The 80% Rule is highly sensitive to the way in which the question is framed. Opposing inferences can be generated from the same underlying data depending on whether one is examining positive employment outcomes (e.g., selection) or negative employment outcomes (e.g., rejection). As a result, the 80% Rule can easily be misapplied or misunderstood.

For example, some courts have applied the 80% Rule to rejection rates, rather than selection rates, without any appreciation of the difference. The following example, taken from Paetzhold and Willborn's Statistics of Discrimination, illustrates this issue.

In all five of these examples, the minority to majority selection rate is 80%. Applying the 80% Rule to any of these selection rates, one would conclude that no disparate impact exists. But the *rejection* rates differ quite dramatically between the five cases. The rejection rates are also affected by whether we take the majority rejection rate as the denominator (column 4), or whether we take the minority rejection rate as the denominator (column 5).

According to Patezhold and Willborn, some courts have found disparate impact based on the proportion of favored and disfavored group members in the pool of rejected applicants. As the following example demonstrates, this alone is not enough information to make a determination of disparate impact.

Based on the proportion of rejected minority applicants and rejected majority applicants alone, one may be tempted to infer disparate impact. But in order to make any inferences* *using the 80% Rule, one needs to know the total number of minority and majority applicants. In the above example, if we assume 100 minority applicants and 100 majority applicants, we infer disparate impact, since 29.9% is less than 80%. However, if we assume 1000 minority applicants and 1000 majority applicants, we *do not* infer disparate impact, since 93.18% is greater than 80%.

Note that in this example, we have two opposing inferences regarding disparate impact based on the same rejection information. This demonstrates why having just rejection information alone is not enough. It also demonstrates that the application of the 80% Rule to rejection rates, rather than selection rates, can lead to Type I and Type II errors.

There are a variety of other statistical tests available that are insensitive to the framing of the question, and work equally well whether one is examining positive employment outcomes (e.g., selection) or negative employment outcomes (e.g., rejection). A review of these alternative statistical tests will be provided in a subsequent post.

### Shortcomings of the 80% Rule - Highly Sensitive to Framing of Question

Subscribe to:
Post Comments (Atom)

## 0 comments:

Post a Comment