By Janet Currie and Molly Schnell
The toll that the opioid epidemic has taken on the United States is undeniable. On average, 115 Americans die every day from a drug overdose involving an opioid,and even more suffer the debilitating effects of addiction. Despite state and federal efforts to curb the crisis, there is no sign that the epidemic is letting up. Whichever way the data is sliced, things look bad and are getting worse.
Inevitably, the effects of this crisis touch multiple aspects of people’s lives: their families, their communities, and, of course, their workplaces. One narrative suggests that addiction leads to job loss and lower labor force participation. In fact, the OECD recently stated that the opioid epidemic is responsible for recent declines in labor force participation in the U.S. At first glance this claim would appear to be supported by the data: a study by Alan Krueger at Princeton University showed that among prime-age white men who are out of the labor force, over 50% report taking prescription opioids daily.
Other data, however, points to a different reality. As the epidemic continues to rage, unemployment is at its lowest level in decades. Furthermore, the numbers suggest that many people taking opioids are actually employed: comprehensive prescription data reveals that nearly 85% of opioids prescribed for working age people are paid for by private health insurance, which is overwhelmingly employer provided. While not everyone who uses opioids gets them directly from a physician—some prescriptions are illegally diverted to other users and an increasing number of addicts turn to heroin or illicit fentanyl—the fact remains that many people who take opioids either begin by using or continue to use legally prescribed medications that are paid for by employer-provided health insurance.
So what’s the actual connection between prescription opioids and the labor market?
To answer this question, we analyzed data on all opioid prescriptions filled at pharmacies across the U.S. from 2006 to 2014. This data includes the gender, age group, residential zip code, and payer (public or private) for the prescription. We aimed to identify the causal effect of opioid prescriptions on employment—that is, getting beyond mere correlations—which is a difficult task for at least two reasons.
First, the areas that have been hardest hit by the opioid epidemic are different than areas that have seen less dramatic rises in opioid abuse for many reasons other than employment opportunities. For example, West Virginia has higher rates of both opioid abuse and unemployment than California. While opioid abuse and unemployment will therefore be correlated when comparing West Virginia to California, this does not mean that opioid use causes unemployment or vice versa. The two states are different for a variety of reasons, such as demographic composition and educational attainment. Any of these factors, or a combination of them, could really be to blame for both high substance abuse and poor labor market conditions.
Since areas are different, we examine how employment within an area changes as prescription rates fluctuate. That is, instead of comparing West Virginia to California at a given point in time, we compare West Virginia to West Virginia and California to California over time. Perhaps surprisingly, this within-location analysis shows that changes in opioid prescriptions per capita are not associated with changes in employment. That is, increases in opioid prescribing in a particular place don’t seem to reduce employment there.
Second, while this kind of analysis controls for any time-invariant differences across locations, another complication remains. Let’s say, for example, that Charleston, West Virginia, unveils a new public transportation system that safely and affordably connects the greater metropolitan area. This public transportation system allows those who were previously isolated to connect with employment opportunities, thereby increasing employment. It also reduces traffic accidents since fewer people opt to drive, thereby reducing opioids prescribed for post-accident pain. In this case we would find that opioid use and employment are correlated within West Virginia over time, although this relationship is still not causal: there’s really a third factor—the opening of the new public transportation system—that is behind the two.
To identify what’s really going on, we need to find something that affects opioid prescribing but has no independent effect on employment. To understand how this might work, imagine a helicopter drop of opioid prescriptions on a town. This drop will increase opioid consumption, but it will not have any effect on employment except through this channel. Measuring how employment changed as a result of this helicopter drop would therefore tell us how increasing opioid consumption causally affects employment.
In our analysis, we treat opioid prescriptions to adults 65 and older as this helicopter drop. Why? We found that doctors who have a high propensity to prescribe opioids to the elderly also on average have a high propensity to prescribe opioids to working age people – and opioid prescriptions to the elderly should have no direct effect on the employment of working age people. Even though some elderly people work, and opioids may have some impact on their employment, it is unlikely that competition from the elderly is a major factor affecting employment of prime age adults. We can therefore use fluctuations in prescriptions to the elderly to isolate changes in opioid consumption that are driven by fluctuations in local prescribing practices rather than by changes in local economic conditions. This methodology of finding shifting—or helicopter drop—variables is referred to as “instrumental variables.”
Our instrumental variables analysis demonstrates that there is no simple causal relationship between opioids and employment. While there is a positive, but small, relationship between changes in opioid prescribing and changes in employment for females in areas with low levels of education, this relationship disappears among women in counties with higher levels of education. Furthermore, regardless of local education, there is no systematic relationship between changes in opioid prescribing and changes in employment rates for men.
Many observers have noted that areas that experienced the largest increases in opioid use over the past decade, like Appalachia, have had persistently low employment. However, it is important to keep in mind that these areas had low employment for decades prior to the opioid epidemic. Our results indicate that the correlation between opioid use and low employment in these areas is largely a coincidence and could be due to other factors, such as the prescribing practice styles of physicians in those areas.
Similarly, some studies have found that a high fraction of people who are out of the labor force take pain medication. However, this this does not prove that taking pain medication causes people to drop out of the labor force. For example, someone with chronic back pain might drop out of the labor force due to this condition and then be prescribed opioids. In this case, it would be the patient’s back pain, not their opioid use, that caused them to leave the labor force. More research is necessary into the reasons why chronic pain in middle age seems to be on the rise, as Angus Deaton and Anne Case have noted.
In short, while the opioid epidemic has caused wide-reaching devastation, aggregate employment appears not to be one of its victims. Furthermore, evidence suggests that poor economic conditions cannot be blamed for the crisis itself. What this means is that we must look at the opioid epidemic for what it is: a self-inflicted perfect storm that arose from a combination of newly available opioids, new attitudes about the importance of pain management, loose prescribing practices, and a lack of professional accountability. The solution to the problem must lie in addressing some of these root causes.