Implications of the 4/5’s (80%) rule: Check the speedometer while you drive, not after the police officer stops you!
Imagine you were caught speeding by a police officer. Think you can tell her that you were too busy to check the speedometer? My guess is that excuse won’t work! Same is true here….
So, one implication for companies about the 80% rule is to keep an eye on and monitor selection rates, promotion rates, and termination rates of different ethnic and gender groups at different facilities in your organization. Make sure you retain the necessary documentation so that you can monitor the 80% rule for these processes (the Dukes v. Wal-Mart case has some interesting information on this issue!).
If there is a problem (i.e., you are not meeting the 80% rule), you should try to figure out what is going on. Perhaps you need to improve your recruitment of qualified women or minorities. Or, perhaps you need to change how you make your selection, promotion, or termination decisions. There has been research, for example, on whether certain types of interviews or tests are less prone to disparate impact. Indeed, some interviews and tests are less prone than others. You would do well to determine what the barriers are here.
A second implication is that you should always keep in mind the importance of job relatedness in your selection, promotion, and termination processes. I will do some more in-depth writing at a later date on this subject, but for now, let’s say that you should be able to make a convincing case for job relatedness if your decision processes were challenged by EEOC, OFCCP, or a judge. If you are using a standardized test, you must be especially sure that it has been properly validated (see my earlier column for some suggested sources of information on this issue).
Remember, if your personnel process produces a selection, promotion, or termination rate that misses the 80% rule, then you must be able to show that the process is valid or job related, in order to support its use. How sure are you that you could argue that the process really is valid or job related?
FOOTNOTE: Just like in driving, one can get a ticket for things other than speeding. Similarly, there are other ways and approaches to assessing disparate impact, in addition to or instead of, the 4/5’s rule. Over the coming months, I plan to address those other ways and approaches.
If you enjoyed this post, please consider to leave a comment or subscribe to the feed
or Subscribe it directly to your email inbox.
Comments
Ah, George, you are always the cynic. I’d like to respond this way. Companies should always do their best to have job-related selection, promotion, and termination systems. So, if the company is doing a good job in this regard, they should have made sure that the system is job-related. If they are going to have barriers in place to various protected groups, they better have done their homework on job relatedness.
How does this fit with multivariate statistics? Well, when it comes to pay raises, for example, several factors may come into play, besides job performance (there is some good research on this), such as Point-in-grade, seniority, and possibly other factors. That is where multivariate statistics come into play. But for a test, the key is: Was the test really job-related or not?
Much as I’d like to make this a vicious debate ;-), I don’t disagree with Michael’s last comment one bit.
As someone knowlegeable in testing and selection procedure validation, Michael seems to think more in terms of impact of specific parts of a decisionmaking process, such as a test. Perhaps it would be useful to think of this as “micro” level impact analysis?
I was thinking more along the lines of the Wal-Mart case, about situations in which one looks at a “macro” level — comparing, for example, the minority population of the county or of an actual applicant pool to that of the workforce or that of employees hired or promoted in the last year, or holding certain positions. At this level, the picture is much more complex, and I suspect a good deal of challenging statistical detective work is required to determine causes of differences — and whether they may be justified based on job-related factors.
I like George’s distinction between macro and micro levels of analysis of disparate impact. I think that indeed the micro analysis (e.g., 4/5’s rule) is alot “cleaner and tighter” than the macro analysis. In future postings, I will delve into the macro analyses that have been used, along with specific case examples!
Stay tuned!
Call me a cynic, call me a realist, or just call me George, but here’s my thought.
Life and society being what it is, I suspect “violations” of the 80% rule are about as common as drivers doing > or = 70mph on a 65mph highway. (Which is to say, expect to find them everywhere you turn.)
Striving to improve recruitment and to improve job relatedness in your selection, promotion, and termination processes, as Michael advises, is good business sense — don’t you want to be utilizing the best people to do what they do best?
But this focus on statistics can all too easily become a quota system in which there is reverse discrimination against better candidates, in order to improve the numbers. This is not only unfair and unlawful, but also contrary to the “best people” principle mentioned above.
The speeding analogy only goes so far. You can avoid a ticket only by slowing down. As we shall see in more detail when we explore the case law, you can, at least theoretically, avoid disparate impact liability without necessarily dramatically improving the numbers, if you can “make a convincing case for job relatedness” (quoting Michael).
Part of my cynicism arises from having read cases in which job relatedness has proven virtually impossible to prove. Tests and standards seeming quite fair to the layman (me) were nitpicked to death by experts.
Part of it arises from the fact that I suspect there are always so many other factors in play (besides race, gender, etc.), that frequently even “perfect” job-relatedness will not yield “perfect” numbers.
This brings us back around to multivariate analysis, and a legal question the answer to which I haven’t a clue (yet): where in the disparate impact legal analysis would one fit a multivariate study that showed, for example, that once certain other factors were taken into account, the numbers fell within accepted standards?