Abstract
Introduction
Medical cannabis is currently a topic of intense interest within the medical and population health communities as a large number of countries and jurisdictions move forward on relaxing restrictions on either medical or recreational cannabis. A rapidly growing literature, starting with the medical effects of marijuana, is consequently branching out to consider a broader range of impacts and effects of cannabis liberalization.
Labor force status is not generally considered in strictly medical studies, but is of immediate concern to policy makers for a variety of reasons, including sustainability of disability programs and medical services and other fiscal reasons such as tax revenues. It also serves as a proxy for general well-being and ability to engage in a variety of activities.
This article investigates whether Canadian medical cannabis patients change their labor supply after being prescribed medical cannabis. To the best of my knowledge, this is the only article that directly observes both medical cannabis usage and labor force status over time in a large sample (
I find that employment increases marginally after prescription of medical cannabis, which may be especially notable as the labor supply of patients with chronic illnesses may, in general, trend downward over time. 1 Patients taking antiepileptic medications on initial prescription of marijuana do particularly well, whereas patients taking nerve modulators do relatively poorly. These findings serve to both motivate further medical research into the interactions of cannabis with a variety of medical conditions and pharmaceutical therapies and contribute to the public health debate over cannabis.
I also examine whether there is any connection between a patient’s diagnosis, the characteristics of their medical cannabis prescription (dose, tetrahydrocannabinol [THC] levels, or cannabidiol [CBD] levels), and their labor market outcomes. None of these factors shows any significant correlation to labor supply, although I cannot clearly identify these effects as prescriptions are subject to a variety of endogeneity concerns.
Related Literature
In the context of cannabis use and labor supply, a literature has focused on identifying the effect of marijuana consumption on employment. In terms of studies that observe cannabis use, these include Popovici and French (2014) and van Ours (2006), who find an insignificant effect, whereas J. Williams and van Ours (2017) find evidence of increased labor market impatience among young men—accepting lower wage jobs more quickly—but no overall reduced labor supply. Adjacent to the labor market, Marie and Zölitz (2017) find a causal negative connection between access to marijuana and academic outcomes, exploiting a natural experiment in the Netherlands. Overall, van Ours and Williams (2015) find in a survey of the literature that “there remains substantial uncertainty as to whether using cannabis has adverse labor market effects” (p. 993). Moreover, as all these authors draw their results from a literature studying recreational use, it is unclear whether these results would generalize at all to a population of medical cannabis users. As far as I can determine, there exist no studies on this subject explicitly dealing with medical cannabis patients, which makes this article a first in this strand of literature.
Conversely, papers dealing with medical cannabis and labor supply do not observe medical cannabis use. 2 For example, Hersch Nicholas and Maclean (2017) exploit changes in medical cannabis legislation across U.S. states to examine the effect of legalizing medical cannabis on the labor supply of older workers, finding a mild positive effect on female labor supply. However, because they cannot observe medical cannabis use in their data, “like all other economics studies examining [medical marijuana legislation] effects of which we are aware” (p. 13), there are limitations to the interpretation of their results. Sabia and Nguyen (2016) find generally no effect of medical marijuana legalization, and indeed a negative effect for young men (see J. Williams & van Ours, 2017, above), though their study does not have the explict casual differences-in-differences identification structure as Hersch Nicholas and Maclean (2017).
In terms of medical cannabis use, however, there does exist literature on medical outcomes potentially linked to labor force participation, though the literature is still developing, given the history of marijuana as medicine. Wang, Collet, Shapiro, and Ware (2008) find a moderate increase in negative outcomes following use of medical cannabinoids. Borgelt, Franson, Nussbaum, and Wang (2013) concur with not only mild negative effects but also positive results for treating pain and spasticity, echoed by Hill (2015) and Whiting et al. (2015).
Finally, given the paucity of studies on medical cannabis, basic descriptive studies of cannabis users have also found a place in the literature. For example, Reinarman, Nunberg, Lanthier, and Heddleston (2011) and Ilgen et al. (2013) simply discuss the characteristics of a population of medical cannabis users.
Data Description
The data for this study are drawn from patient records of a major Canadian physician cooperative specializing in prescription cannabis.
The data collection is built around a clinic model. Patients arrive solely through a referral from either their family doctor or some other medical professional. Clinic physicians may or may not prescribe cannabis based on their medical judgment, though for patients who do not receive cannabis, data are not collected and the share of such patients cannot be estimated. Data are collected via electronically administered surveys, typically while under staff supervision at a clinic. Patients in the data set are evaluated multiple times. Initially, before any prescription of medical cannabis, patients complete a questionnaire designed to assess their suitability for cannabis. If a prescription is written, the patient is obliged to complete follow-up evaluations to allow the prescribing physician to monitor their health and the effectiveness of cannabinoid therapy. This gives the data set a panel element that contains information both before and after prescription of cannabis, allowing direct investigation of changes over time.
All patients are observed twice. One, patients are observed before medical cannabis on their initial assessment for prescription cannabis. First visits occurred between August 2014 and September 2016, where patients reported labor force status and a variety of other information. Patients were observed a second time between October 2016 and April 2017 on a follow-up survey, having already been consuming medical cannabis. There is, thus, significant heterogeneity in how long patients had been using medical cannabis at this check-in point. The change in labor force status is being measured for some patients after as little as 30 days, whereas the longest patient was initially prescribed cannabis 928 days previous to the labor-force check-in data used in this study. The mean time between follow-ups is 242 days, and the distribution of time elapsed between the two observations is given in Figure 1.
Histogram of duration between observations of patient labor force status, in days.
The sample is drawn from the Canadian provinces of Ontario, Alberta, Nova Scotia, and Newfoundland, but the majority of the sample comes from Ontario. The average age of patients is 48.1 years, with 50.7% reporting female, 49.3% male, and 0.04% other.
The labor force composition of the sample across all ages is given in Table 1. Note that the underlying sample exhibits considerably more churn than is at first apparent, because there are people moving in and out of each category. For example, of the 6.5% unemployed, 40% leave the unemployment state, which would bring the rate down to 3.9%, but others move into unemployment over time.
Occupational Status Among Canadian Medical Cannabis Patients Before and After Prescription.
In terms of the prime-age working population presented in Table 2, the picture remains very similar. There is a small increase in employment and voluntary out of the labor force offset by a decrease in unemployment.
Occupational Status Among Canadian Medical Cannabis Patients Before and After Prescription, Age 25 to 55 Only.
Overall, the descriptive statistics do not indicate any significant, across-the-board increase or decrease in labor force participation. Given the decrease in “other”—patients do not specify what “other” might be—it is difficult to interpret exactly what might be changing. It is also unclear what the counterfactual would be, given the medical conditions of these patients. Absent cannabis, it is likely that labor supply would be trending downward over time, for example, as in Wolfe and Hawley (1998), who find a consistent negative time trend in labor supply among arthritis patients. This is particularly true given the mean age of the sample is close to 50 years. Alternatively, in the other direction, changes in the Canadian labor market may also be influencing the distribution across labor force types. 3
The primary medical conditions being prescribed for are varied. The distribution is presented in Table 3. 4
Primary Medical Disorder Motivating Prescription of Cannabis.
Beyond these rough classifications, the data also permit some (but not exhaustive) breakdowns of the given primary conditions in the context of pain, psychiatric, and neurological patients. 5 For example, 13.2% of pain patients are seeking help with fibromyalgia and 15.0% for arthritis, though the majority simply report chronic pain, particularly lower back pain. For psychiatric patients, 40.2% identify as anxiety patients, 20.1% for depression, 16.2% for sleep disorders, and 16.0% for post-traumatic stress disorder. For neurological patients, 22.0% of patients are dealing with spinal cord injuries, 19.8% with seizures, and 19.5% with multiple sclerosis.
Furthermore, these conditions are overwhelmingly chronic: 75.6% of the sample reports that their primary condition has been symptomatic for more than 3 years, with only 2.7% reporting that their condition is less than 6 months old.
One final material variable in the study is prior cannabis usage. A large fraction of the sample—56.4%—reports at least occasional cannabis usage on evaluation for medical cannabis. This includes patients using both recreationally and those self-medicating without a prescription.
Results
More formally, I now estimate a series of regressions designed to examine connections between changes in labor supply from before and after medical cannabis and various patient characteristics. Given the richness of the labor supply variable, there are several ways to estimate whether such differences exist. I consider three primary dependent variables—full-time employment, employment (full time or part time), and labor force participation (employed plus unemployed). Patients reporting a labor force status of “other” during either survey are omitted from all of what follows.
Male pain patients in Ontario who start medical cannabis in the fourth quarter of 2014 are taken as the base group. Included control variables used are defined as follows:
age: age, in years;
sex: 1 if female, 0 if male (the two patients reporting a gender of “other” are omitted);
psych (neuro/cancer/gastro/other): 1 if primary condition is psychiatric (neurological, cancer, gastrointestinal, other), 0 otherwise;
prioruse: 1 if using marijuana before medical prescription of cannabis, 0 otherwise;
chronic: 1 if primary condition symptomatic for at least 3 years, 0 otherwise;
antidep (opio/antispas/epilept/nerve): 1 if using antidepressants (opioids, antispasmodics, antiepileptics, nerve modulators) on initial prescription of medical cannabis, 0 otherwise;
highdose (meddose): takes value 1 if patient’s dose is more than 1.5 g/day (0.5-1.5 g/day), 0 otherwise;
highTHC (medTHC): takes value 1 if THC level of patient’s primary strain is more than 15% THC (5%-15%), 0 otherwise;
highCBD (medCBD): takes value 1 if CBD level of patient’s primary strain is more than 12% CBD (4%-12%), 0 otherwise;
timediff: time elapsed between initial prescription and follow-up check-in;
Alberta/Nova Scotia/Newfoundland (AB/NS/NL): provincial dummies;
unem1/unem2: the corresponding provincial unemployment rate measured when the patient starts medical cannabis and when he or she is observed postprescription; and
Q
Given that I am particularly interested in whether patients improve their employment or reduce the severity of their disability, I estimate regressions of the following form, using the change in the dependent variable:
where
Given the discrete nature of the dependent variables, I also estimate the same equation by ordered logit in several specifications. In general, the results are very robust to this choice. All
Table 4 presents a set of estimates for the change in labor force status across these three variables. In general, there is minimal variance in whether coefficients are estimated via ordered logit or via ordinary least squares (OLS)—the
Estimates for Difference in Labor Force Statuses, All Ages.
Looking at column 1, patients taking antiepileptics are approximately 5% more likely to improve their labor force status relative to average, recalling from Table 2 that on average there is a small increase in employment across the board. A similarly significant coefficient also results from ordered logit, and there is some weak significance when extending the dependent variable to employment but not to labor force participation. I also observe consistently negative and usually significant coefficients across specifications for nerve modulators. Given these medications are somewhat more common among the patient base, significance is achieved with smaller coefficients.
These specifications can also be estimated by stratifying across control variables of interest. In particular, I estimate a set of regressions for the change in full-time employment in Table 5 for each gender (columns 1 and 2 for male and female patients), for pain patients (columns 3 and 4 for primary condition other than pain and primary condition of pain), based on whether a patient was using cannabis ex ante (columns 5 and 6 for patients not using cannabis ex ante and for patients already using cannabis at the time of prescription, respectively), and based on whether the condition is chronic (specifications 7 and 8 for acute and chronic). In each case, corresponding control variables drop out based on collinearity.
Stratified OLS Results Based on Change in Full-Time Employment, All Ages.
In general, again most of the variables are insignificant. Although the coefficients on using antiepileptics are mostly marginally short of standard levels of significance, the magnitude is quite consistent at roughly 5% across the board, suggesting a degree of robustness to this finding. Conversely, however, stratification shows that the coefficient picked up in Table 4 on nerve modulators is gendered, with men on nerve modulators doing significantly worse than average and women doing significantly better than average.
Although Table 5 does present a few other scattered coefficients meeting a 5% significance threshold, none of these coefficients seems robust in any way across specifications, and a 5% significance finding is not particularly compelling in a table of 106 coefficients. Perhaps notably, the fact that prior users of cannabis do not seem to do better than cannabis-naive patients suggests that personally directed cannabis use is not a good substitute for medical cannabis therapy.
I now include prescription characteristics in the regressions and present both OLS and ordered logit results in Table 6. However, the prescription is clearly endogenous to the patient’s condition and severity of his or her condition, which cannot be easily controlled for—patients with more severe health problems are likely to be given stronger prescriptions, biasing the estimation results.
Estimates for Difference in Labor Force Statuses, With Prescription Characteristics, All Ages.
The corresponding base group for estimates in Table 6 including prescription characteristics is, thus, male pain patients in Ontario using less than 0.5 g/day of low CBD and low THC cannabis. Note also that sample sizes are decreasing due to a number of patients not knowing the characteristics of their medical cannabis. 7
In adding prescription characteristics, the relatively better performance of patients taking antiepileptic medications wanes somewhat, without producing any other strong findings, and the results become more difficult to interpret. For example, while there is some evidence that medium THC prescriptions are correlated with labor force success, the direction of causality may likely run the other way—patients are receiving such prescriptions as their conditions are relatively less severe.
In Table 7, I extend the analysis of the article to include just the prime working-age population, which should be compared with the results of Table 4. The magnitude of the coefficients on antiepileptics remains similar, though the finding of relatively poor outcomes among patients using nerve modulators vanishes, suggesting the coefficients in Table 4 were driven mostly by older patients.
Estimates for Difference in Labor Force Statuses, Ages 25 to 55.
Conclusion
I analyze the labor supply of approximately 4,000 Canadian medical cannabis patients, providing descriptive evidence on a important and rapidly growing medical constituency, finding mild increases in employment following adoption of prescription medical cannabis therapy. Patients using antiepileptic medications are relatively more likely to see positive changes in their labor force status, whereas patients using nerve modulator pharmaceuticals do relatively worse compared with the patient population.
Although there is not a direct casual estimate on the magnitude of the effect cannabis has on employment, it is notable that there is no observed decline as might be expected in a population of aging individuals with chronic diseases (Wolfe & Hawley, 1998). The article also makes direct contributions to understanding the population of medical cannabis users and relative treatment outcomes in terms of pharmaceutical use and other variables.
This work motivates further study of cannabis, particularly interactions between different medical conditions, pharmaceutical use, and marijuana. Clinical trials or other frameworks that address endogeneity concerns will provide further insight on key questions of health and policy.
Footnotes
Declaration of Conflicting Interests
Funding
ORCID iD
