Saturday, October 25, 2014

Research update

Okay, so it been a couple months, huh?  Well, what say we do a research update then.

But before I dive in, I discovered something interesting and important.  Longtime readers know that one of my biggest pet peeves is how difficult research articles are to get a hold of.  And by difficult I mean expensive.  Historically, unless you were affiliated with a research institution or were a subscriber, you had to pay exorbitant (IMHO) fees to see research articles.  So imagine my pleasure when I discovered that at least one publisher--Wiley, who publishes several of the research journals in this area--now allows you to read-access for an article for as low as $6.  Now that's only for 48 hours and you can't print it, but hey--that's a heck of a lot better than something like $30-40, which historically has been the case!  So kudos.

Moving on.

Let's start with a bang with an article from the Autumn 2014 issue of Personnel Psych.  A few years back several researchers argued that the assumption that performance is distributed normally was incorrect; and it got a bit of press.  Not so fast, say new researchers, who show that when defined properly, performance is in fact more normally distributed.

For those of you wondering, "why do I care?"  Whether we believe performance is normally distributed or not significantly impacts not only the statistical analyses performed on selection mechanisms but theories and practices surrounding HRM.

Moving to the July issue of the Journal of Applied Psychology:

- If you're going to use a cognitively-loaded selection mechanism (which in many cases has some of the highest predictive validity), be prepared to accept high levels of adverse impact.  Right?  Not to fast, say these researchers, who show that by weighting the subtests, you can increase diversity decisions without sacrifice validity.

- Here's another good one.  As you probably know, the personality trait of conscientiousness has shown value in predicting performance in certain occupations.  Many believe that conscientiousness may in fact have a curvilinear relationship with performance (meaning after a certain point, more conscientiousness may not help)--but this theory has not been consistently supported.  According to these researchers, this may have to do with the assumption that higher scores equal more conscientiousness.  In fact, when using an "ideal point" model, results were incredibly consistent in terms of supporting the curvilinear relationship between conscientiousness and performance.

- Range restriction is a common problem in applied selection research, since you only have performance data on a subset of the test-takers, requiring us to draw inferences.   A few years back, Hunter, Schmidt, and Le proposed a new correction for range restriction that requires less information.  But is it in fact superior?  According to this research, the general answer appears to be: yes.

Let's move to the September issue of JAP:

- Within-person variance of performance is an important concept, both conceptually and practically.  Historically short-term and long-term performance variance have been treated separately, but these researchers show the advantage of integrating the two together.

- Next, a fascinating study of the choice of (and persistence in) STEM fields as a career, the importance of both interest and ability, and how gender plays an important role.  In a nutshell, as I understand it, interest and ability seem to play a more important role in predicting STEM career choices for men than for women, whereas ability is more important in the persistence in STEM careers for women.

Let's take a look at a couple from recent issue of Personnel Review:

- From volume 43(5), these researchers found support for ethics-based hiring decisions resulting in improved work attitudes, include organizational commitment.

- From 43(6), an expanded conceptual model of how hiring supervisors perceive "overqualification", which I would love to see more research on.

Last but not least, for you stats folks, what's new from PARE?

- What happens when you have missing data on multiple variables?

- Equivalence testing: samples matter!

- What sample size is needed when using regression models?  Here's one suggestion on how to figure it out.

The December 2014 issue of IJSA should be out relatively soon, so look for a post on that soon!

No comments: