As we have discussed in previous blogs, federal contractors are required to use the 2006-2010 American Community Survey (ACS) data to estimate external availability in Affirmative Action Programs (AAPs) dated January 1, 2014 or later. Now that we have been using these new data for a few months, there are some interesting aspects to this data we have observed and would like to share.
One of the key differences between the 2000 Census data that contractors used prior to this year and the ACS is that the ACS is a sample of approximately 1% of the population. Conversely, the 2000 Census data was a survey of the entire population. As a result of the smaller sample size, the ACS data can be extremely unreliable, particularly in less densely populated areas or when referencing uncommon census occupational codes.
Why is this data so important? Every year federal contractors must assess their employment of females and minorities by conducting a utilization analysis. This utilization analysis is performed by assessing the availability of qualified females and minorities both within the organization (i.e., potential for promotion) and external to the organization (i.e., available within a reasonable recruitment area). The ACS data provides the external availability figures, which directly affect whether the contractor appears to be underutilizing females or minorities in any given job group.
For example, consider a contractor with a warehouse in a rural county. This contractor has several engine and other machine assemblers, a rare occupation in this county. When the contractor needs to fill one of these positions, the contractor recruits employees from within the county. Thus, when calculating external availability, the contractor would pull the percentage of females and minorities the ACS data reports for engine and other machine assemblers within this county.
Because the ACS is a sample of about 1% of the population, if there are fewer than 100 individuals in the area who match that census occupational code, it is likely that only one, or even none, of those individuals would be selected to be part of the ACS sample. If there is only one response and that person is a Black female, then the external availability for females and minorities will be reported as 100%. If that individual is a white male, the external availability of females and minorities will be 0% because the estimate is based only on those individuals sampled by the survey, not on the full demographics of the area.
Even worse, it is possible that the 1% sample did not survey any of the individuals in that occupational code. This may cause computational headaches as some AAP software will have difficulty computing a non-existent figure. In the example above, the federal contractor may be analyzing utilization against 100% availability, 0% availability, or a null value, because of the level of granularity involved in the combination of the occupational code and the rural county.
Because the ACS is a survey, and not a census, there is a varying amount of uncertainty involved in the estimates it provides. As the population in the geographic area increases, the uncertainty in the estimate decreases because there is more data factored into the estimate. The ACS includes margin of error calculations to help identify how uncertain each estimate is. For example, Allen County, Indiana is estimated to have approximately 179,000 people. There are 220 sheet metal workers (code 6520) estimated to work in Allen County, but the margin of error is +/- 113 (or +/-13.7%). That’s a potentially big difference. Now, if we look at the number of sheet metal workers in all of Indiana, we get 3,375, with a margin of error that is +/- 342 (or +/-1.0%), meaning we are more certain that the estimate is sound. By the time we expand to the entire United States, the margin of error is down to +/-0.1%.
It may also be the case that in areas dominated by a single employer in an industry, all respondents in the ACS data with those Census Occupational Codes are in fact already working for that organization. If that is the case, the "external" availability estimate is really just a sample of the employees already included in the "internal" availability figures.
Contractors who experience issues with the ACS data may be able to employ strategies to deal with these problems. One strategy is to expand the external recruitment area, especially if a more specific area is returning vastly different estimates from previous years. By adding neighboring counties or expanding to include an entire metropolitan statistical area (MSA), the sample size may increase enough to provide more reliable estimates of the population. It may also be necessary to establish or adjust recruitment area weighting schemes to more accurately reflect actual recruitment practices, if few applicants actually are sourced from certain areas. In addition, it may be helpful to re-evaluate census occupation codes to make sure they are accurate, or to reassess whether a more broad/common occupational code better fits some positions, allowing for more reliable external availability data.
As we continue to use the ACS Census data, we will share any additional observations and best practices. Stay tuned!
By Dave Sharrer, M.S., and Kristen Pryor, M.S,. Associate Consultants at DCI Consulting Group