•I’m aware that companies have to provide proof of the validity of their selection procedures, but this article showed how it is also important in the statistical analyses they use to determine if adverse impact had occurred. Are companies required to keep track of the analyses they run on their selection decisions in order to prove that adverse impact did not occur? In addition, should the court require selection personnel to understand these types of statistics in order to prevent adverse impact during hiring?
Although the authors recommended using the Z-test in testing statistical significance with small samples, they also recognized that low power remains a problem with significance testing small samples. Are confidence intervals accepted by the legal system? When making decisions about the presence of adverse impact, how much weight is placed on mean differences and on statistical significance inferences?
Katie's comment made me think about when these statistics are actually run. Are these only done I'd an AI case is brought to the courts? I would guess most HR personnel don't know how to do these analyses, would they hire a statistician to do this kind of work for them?
One of the specific recommendations here was to use one-tailed tests to improve power since we know the direction of the prediction. How much does this help? This article is another example of the seriousness of research in this area searching to improve the use of test statistics for the purpose of helpful selection in the area of employment. What a challenge - 40 years of this. Will computer modeling techniques being worked on now help us with this challenging area?
We learned last semester in methods that even professionals in research make mistakes when it comes to applying the correct statistical test and interpreting results. The issues brought up in this article are quite statistically meaningful, but complicated and need a careful, statistically trained person to apply the right test. Is it practical that organizations would have this type of person? Or would this be done by an outside consultant? ....In other words, as Dr. T once wrote on my paper "what is the utility of this!?!" or more appropriately when is this utilized?
One specific recommendation the authors make is to use one-tailed test to gain power. How much does this help?This article is an example of the seriousness of research searching for improved use of test statistics for the purpose of helpful selection. What a challenge - 40 years of this. Will new forms of computer modeling being developed now help us with this in the future or do we need to put more thought into other approaches as Aguinis et al. (2010) propose?
When the sample size is so small, perhaps alternative predictors (such as job experience, references) should also be considered? If your sample size is already small, it is probably practical to look for additional selection criteria. What do you think?
The authors address the issue of what is to be considered a representative sample of job applicants when doing significance testing. Certainly, it seems unreasonable to fault an organization for low selection ratios of minority groups (e.g. nurses) if the ratio of minority to majority graduates from local B.S.N. programs shows an underrepresentation of minorities graduating from nursing school. Doesn't it seem reasonable to consider a representative sample to be based on the ratio of minority to majority of potentially qualified applicants?
•I’m aware that companies have to provide proof of the validity of their selection procedures, but this article showed how it is also important in the statistical analyses they use to determine if adverse impact had occurred. Are companies required to keep track of the analyses they run on their selection decisions in order to prove that adverse impact did not occur? In addition, should the court require selection personnel to understand these types of statistics in order to prevent adverse impact during hiring?
ReplyDeleteAlthough the authors recommended using the Z-test in testing statistical significance with small samples, they also recognized that low power remains a problem with significance testing small samples. Are confidence intervals accepted by the legal system? When making decisions about the presence of adverse impact, how much weight is placed on mean differences and on statistical significance inferences?
ReplyDeleteKatie's comment made me think about when these statistics are actually run. Are these only done I'd an AI case is brought to the courts? I would guess most HR personnel don't know how to do these analyses, would they hire a statistician to do this kind of work for them?
ReplyDelete*when
ReplyDeleteOne of the specific recommendations here was to use one-tailed tests to improve power since we know the direction of the prediction. How much does this help?
ReplyDeleteThis article is another example of the seriousness of research in this area searching to improve the use of test statistics for the purpose of helpful selection in the area of employment. What a challenge - 40 years of this.
Will computer modeling techniques being worked on now help us with this challenging area?
We learned last semester in methods that even professionals in research make mistakes when it comes to applying the correct statistical test and interpreting results. The issues brought up in this article are quite statistically meaningful, but complicated and need a careful, statistically trained person to apply the right test. Is it practical that organizations would have this type of person? Or would this be done by an outside consultant? ....In other words, as Dr. T once wrote on my paper "what is the utility of this!?!" or more appropriately when is this utilized?
ReplyDeleteOne specific recommendation the authors make is to use one-tailed test to gain power. How much does this help?This article is an example of the seriousness of research searching for improved use of test statistics for the purpose of helpful selection. What a challenge - 40 years of this. Will new forms of computer modeling being developed now help us with this in the future or do we need to put more thought into other approaches as Aguinis et al. (2010) propose?
ReplyDeleteWhen the sample size is so small, perhaps alternative predictors (such as job experience, references) should also be considered? If your sample size is already small, it is probably practical to look for additional selection criteria. What do you think?
ReplyDeleteThe authors address the issue of what is to be considered a representative sample of job applicants when doing significance testing. Certainly, it seems unreasonable to fault an organization for low selection ratios of minority groups (e.g. nurses) if the ratio of minority to majority graduates from local B.S.N. programs shows an underrepresentation of minorities graduating from nursing school. Doesn't it seem reasonable to consider a representative sample to be based on the ratio of minority to majority of potentially qualified applicants?
ReplyDelete