Dissertation Proposal - What is the relationship between cannabis use and

youth unemployment?

Student ID: 14331511

The use of cannabis has been increasing over the past decades1. This is mainly due to widespread
media glorification and public de-stigmatization, which are reflected in the worldwide trend to relax
cannabis control policies. An example of this trend is Canada legalising the substance in 20182.
With the united nations recently recognising Marijuana’s medical potential, it is clear that the trend
will continue. However, the literature concerning the effects cannabis use has on the labour market
is contrasting. Moreover, it is unclear how this effect changes with varying degrees of legality. By
looking at data across all European countries I will be analysing the relationship between cannabis
use and unemployment. I will then further my research by interpreting my previous findings in the
context of several European countries with distinct cannabis regulations.

Literature review

Starting with the work by Catalano et al. (2011) “The health effects of economic decline” we see
that unemployment is frequently linked to psychiatric problems such as drug abuse. This
association is thought to arise through two pathways: social causation, where unemployment
induces drug use (Boden et al., 2014) and social selection, where the drug is responsible for job loss
(Sareen et al., 2011).

Social causation:
The literature on the causal relationship between unemployment and drug use is vast, however there
appears to be a lack in research consensus on the direction and magnitude of the effect. French,
Roebuck, and Alexandre (2001) looked at data from a 1997 drug abuse survey to analysis how
chronic drug use changed with unemployment rates. They concluded that chronic illicit drug use
was associated with higher unemployment rates while occasional drug use wasn’t. In another study
Alexandre and French (2004) concentrate on poor neighborhoods in Miami. They observed that
chronic illicit drug use was significantly related to unemployment. This result is in line with the
thoughts of Bradford and Lastrapes (2014) where social causation acts through two main effects:
economic stress and substitution effect. The economic stress effect suggests substance abuse arises
with psychological distress. Thus, stressed caused by a poor economic environment. i.e.,

1
Roth C. 2019. “Global Marijuana Use Rose by 60 Percent over the Past Decade.”
2
BBC News. 2018. “Canada Becomes Second Country to Legalise Recreational Cannabis.”
uncertainty about the future, induces drug use. The substitution effect refers to the opportunity cost
of using drugs. With poor economic conditions and lack of job supply there is a smaller opportunity
cost of drug use, therefore consumption increases. These papers however do not look specifically at
cannabis. When focusing on cannabis Melchior et al. (2015) report a null finding in the correlation
between unemployment and consumption in individuals ageing 20-37. This however became
statistically significant with participants that had not completed a higher education.

Ayllón and Ferreira-Batista (2018) focused on cannabis use in Europe in the context of the 2008
great recession by using four waves of the Eurobarometer survey, “young people and drugs”. They
demonstrated that a 1% increase in regional unemployment is associated with a 0.7 percent point
increase in the probability the respondent had tried cannabis. While this suggests that
unemployment causes cannabis use to increase, the result considered participants who had admitted
to trying marijuana at any point. When specifying between use patterns the results were less
conclusive.

The contradicting results could be driven by several elements. Firstly, the use of different survey
techniques, it is possible that some surveys made the respondents feel more comfortable therefore
respondents were more likely to answer honestly. Secondly sample heterogeneity and different
economic situations. Thirdly the econometric models used differed across studies. Lastly different
scales for drug use disorders and use patterns i.e., what was chronic for some might have been
frequent for others.

Social selection:
Evidence for social selection is given to us by Bruno Casal et al. (2020). They used data from the
EDADES3 to conduct fixed effects analysis on the relationship between drug use and employment
status in Spain during the great recession. They concluded that drug use decreases an individual’s
capacity and availability when he or she is trying to enter the labour market (Bruno Casal et al.,
2020). However, the authors make no distinction in the degree of cannabis use and only consider if
the participant had ever consumed cannabis a year prior to the interview. Furthermore, when
considering the effects on the labour market they looked at only participants that had admitted to
using both cocaine and cannabis. This will have had implications on their findings as use of hard
drugs such as cocaine could be correlated with unobserved individual characteristics (e.g., personal

3
Encuesta Domiciliaria sobre Alcohol y Drogas en España
attitude toward risk) that could also affect the labour market outcomes (Popovici et al., 2014). This
endogeneity problem is particularly relevant in my analysis as cannabis normalization means that
properties like personal attitude towards become less relevant with regards to cannabis
consumption.

Popovici et al. (2014) distinguish between various cannabis use patterns such as daily, weekly etc.
They used data from Waves 1 and 2 of the NESARC4 to run fixed effect logit models in order to
determine the probability of being unemployed as a cannabis user. They found that probability of
being employed across all degrees of cannabis users was lower compared to non-users. Moreover,
the odds of being employed for men who used cannabis daily were 0.217 times the odds for those
that did not use (p<.10), (Popovici et al, 2014). However, this study, like the previous, concentrates
on a specific country. Therefore, country specific labour market structure could be a source of bias.

Register and Williams (1992) also distinguish between use patterns in cannabis. When using data
from the 1984 wave of NLSY5 they observed that when cannabis was used at any time on the job in
the last year, it was positively associated with the individual’s employment, across a given time
period, whilst cannabis use in the last 30 days was negatively associated to employment. Gill and
Michaels (1992) found similar results; suggesting that consumers make use of illegal substances to
mitigate stress therefore their productivity increases in the short run.

It is clear that both social causation and social selection play a role in the relationship between
cannabis and the labour market. However, because I want to analyse how the association varies
with legality, I will be conducting my analysis in the context of social selection so to look at the
substance’s direct effect.

4
National Epidemiological Survey of Alcohol and Related Conditions
5
National Longitudinal Survey of Youth
Data and variables

The data I will use comes from a series of Eurobarometer surveys. I will be specifically looking at
the Eurobarometer conducted in 2014 (Flash EB 401)6 which is the most recent one and allows me
to avoid effects of the 2008 great recession. The survey is titled “young people and drugs” and
contains substance-abuse related information on young adults with ages ranging from 15 to 24. This
age group suits my analysis as it is highly susceptible to social media influence. When looking at
cannabis it distinguishes between used in the last month, used in the last year, used more than a year
ago and never used. While information on daily use would have been useful it is less reliable as
respondents who use the substance daily are reluctant to admit their illegal habit. It contains other
characteristics of the respondents which I will specify when talking about my control variables.
Some variables used in the previous literature are not present in this survey however they are not
relevant for this age group. For example, marital status (Ayllón and Ferreira-Batista, 2018). Overall,
the pooled sample contains 13128 observations. 17% of respondents admitted to using cannabis in
the last 12 months, 3% point more compared to 2011.

To investigate how the effect cannabis use has on the labor market varies with different degrees of
cannabis legality I looked at data from the EMCDDA7. Due to the ambiguousness surrounding
cannabis laws the EMCDDA divides countries into two categories: the ones where incarceration for
minor possession is possible and the ones where it is not. See figure 18.

Figure 1

6
https://data.europa.eu/euodp/en/data/
7
European monitoring centre for drugs addiction
8
Cannabis Policy: Status and Recent Developments. 2020.
While this does not distinguish whether the substance is legal or not, it serves as a good indicator on
the degree of legality. Countries where incarceration for small possession is not possible have either
decriminalized or depanelized the substance. I will use this data to group all 28 countries into a
dummy variable (incarpos) where inacrpos =1 if incarceration is possible and 0 if otherwise.

Dependent variables
The dependant variable will be Unemployment status in binary form, I will therefore harmonize my
data so that the respondent is either unemployed or employed (self-employed and part time
respondents will be considered as employed). Unemployed = 1 and employed = 0

Independent variables

Explanatory variables: Cannabis use will be the key explanatory variable. This will be divided into
used in the last 30 days, used more than 12 months ago, never used. The other explanatory variable
will be the dummy variable incarpos.

Control variables: sex, age, age squared, level of education, student status, circumstances of
consumption, subjective urbanization, household composition and 28 country dummies to correct
for country-varying labour market structures.

Model

I will be using the model, The demand for health care by Grossmann (1972) as a starting point. The
model provides evidence that employment is positively associated with better health. Therefore,
because substance abuse is related to health status, we would expect a significant correlation
between unemployment and cannabis use. This could be due to direct mechanisms such as, higher
probability of injury and depression or indirectly because of things like lower levels of education. In
my analysis I focus on the direct mechanisms by controlling for things like educational attainment.

The variable of interest is unemployment status, which is a binary variable, therefore the
regressions will yield probabilistic results. I will therefore run the regressions using standard
ordered logit models in order to better interpret the yielded values. The probability of
unemployment will therefore be in log odds form.
The basic model will therefore be:

𝑝𝑟𝑜𝑏(𝑢𝑛𝑒𝑚𝑝)+ = 𝛼 + 𝛽1 𝑢𝑠𝑒+ + 𝛽3 𝑖𝑛𝑐𝑎𝑟𝑝𝑜𝑠+ + 𝛽7 (𝑢𝑠𝑒+ ∗ 𝑖𝑛𝑐𝑎𝑟𝑝𝑜𝑠+ ) + 𝛽9 𝐶𝑣+ + 𝑢+

Where 𝑖 denotes the individual 𝑝𝑟𝑜𝑏(𝑢𝑛𝑒𝑚𝑝) is the probability of being unemployed, 𝛽1 𝑢𝑠𝑒 is the
measure of cannabis use, 𝛽3 𝑖𝑛𝑐𝑎𝑟𝑝𝑜𝑠 to account for legality and the interaction term 𝛽7 (𝑢𝑠𝑒 ∗
𝑖𝑛𝑐𝑎𝑟𝑝𝑜𝑠) to measure how these affect unemployment changes for cannabis users. Finally, 𝛽9 𝐶𝑣 is
a vector of my control variables and 𝑢 captures the unobserved individual characteristics.

I would like to point out that in countries where cannabis is more legally/socially accepted the
direct effects of cannabis use will become more relevant as the cofounding bias of behavioural traits
will become less relevant. This is because in these countries participants who admit to use are more
likely to represent the whole population in terms of behavioural traits. Therefore, I will only
consider occasional cannabis use in countries where incarceration due to small possession is not
possible as occasional use effects are the most likely be the driven by this endogeneity.

Hypothesis

I hypothesis that cannabis use will have a negative effect on the individual’s probability of being
employed (𝛽1 will be positive and significant). Furthermore, I assume that this effect will be more
severe in countries where incarceration due to small possession is possible (𝛽7 > 𝛽1 and
significant).

Dissertation structure

• Introduction
• Literature Review
• Econometric Analysis
• Possible explanation for results
• Conclusion
• Evaluation
• Bibliography.
Closing Remarks

It is hard to isolate social selection from social causation when looking at probability of
unemployment. The biggest threat to the validity of my results would be the effects of uncontrolled
confounding. Popovici et al. (2014) solve this issue by using a fixed effects model thereby
eliminating time invariant unobserved individual heterogeneity. However due to the lack of
longitudinal data this is not possible in my analysis. Eurobarometer have also conducted a survey in
2012 however I have still not been able to access it, should it become available to me I will
implemented in my analysis.
Bibliography

Alexandre PK, French MT. 2004. “Further evidence on the labor market effects of addiction: Chronic drug
use and employment in metropolitan Miami.” Contemporary Economic Policy 22 (3):382–393.

Ayllón, S. and Ferreira-Batista. 2018. “Unemployment, drugs and attitudes among European youth”, Journal
of Health Economics, Vol. 57, pp. 236-248.

Bradford, W. D. and Lastrapes, W. D. 2014. A PRESCRIPTION FOR UNEMPLOYMENT? RECESSIONS

Casal, B., Rivera, B. and Currais, L. 2020. "Economic crisis, unemployment and illegal drug consumption in
Spain", Applied Economic Analysis, Vol. 28 No. 83, pp. 153-170.

Catalano R, Goldman-Mellor S, Saxton K, Margerison-Zilko C, Subbaraman M, LeWinn K, and Anderson E.

French MT, Roebuck MC, Alexandre PK. 2001. Illicit drug use, employment, and labor force
participation. Southern Economic Journal. 68(2):349–368.

Gill AM, Michaels RJ. 1992. Does drug use lower wages? Industrial and Labor Relations Review.
45:419–434.

Grossman M. 1972. On the concept of health capital and the demand for health. Journal of Political
Economy. 80:223–255.

Joseph M. Boden, David M. Fergusson, L. John Horwood. 2014. Associations between exposure to stressful
life events and alcohol use disorder in a longitudinal birth cohort studied to age 30, Drug and Alcohol
Dependence, ISSN 0376-8716.

Melchior M, Chollet A, Elidemir G, Galéra C, Younès N. 2015. Unemployment and Substance Use in Young
Adults: Does Educational Attainment Modify the Association? Eur Addict Res, 21:115-123.

Popovici, Ioana, and Michael T French. 2014. Cannabis use, employment, and income: fixed-effects analysis
of panel data.” The journal of behavioral health services & research vol. 41, 2:185-202.

Register CA, Williams DR. 1992 Labor market effects of marijuana and cocaine use among young
men. Industrial and Labor Relations Review. 45:435–448.
Sareen J, Afifi TO, McMillan KA, Asmundson GJG. 2011. Relationship Between Household Income and
Mental Disorders: Findings from a Population-Based Longitudinal Study. Arch Gen
Psychiatry. 2011;68(4):419–427.