Union Numbers Are Accurate Nationally, Less So by State

Published June 1, 2008

The modern era of the Current Population Survey (CPS), on which the Bureau of Labor Statistics’ union membership numbers are based, began in 1983 when the survey methodology was changed. There were no data on unionism in 1982, and CPS data before 1983 are inconsistent with the later data.

At the national level, CPS data are very accurate, with a confidence rating of about plus or minus 0.2 percent. They are, however, survey data, and therefore subject to the statistical and sampling errors inherent in all survey data.

Five-Year Average a Better Measure

At the state level, particularly in lower-population states and when looking at a sector of employment within a state, the report can’t be as accurate as on the national level. This sometimes leads to wide fluctuations from year to year.

Those fluctuations can be reduced and the data made more accurate by applying a five-year moving average. The moving average, however, tends to backload trends. If the trend is downward, the figure for the average will be higher than is probably the case, and vice versa.

For example, from the raw numbers it appears that between 2006 and 2007 public-sector union density in West Virginia fell from 30.0 to 22.5 percent. That probably didn’t happen. The sample size for West Virginia’s public sector in 2007 was 423, giving a confidence rating of plus or minus 4.7 percent. Applying a five-year moving average produces figures of 27.5 for 2006 and 26.9 for 2007, a drop of just 0.6 percentage points.

David Denholm