There are two cases here. The first case was in 1993 and the second one covered 2002 to 2005. An Administrative Law Judge (ALJ) found Bank of America (BOA) (previously Nations Bank) guilty of a pattern or practice of discrimination of race discrimination in hiring for entry level jobs. BOA was ordered to pay $964,033 for back pay for 2003 and $1.2 million for back pay for 2002 to 2005. Basically, the ALJ ruled that there were statistically significant differences in applicants first interviewed and applicants then hired, even when controlling for critical covariates. Also at issue in the 1993 case was disproportionate automatic exclusions for code violations (i.e., credit checks and incompatible work schedules). However, as later determined by a three-panel review board, the code violation exclusions were not disproportionate.
The three-panel review board upheld the 1993 award, but overturned the 2002-2005 ruling (and $1.2 million award). The review panel was unanimous on the earlier ruling and divided two to one on reversing the 2002-2005 award.
The basis for upholding the 1993 award was the finding of statistical differences of 6.9 standard deviation (SD) (actually, the correct metric is standard error of the mean, but that is another blog for another time). As for the 2002-2005 case, the ALJ ruling was based on a 4.0 SD difference across the four years and the basis for the review board’s reversal was that there were no statistically significant differences in any of the four years. Another important factor is that the 1993 ruling was based on two stages of the hiring process (application to interview & interview to job offer), whereas the 2002-2005 ruling was based on only hiring rates. Another factor in the 2002-2005 reversal was the opinion of one of the board members that there was no legitimate basis for the OFCCP to extend its initial review to 2002-2005.
And as a final, but critical point, the review board noted that in 2003, there were 44 Black hires in comparison to a statistical expectation 47.9, and that in 2005, there were 32 hires in comparison to a statistical expectation 34.5. These were considered “small shortfalls”, which is code for lack of practical significance.
I have only one comment here. The fact that there was a statistically significant disparity in a four-year period, but not in any of the individual years, is a strike against data aggregation. The same would hold for the sum total hires to sum total expected hires. Data aggregation, or lack thereof, is itself an important factor we need to follow.
By Art Gutman, Ph.D., Professor, Florida Institute of Technology