From a lively lunchtime discussion:
We know, from the experiences of the Standardized Aptitude Test (SAT), commonly used for college admissions, and from the folks that develop the various IQ tests, that aptitude tests can exhibit a bias against people who come from a different ethnic or cultural background (i.e., different life experiences and emphasis) than the test creators. For reference, let's refer to this type of bias as Test Bias.
We also have experience that education is not evenly distributed (at least in America). We have experience that where schools are poor, where often minority ethnic or cultural groups reside, that the education they receive is not as effective at imparting the skills and knowledge they need to pursue the more technical and often higher paying jobs. Let's refer to this type of bias as Education Bias.
Assume that you are the head of HR for a large, national corporation. You've observed that a significant minority, comprising 25% of the population, is represented at only 5% in you company. You've gone to the effort of developing a pre-screening test that you administer to all applicants to weed out those unlikely to succeed based upon their not having the required skills and knowledge. Since you are large (over 70,000 employees), you find yourself in the position of hiring nearly 3000 new college graduates each year. After the recent such hiring, you sit back and look at the numbers.
Nearly 20% of the applicants did identify with the minority group, so the percentage of applicants was not too much lower than their representation in the population. However, you observe that only 1/2, 10% of those applicants who passed the pre-screening skills test and qualified for a personal interview where of the minority.
The question you have: Is it true that a much lower percentage of the minority group have acquired the necessary skills and knowledge to work in your business (evidence of Education Bias), or is your aptitude test biased against this group? How would you go about discerning which?
(Extra credit: After the personal interviews, you observe that only 5% of the applicants who receive offers are of the minority group. Is that further evidence of Education Bias (the minority applicants do not have the necessary skills and knowledge), or is your company reducing its minority representation through a 'personal' bias exhibited by your interviewers?)
I just read a good article that was very similar to this.
ReplyDeleteAssume (fairly certainly) that intelligence within a group of people is bell-curve shaped. Then, assume that one group (such as a minority) has a slightly smaller mean, but the same deviation (the "lower" group). If you draw a cut-off at a high point with a test, you'll find that the "lower" group has a vast minority of people passing the test.
Now let's assume that you've hired a set of people in the "lower" and "higher" groups. If you test for promotions, the same result will be seen again: fewer people in the "lower" group will pass and better positions will be filled mostly by the "higher" group. Almost paradoxically, if you lower the bar for promotions, you'll find that the "lower" group starts to make up an even larger percentage of the lower positions.
This article was describing that bias is real, but that we shouldn't confuse it with a natural mathematical definition.
For me this means that the real solution is to try to get the bell-shaped curves to overlap. But even if we start today, we're talking about a 15 year process.
I would be curious to see the article, too: Perhaps you could post a link?
ReplyDeleteMy contention is that it may be very difficult, if not impossible, to discern the the cause of a difference in the means: Is it a real difference between the two groups (i.e., a lack of solid education for the minority, which leaves the mean lower, even though there are some exemplary individuals who track to the upper tail), or bias on the part of society or the company? I think that bias would appear to have shifted the mean lower for the group suffering from bias...So, I think that bias would appear mathematically identical, and wonder how the author of the article would propose we separate the two (mathematically, of course!)