PUTTING THE GENIE BACK IN THE SPECIMEN CUP
Jacob Sullum has a serious essay on rountine workplace drug testing in November’s Reason. (You won’t want to miss the dueling Scalia quotes on the question of privacy.) An opponent of the idea, Sullum cites reasons (not very convincing ones, I thought) for the belief that use of drug testing has passed its peak. But he gives a good review of the issue from the viewpoint of employers, noting correctly the weak evidentiary base underlying what has become a substantial industry. Sullum points out that even if casual illicit drug use is uncorrelated with workplace performance, a positive test on a pre-employment drug screen might still be a signal of a potential problem employee: someone too dumb figure out a test was coming, or too committed to drug use or too indifferent to the prospect of a job to quit for a few days. (Random, unannounced on-the-job testing, as done in the military, raises a different set of questions: such a test is not only a screen, but also a deterrent.)
There’s actually a simple experiment that would help answer the questions Sullum raises. It could be done by any sizable company or agency that uses drug testing, or any of the contractors that perform it. Here’s the idea:
The result of a drug test as reported back to the employer is simply a “positive” or “negative” for each drug examined. But that’s not what actually happens in the lab. The results as the testing equipment spits them out are quantitative. Some arbitrary line is drawn (so many nanograms of metabolite per deciliter of blood), with any value above the cutoff counting as “positive” and anything else “negative.” That means that there are some testees who fall just below the cutoff, and are consequently hired just as if their tests had been completely clean, and whose test status is unknown to anyone in the company.
If results above the cutoff have predictive power about workplace performance, results just below the cutoff should also have (presumably weaker) predictive power in the same direction. If the cutoff is 50 nanograms/dl, look at the people whose sample tested between 40 and 50 and who were actually hired. Compare them (with appropriate statistical controls for age, education, gender, ethnicity, and whatever else the company knew at the time of hiring) with those who came out squeaky-clean, using whatever the company uses as measures of employee performance: job ratings, promotions, absenteeism, accidents, disciplinary proceedings, quits, and firings. (The same sort of study could be run with respect to on-the-job testing, though it couldn’t address the deterrence issue.)
A positive result — a finding that the just-below-the-cutoff employees performed worse on the job than those showing no evidence of drug use — would be pretty convincing, though you’d still want to look at the magnitude of the effect to get some idea about cost-effectiveness compared to other forms of pre-hiring screening, such as calling references. A negative result wouldn’t be conclusive — maybe most of the action is in the people who test way above the cutoff — but it would require some explaining on the part of testing advocates.
Someone less judicious and fair-minded than your correspondent might be tempted to draw an inference from the failure of the drug-testing industry, or the other advocates of workplace testing, to perform such an obvious validation study. But if I were running a business with a seven-figure drug-testing bill, I’d think about running it on my own company.