Even though the work sample selection procedures were found to have adverse impact, these exercises seem to have more face validity and I would expect would be highly predictive of work performance. I wonder if this would be more predictive of typical or maximum performance?
How important is it for the graders of work sample tests to be blind to the race or gender of the applicant? Do you think this is possible for some forms of work sample tests?
This article made me think hard about the values we use and the conclusions we draw based on effect sizes. For instance, we say that an effect size (d) of .20 is small. But in a selection situation, where we are making important decisions about an individual, is an effect size of .20 really still a small value? Or do effect sizes take on another meaning or must they reach another standard when they are used in selection decisions??
In this article, d means 'standardized ethnic group differences associated with given selection test' (p.2). The authors go on to explain that standardization occurs using the pooled with-group SD and that a d=.5 indicates that the majority group scored, on average, one half of a pooled SD higher than the minority group. Is this another way of explaining effect size Vicki? I was left with this question. Thanks.
I thought that the range restriction corrections typically applied to validity studies based on job incumbents and then subsequently used in VG studies accounted for the impact of range restriction on estimates of adverse impact. This seems to be untrue. Does suggest we should be cautious when assuming that statistical corrections will overcome methodological limitations?
Even though the work sample selection procedures were found to have adverse impact, these exercises seem to have more face validity and I would expect would be highly predictive of work performance. I wonder if this would be more predictive of typical or maximum performance?
ReplyDeleteAre SJTs more valuable in that they have lower adverse impact than work samples?
ReplyDeleteHow important is it for the graders of work sample tests to be blind to the race or gender of the applicant? Do you think this is possible for some forms of work sample tests?
ReplyDeleteThis article made me think hard about the values we use and the conclusions we draw based on effect sizes. For instance, we say that an effect size (d) of .20 is small. But in a selection situation, where we are making important decisions about an individual, is an effect size of .20 really still a small value? Or do effect sizes take on another meaning or must they reach another standard when they are used in selection decisions??
ReplyDeleteIn this article, d means 'standardized ethnic group differences associated with given selection test' (p.2). The authors go on to explain that standardization occurs using the pooled with-group SD and that a d=.5 indicates that the majority group scored, on average, one half of a pooled SD higher than the minority group. Is this another way of explaining effect size Vicki? I was left with this question. Thanks.
ReplyDeleteI thought that the range restriction corrections typically applied to validity studies based on job incumbents and then subsequently used in VG studies accounted for the impact of range restriction on estimates of adverse impact. This seems to be untrue. Does suggest we should be cautious when assuming that statistical corrections will overcome methodological limitations?
ReplyDelete