A few weeks ago, I wrote a blog citing macroeconomic research that claimed one of the big problems with the current economy is the paradox of persistent unemployment in the midst of high job vacancy rates.
A decomposition of the data reveals the aberration can be totally explained by those unemployed for more than 26 weeks: “… all of the increase in vacancies relative to unemployment has taken place among the long-term unemployed.”
Following up on this, several researchers conducted a study in which they sent fictitious resumes for 600 real job postings. One factor varied in this randomized experiment was the length of time individual applicants had been without a job. A clear finding of the research was that the “resumes” of the long-term unemployed received callbacks at significantly lower rates than the short-term unemployed – even when they presented more pertinent job experience.
San Francisco start-up Evolv Inc., which helps companies hire and manage hourly workers, might argue that such discrimination against the longer-term unemployed is bad for business. Among the analytic gold they’ve unearthed from their wealth of HR data is that “People who have been unemployed for a long time are, once they’re hired again, just as capable and stay on their jobs just as long as people who haven’t been out of work.”
Evolv’s leading a movement to augment experts with analytics for the big decisions of HR. Conventional methods of hiring that rely on resumes, interviews and “gut” are often unreliable. The expertise of management may often pan out – but can flame out as well. The analytical approach looks to replace potential expert bias with the cold statistical correlations that derive from experience, demographics, psychographics and personality/psychological testing. And, as data expands over time, analytics can be used to “help companies pick who to advance, who to promote.”
Evolv’s workforce science = data science business model has also debunked other popular HR stereotypes. With their large and growing call center data bank for example, Evolv has concluded that a non-recent criminal record is actually associated with slightly better than average work performance. And in the call center industry, where 60 percent annual turnover is the norm, the data show that past job-hopping is not a predictor of quick future departures. Other interesting call center findings are that job-relevant experienced candidates perform no better than their inexperienced colleagues, while candidates referred by employees are 20 percent less likely to quit than their peers.
As you might expect, Google’s in the vanguard of HR analytics. Senior VP for people operations Laszlo Block’s point of departure is that, historically, there’s virtually no correlation between Google’s interviewer ratings and future employee performance. “We found zero relationship. It’s a complete random mess.”
Block’s established doubt on several other tenets of Google’s recruitment process as well. Always looking for the best and brightest, Google demanded the highest GPA’s and SAT scores of candidates. Block’s found that “G.P.A.’s are worthless as criteria for hiring, and test scores are worthless – no correlation at all except for brand-new college grads, where there’s a slight correlation.”
Block’s also recommended Google abandon its notorious on-the-spot brain teasers and puzzles for behavioral interview probes. Asking a question such as “Give me an example of a time when you solved an analytically difficult problem.” provides “data” on both what she considers difficult as well as how she interacted in a real-world situation.
The findings cited above would paint workforce science in the correlational as opposed to the causational camp of big data analytics. The correlationally-obsessed argue that the sheer magnitude of data today trumps the problem of messiness and allows analysts to confidently work from weak observational designs. Causation-skeptics, in contrast, prefer experiments to help establish cause and effect, minimally demanding more sophisticated designs such as time series panels of treatment and control for their “studies.”
I have to admit I’m torn here. An analytics skeptic, I certainly understand in many cases all that’s available is observational big data with correlations. At the same time, I’d be upset if my career were derailed by a bogus statistical artifact.
A New York Times article captures the concern: “The larger problem here is that all these workplace metrics are being collected when you as a worker are essentially behind a one-way mirror,” says Marc Rotenberg, executive director of the Electronic Privacy Information Center, an advocacy group. “You don’t know what data is being collected and how it is used.”