changes in geographic boundaries when making comparisons. For more information on geographic concepts used in the ACS, visit Geography and the ACS.