The authors make their point that we may not detect slope-based differences with significance testing when the sample size is small or we lack statistical power to find these differences to be statistically significant. While I realize that we are making important decisions and inferences based on selection measures, what is the true impact of a .005, .01 difference in slope on the likelihood that an individual of one racial/ethnic/age/gender group or another is hired? What is the likelihood that we can find ANY predictor that will not produce some sort of difference in slope?
Additionally, I feel like the authors presented this issue and then gave pretty general and not terribly helpful advice in terms of how to deal with the problem – increase predictor reliability (yes, we know that and it’s not easy to do), and minimize differences between groups (exactly how are we to do this?).
Has the prevalence of ADA Act of 1964 naturally minimized the differences between groups? In addition, it would seem that individual differences would limit our ability to, as recommended, minimize differences between groups. Especially since there is evidence of racial differences in IQ (U.S. definition).
I am not sure I understood this article. Are the authors saying that if their simulations are correct, then subgroup mean differences in test scores of GMA may be exaggerated, but that GMA may be less strongly correlated with performance for some subgroups? If this is true, it is rather good news for minorities who have believed that they tend to score lower as a group on GMA tests than do Whites, but bad news for those of us needing to predict performance based on a test such as GMA. (Roni, it would sure be great if you would go over this article in class tomorrow.)
I too had an issue with their recommendations. Decreasing the differences between groups seems like it is out of the hands of selection researchers and practitioners. Do you think is something we could do with training? That is, select minority individuals who may do less well on our selection measures and then make up for it with training. Is this legal?
To me it seems like they are recommending a couple of things - the importance of using power analysis to determine adequate sample sizes for testing in this area, but the other is more radical - really questioning the whole approach and advocating a change in direction for HR selection. They advocate for "an expanded view of staffing process that considers in situ performance and the role of time and context" described in the article we read the first week on staffing twenty-first century orgs (Cascio & Aguinis, 2008). What do you think of this recommendation?
Because they found that performance is over predicted in minority groups, should this be even more evidence to preclude the use of Thorndike model of fairness (a model that allows for increased selection of minorities compared to Cleary) when it comes to selection?
Maybe I am not understanding this article. But did these authors just make up parameters and then make up data and say they just revolutionized the test-bias in selection research? I don't understand how they can come to such extreme conclusions? I'm with Shay, I think we should talk about this one a little more in class.
The authors make their point that we may not detect slope-based differences with significance testing when the sample size is small or we lack statistical power to find these differences to be statistically significant. While I realize that we are making important decisions and inferences based on selection measures, what is the true impact of a .005, .01 difference in slope on the likelihood that an individual of one racial/ethnic/age/gender group or another is hired? What is the likelihood that we can find ANY predictor that will not produce some sort of difference in slope?
ReplyDeleteAdditionally, I feel like the authors presented this issue and then gave pretty general and not terribly helpful advice in terms of how to deal with the problem – increase predictor reliability (yes, we know that and it’s not easy to do), and minimize differences between groups (exactly how are we to do this?).
Has the prevalence of ADA Act of 1964 naturally minimized the differences between groups? In addition, it would seem that individual differences would limit our ability to, as recommended, minimize differences between groups. Especially since there is evidence of racial differences in IQ (U.S. definition).
ReplyDeletemeant to say both ADA Act of 1990 and the Civil Rights Act of 1964(Title VII)
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteI am not sure I understood this article. Are the authors saying that if their simulations are correct, then subgroup mean differences in test scores of GMA may be exaggerated, but that GMA may be less strongly correlated with performance for some subgroups? If this is true, it is rather good news for minorities who have believed that they tend to score lower as a group on GMA tests than do Whites, but bad news for those of us needing to predict performance based on a test such as GMA. (Roni, it would sure be great if you would go over this article in class tomorrow.)
ReplyDeleteI too had an issue with their recommendations. Decreasing the differences between groups seems like it is out of the hands of selection researchers and practitioners. Do you think is something we could do with training? That is, select minority individuals who may do less well on our selection measures and then make up for it with training. Is this legal?
ReplyDeleteTo me it seems like they are recommending a couple of things - the importance of using power analysis to determine adequate sample sizes for testing in this area, but the other is more radical - really questioning the whole approach and advocating a change in direction for HR selection. They advocate for "an expanded view of staffing process that considers in situ performance and the role of time and context" described in the article we read the first week on staffing twenty-first century orgs (Cascio & Aguinis, 2008). What do you think of this recommendation?
ReplyDeleteBecause they found that performance is over predicted in minority groups, should this be even more evidence to preclude the use of Thorndike model of fairness (a model that allows for increased selection of minorities compared to Cleary) when it comes to selection?
ReplyDeleteMaybe I am not understanding this article. But did these authors just make up parameters and then make up data and say they just revolutionized the test-bias in selection research? I don't understand how they can come to such extreme conclusions? I'm with Shay, I think we should talk about this one a little more in class.
ReplyDelete