Federal contractors covered under Executive Order (EO) 11246 are required to conduct several quantitative analyses as part of their affirmative action program (AAP). To determine in which job groups minorities and females are underutilized, 41 C.F.R. § 602.15 directs contractors to compare the percentage of women and minorities currently employed in each job group against the calculated availability for each job group to determine which groups have fewer women and minorities than would be reasonably expected.
However, the regulations do not tell contractors how much of a difference constitutes “underutilization.” This can be difficult when contractors attempt to establish placement goals for underutilized job groups. It is important to remember goals are NOT quotas, but are reasonable targets for representation that may be obtained through a variety of good faith efforts. This blog will discuss several acceptable methods for setting a utilization threshold.
As the name suggests, this method considers any difference between the availability and incumbency rates to indicate underutilization. While it is technically accurate to say that there is a difference, OFCCP‘s stated policy is that exactly proportional representation or equal outcomes are not what EO 11246 requires. Typically, a contractor would prefer to adopt a difference threshold that is more meaningful.
Whole Person
One modification to the Any Difference rule is to require that the difference between the availability and incumbency represent at least one whole person. The rationale is essentially that with a shortfall of less than one, while the representation of women or minorities in the group is not exactly equal, if you were to change the race or gender of just one person, that group would actually then be overrepresented. This method may be appropriate for smaller job groups, where one or two employee changes can make a big difference in the demographics. However, for larger job groups, a difference of one person may not carry much practical significance.
80% Rule
Contractors may also choose the 80% rule of thumb commonly used in adverse impact analyses. If the incumbency rate for the protected class is less than 80% of the availability, the protected class is underutilized. Larger job groups will almost always have a shortfall of more than one person, so the 80% rule may reduce the number of flagged large job groups. However, with very large groups, this method may permit some fairly large shortfalls.
Standard Deviation
While the above methods are useful in determining the size of the difference between the availability and incumbency percentages in each job group, statistical testing allows us to determine the likelihood that the difference represents underutilization rather than a chance occurrence. The statistical threshold most often used by the courts and OFCCP is the “two standard deviation rule,” where a difference of two standard deviations or greater is considered statistically significant. This method is likely the preferred method with job groups of 30 or more people, although you will notice that with very large job groups, even small differences will often exceed two standard deviations.
Exact Binomial
While the two standard deviation rule is widely used for large job groups, it is not appropriate for smaller job groups. Practitioners wanting to use statistical tests on small job groups can use the exact binomial test to calculate a probability that the incumbency rate for women or minorities is less than would be reasonably expected given the availability percentage. Probabilities less than or equal to 0.05 are interpreted as underutilization.
The table below shows 4 different job groups, all with calculated availability of 45%. Notice that even though job groups A, B, and C all have the same incumbency rate, the results of the different utilization tests vary with the size of the job group. Also note that while job group D violates the 80% rule, the difference does not surpass the 2 standard deviation threshold.
Job Group

Availability

Incumbency

Any Difference?

Whole Person?

80%

2 SD?

Exact Binomial?


A

45%

4

40%

yes

no

no

no

no

10


B

45%

40

40%

yes

yes

no

no

yes

100


C

45%

400

40%

yes

yes

no

yes

N/A*

1000


D

45%

21

35%

yes

yes

yes

no

yes

60


*Because of the way the exact binomial is calculated, specialized statistical software is often needed to calculate the probability with large job groups
By Dave Sharrer, M.S., Associate Consultant and Yesenia Avila, M.P.S., HR Analyst, DCI Consulting Group
