Like our site's new design? In April 2023, Circa was acquired by Mitratech.
Learn More

The economy is going strong and the unemployment rate is at an all-time low. The last thing on anyone’s mind is workforce downsizing or Reduction in Force (RIF). However, for the experienced Human Resources (HR) practitioner, s/he understands that RIFs do not occur solely during economic downturns. RIFs are a normal and (arguably) healthy part of maintaining a high performing workforce that supports organizational strategy.

Although the employment condition is largely understood as “at will,” termination actions are not without legal exposure. As one can imagine, there are laws written to protect employees’ rights against potential discrimination in RIFs1. Assuming the termination decisions were not intentionally discriminatory (disparate treatment), RIFs can nonetheless be evaluated within an Adverse Impact (AI) framework for discrimination. In plain English, claimants can establish a prima facie case of AI discrimination with statistical evidence of termination rate differences between groups (e.g., gender, race, age).

Classic RIF Analysis

Analyzing for AI in gender and race is fairly straightforward. This can be evaluated through traditional selection rate analyses – classic 2×2 analyses, where there are 2 groups (e.g., male, female) with 2 outcomes (e.g., pass, fail):

Table 1. Sample 2×2 Selection Rate Analysis Table

  Pass (Retained) Fail (Terminated)
Male 8 5
Female 8 9

In this example, there are a total of 13 males (8+5) and there are 17 females (8+9). The retention rate for males is 62% (8 out of 13) and the retention rate for females is 47% (8 out of 17). The female retention rate (47%) is lower than the retention rate for males (62%), but is this difference statistically significant? This question can be answered using Fisher’s Exact Test (FET) to evaluate the observed 2×2. For this example, the FET finds that the selection rate differences are not statistically significant (p=0.48)2, which undermines AI discrimination claims.

Age-Based RIF Analysis

RIF analysis among “crisp” and well-defined categories (e.g., race, gender) is easy and straightforward – simply apply the 2×2 selection rate analysis method. For age-based RIF analyses, this is less clear since age is on a numeric continuum; fortunately, the Age Discrimination and Employment Act (ADEA) has defined individuals who are 40 years old and above as a protected group. Applying this law, many analysts have set up the two groups in their selection rate analyses as: Less Than 40 (<40) and 40 or Greater (≥40).

Table 2. Sample 2×2 Selection Rate Analysis Table for Age (FET p=0.48)

  Pass (Retained) Fail (Terminated)
<40 8 5
≥40 8 9

These 2×2 selection rate analyses were also straightforward and fairly simple. However, after the O’Connor v. Consolidated Coin Caterers Corp. Supreme Court decision (517 U.S. 308, 116 S. Ct. 1307, 134 L. Ed. 2d 433 [1996]), age-related AI discrimination cases have become slightly more complicated. The Supreme Court recognized that age is a continuum and, more importantly, noted that if age can be linked to employment decisions, then those decisions are eligible to be scrutinized for AI discrimination, regardless of class membership (<40, ≥40).

In practice, what this means is that 2×2 selection rate analyses may not necessarily be the most probative analytical method for evaluating age-based AI discrimination after the O’Connor v. Consolidated decision. Instead, regression-based methods are more appropriate for age-based AI discrimination investigations. Regressing age into termination status (retained/terminated), one can evaluate the relationship between termination status and age. Applying regression methods to evaluate the data in Table 2, we find the following:

Table 3. Regression Descriptives of Table 2 Example

  Retained Terminated
Count 16 14
Average Age 40 years 51 years

A regression analysis on this data3 would find that on average, terminated employees were 11.54 years older than retained employees (p=0.002, SD=3.34). At this point, it would be natural to wonder how is it possible that the regression-based findings are significant, while the 2×2 analyses are not significant? The answer is quite simple. The regression analysis was able to evaluate age and all the rich, numeric information associated with age. In the 2×2 analysis, on the other hand, all that age information was collapsed into just two categories (<40, ≥40). Consequently, the regression analysis had significantly more information and statistical power to detect group differences.

Another notable advantage of regression-based methods is the ability to control for explanatory factors (e.g., performance). Although the above example did not include an explanatory factor in the regression analyses, inclusion of explanatory factors is fairly simple to model. This is a significant advantage over traditional 2×2 analysis, which does not easily lend itself to controlling for explanatory factors.

Conclusion

An important take-away from this paper should be that simple 2×2 selection rate analyses may not be the most probative analysis in age-based AI discrimination investigations, especially after the Supreme Court O’Connor v. Consolidated decision. Note that the 2×2 retention analyses were not statistically significant (p=0.48), while the regression analyses, which accounted for gradations in age, were statistically significant (p=0.002, SD=3.34).

Currently, age-based AI discrimination is not a concern among federal contractors in their compliance efforts. This is because age is not a protected group covered by EO 11246. So, aside from describing the proper methods for age-based AI analyses, the second goal of this paper is to raise the awareness among federal contractors (and non-federal contractors alike) of major risk exposure they might not be aware of. This is particularly concerning due to the increase in age of the workforce as many employees have decided to delay retirement. Although the OFCCP does not currently have age-based enforcement authority on federal contractors, it still makes good business sense for employers to manage their risk exposure by conducting proper AI analyses in RIF situations. EO’s have expanded the scope of OFCCP’s enforcement authority (e.g. veterans, sexual orientation, gender identity) where necessary, and with an ever-aging workforce, it is only a matter of time before the OFCCP is empowered to protect against age-based discrimination. Who knows, when the OFCCP realizes how much age-based AI there is (in hires, promotions, and terminations), enforcement momentum can surely follow. Federal contractors are advised to get ahead of this wave.

1. e.g., Title VII of Civil Rights Act of 1964, Age Discrimination in Employment Act of 1967
2. Probability values of 0.05 or less are considered statistically significant (p≤.05).
3. Readers who wish to obtain a copy of the dataset used in this example are invited to contact the authors.

Author

Skip to content