Analytical Studies: Methods and References
Constructing Historical Cannabis Consumption Volume Estimates for Canada, 1960 to 2015

Skip to text

Text begins

Acknowledgements

We would like to thank Isabelle Amano, Danny Leung and Claudia Sanmartin for their support, suggestions and discussions. We would like to thank Cindy Renaud and Louise Simcox for their tireless effort towards publishing this manuscript. And, we would like to thank Nigel Wodrich for his comments; Naz Kazi, Conrad Barber-Dueck, Tony Peluso, Jim Tebrake and Philip Smith for their input; as well as the participants of seminars where preliminary results were presented.

Abstract

This paper discusses the process for estimating the volume of cannabis consumption in Canada by age group from 1960 to 2015. Cannabis consumption is estimated using a model that first estimates the number of cannabis consumers among 15- to 17-year-olds, 18- to 24-year-olds, 25- to 44-year-olds and 45- to 64-year-olds. This is accomplished by estimating cannabis consumption prevalence based on multiple survey data sources. For each age group, consumers are divided into categories based on annual frequency of consumption: once in the past year, less than once a month, one to three times a month, weekly (excluding daily) and daily. Each category of frequency of consumption is then associated with a quantity of cannabis consumed. 

1 Introduction

On April 13, 2017, the Government of Canada tabled legislation to legalize the recreational use of cannabis in Canada. As a result, it will be necessary to adjust statistical measures of economic activity in the Canadian System of Macroeconomic Accounts (CSMA) to include the volume, price and value of cannabis produced, distributed and consumed by Canadians (Statistics Canada 2017, Tebrake 2017). To do so, Statistics Canada has been releasing a sequence of reports and data sets to help inform about the size and composition of the cannabis market.  The first estimates for the volume of cannabis consumption were published in December 2017 (Macdonald and Rotermann 2017).  These estimates documented the progression of cannabis consumption by age group in Canada from 1960 to 2015.  The Cannabis Stats Hub subsequently incorporated these estimates by basing its trend estimates for cannabis consumption on them, and extending the data up to 2017.  A detailed discussion of available survey data sources and an examination of recent trends in the prevalence of cannabis use in Canada can be found in Rotermann and Macdonald (2018). This paper provides a detailed methodology for how the time series for cannabis consumption between 1960 and 2015 are estimated.

Cannabis consumption presents an unusual challenge for integration into the CSMA because national accounts estimates do not include illegal activity.^{Note 1} And, while the recent development of a medicinal cannabis market is captured in some national accounts statistics, there is no formal designation for the medical cannabis market. Moreover, cannabis has been widely available in Canada despite its prohibition, and many people regularly consume it recreationally. Consequentially, a fully formed market for recreational cannabis exists outside the boundaries of what is measured as economic activity. Additionally, existing data sources for consumption do not permit a disaggregation between the legal medical market and the illegal, recreational market. As a result, including cannabis consumption in the CSMA requires integrating an estimate for the entire cannabis market. This may create a discontinuity through time if the market value of cannabis consumption is simply added to current consumption estimates once recreational cannabis consumption is legalized.

Some policy questions also require information on the shares of activity in the black market and the legal market. For example, information about total market size and composition is important for evaluating the degree to which the legal market is replacing the black market.

To address these challenges, modern and historical data sources are integrated to estimate the volume of cannabis consumption from 1960 to 2015. This period covers the illegal market and the legal market as it has developed. It also allows for the creation of a time series long enough that including it, or portions of it, will maintain the time series consistency of the CSMA.

It is critical to recognize explicitly at the outset that estimating cannabis consumption involves considerable uncertainty. This uncertainty increases the further into the past one goes because sources of information are more limited and data sources are less consistent. This occurs because cannabis consumption was illegal during these periods, so limited information exists about the size and composition of the cannabis market in Canada. Moreover, the information that does exist exhibits a tendency towards underreporting because of the legal penalties and social stigma associated with cannabis consumption (Rotermann and Macdonald 2018). As a result, strong assumptions must be used to construct estimates. These assumptions are difficult to test, and their validity can affect the estimate of the volume of consumption.

Two main types of data sources can be used to estimate the volume of consumption of illegal narcotics in Canada: health, addiction, and drug/tobacco/alcohol use monitoring surveys; and law enforcement data. To estimate the volume of cannabis consumption for this paper, health, addiction, and drug/tobacco/alcohol use monitoring surveys data are used (e.g., OPBO 2016; ESPAD 2015; Abramsky and Drew 2014; Kilmer et al. 2013, 2011). To do so, population estimates are combined with an estimate of the proportion of cannabis consumers by age group, an estimate of the frequency of use and an estimate of consumption by frequency. The age groups used are 15 to 17 years, 18 to 24 years, 25 to 44 years and 45 to 64 years. Combining these estimates and aggregating across groups allows a volume estimate of cannabis consumption to be tabulated.

The results show that the total consumption of cannabis has tended to increase for a long time. In the early to mid-1960s, before the rapid spread of cannabis consumption among 15- to 17-year-olds and 18- to 24-year-olds, it is estimated that 24 to 29 tonnes of cannabis were consumed annually in Canada. By 1972, this had increased to 223 tonnes. The increase slowed over the rest of the 1970s and the early 1980s, but consumption reached 361 tonnes in 1984. The volume of consumption fell as the adjusted and benchmarked estimate for consumption prevalence and the number of adjusted consumers declined through the latter half of the 1980s and the early 1990s, reaching a low of 203 tonnes in 1992. Volumes quickly recovered over the next decade and reached 483 tonnes by 2001. The volume of consumption stabilized around an average of 475 tonnes from 2002 to 2010. A period of rapid increase then occurred from 2010 to 2015, culminating in an estimated 697 tonnes of cannabis being consumed in Canada in 2015.

The granularity of the estimates permits an examination of the age groups responsible for growth in cannabis consumption across periods. The growth in the cannabis market in the late 1960s and early 1970s comes from younger age groups (15- to 17-year-olds and 18- to 24-year-olds). However, after the early 1970s, growth comes from older age cohorts, as the volume of consumption by younger groups does not exhibit an upward tendency.

The remainder of this paper is structured as follows. Section 2 discusses the data sources, and Section 3 outlines the methodology. Section 4 describes how the necessary parameters are estimated, while Section 5 reports the results. Section 6 then examines the sensitivity of the estimates, and Section 7 concludes the paper.

2 Data sources

This study uses data from a range of sources. These include the report by the Office of the Parliamentary Budget Officer (PBO) (OPBO 2016), whose methodology this report expands upon; estimates of cannabis consumption prevalence and frequency from numerous surveys; and Statistic Canada’s population estimates.

Data drawn from the PBO report are noted in the discussion of parameter estimates and are used here because they facilitate comparisons between this work and that of the PBO. They are also used because they are professionally sourced and examined, and because there are no reasonable grounds to challenge these estimates, given the paucity of information regarding some cannabis consumption questions.

Survey estimates are drawn predominantly from national surveys, with the exception of earlier periods when less data are available.

The national surveys include the 1970 surveys for the Commission of Inquiry into the Non-Medical Use of Drugs, commonly referred to as the Le Dain Commission (Le Dain 1972); the Health Promotion Survey (HPS) (Statistics Canada n.d.a, n.d.b); the National Alcohol and Drug Survey (NADS) (Statistics Canada n.d.c); the General Social Survey (GSS) (Statistics Canada n.d.d); Canada’s Alcohol and Other Drugs Survey (CADS) (Statistics Canada n.d.e); the Canadian Community Health Survey (CCHS) (Statistics Canada n.d.f, n.d.g); the Canadian Addiction Survey (CAS) (Adlaf, Begin and Sawka 2005); the Canadian Alcohol and Drug Use Monitoring Survey (CADUMS) (Health Canada n.d.); the Canadian Tobacco Use Monitoring Survey (CTUMS) (Statistics Canada n.d.h); and the Canadian Tobacco, Alcohol and Drugs Survey (CTADS) (Statistics Canada n.d.i, n.d.j).

Early-period estimates are taken from surveys conducted by the Centre for Addiction and Mental Health (CAMH) and by independent researchers.

The CAMH data are used because they report on Ontario, which is the most populous province, and because of their extended time series. Both the Ontario Student Drug Use and Health Survey (OSDUHS) of high school students (Boak et al. 2015) and the CAMH Monitor (Ialomiteanu et al. 2016), which surveys people aged 18 and older, began in 1977. The former is conducted every two years, while the latter was conducted in particular years up to 2006, after which it has been conducted annually.

Published results from independent researchers are taken from Lanphier and Phillips^{Note 2} (1971); Smart and Fejer (1971); Smart, Fejer and White (1972); Rootman (1972); and Currie, Perlman and Walker (1977). Most of these studies focus on high-school-aged people, so this age group has the majority of the data points for this period.

The surveys from which the data are drawn typically target the Canadian population aged 15 and older, but a number of people are excluded. People younger than 15 are assumed to consume sufficiently small quantities of cannabis that their exclusion will not have an impact on the results. Common survey exclusions are people living in institutions, people living on reserves, people serving in the Canadian Forces, people living in the territories and people without a landline telephone (see Rotermann and Macdonald 2018). To produce an estimate of the cannabis-consuming population, the estimate of the Canadian population aged 15 and older is used instead of the population estimate from the individual surveys. This imposes the assumption that the excluded population has the same characteristics as the surveyed population.^{Note 3}

Population estimates by age for the years 1960 to 1970 are taken from CANSIM Table 051-0026 (Statistics Canada n.d.k), while estimates for the years 1971 to 2016 are taken from CANSIM Table 051-0001 (Statistics Canada n.d.l). These population estimates are aggregated into five age categories. The first four correspond to the age categories used to estimate prevalence: 15 to 17 years, 18 to 24 years, 25 to 44 years and 45 to 64 years.

The fifth category is for people aged 65 years and older. People in this category are included in order to match the target population from the national surveys that are employed as data sources. This age group is assumed not to engage in noteworthy levels of cannabis consumption, and are included only for the purpose of examining the rate of cannabis use for the Canadian population aged 15 and older.

3 Methodology

There are three steps for combining the information on cannabis consumption to produce an estimate of the volume of cannabis consumed.

First, the number of cannabis consumers is estimated. This is accomplished by multiplying a benchmarked prevalence estimate by the population estimate. The prevalence estimate is constructed from survey responses where participants indicated whether or not they had consumed cannabis in the previous year. For some early studies, the prevalence question refers to the previous six months.

To estimate prevalence, four age groups are used: 15 to 17 years, 18 to 24 years, 25 to 44 years and 45 to 64 years. Each age group is treated separately, allowing for the examination not only of the total quantity of cannabis consumed, but of the evolution of the age composition of cannabis consumption over long periods of time.

For each age group, the number of consumers in each year is estimated as follows:

where $p_{i,t}$  is the population of each age group and $\%p_{i,t}$  is the benchmarked prevalence of people in each age group who reported consuming cannabis in the previous year.

Second, the frequency distribution of cannabis consumers for each age group and the frequency-specific underreporting factor are used to allocate the number of consumers into the frequency categories and adjust for underreporting. These estimates are assumed to be fixed through time. The number of adjusted consumers in each age group in each year by frequency of consumption is expressed as follows:

where  $Consumer{s}_{i,t}$  is the number of consumers in each age group as estimated in step 1, $F$  is the frequency-specific adjustment factor for underreporting, $\%F$  is the proportion of consumers from each age group reporting different frequencies of consumption and $\widehat{C}onsumers$  is the adjusted number of consumers.

Third, the volume of consumption for each age group in each year by frequency type is calculated as follows:

where $\widehat{C}onsumer{s}_{k,i,t}$  is the adjusted number of consumers from Equation (2), ${\text{Average consuming days}}_k$  is an estimate of the average number of days on which consumption occurs for each frequency category and ${\text{Grams/Day}}_k$ is the quantity of cannabis consumed each day for each frequency group.

4 Parameter estimates

4.1 Prevalence of consumption

To estimate the benchmarked prevalence of cannabis consumption, a meta-analysis and instrumental variable approach is used. In this approach, estimates for growth rates and levels are combined, allowing data sources to be used based on their relative strengths. This effectively decomposes the problem into two distinct elements: estimating as accurately as possible the level of the prevalence of cannabis consumption and estimating as accurately as possible the growth rates or path through time of the prevalence of cannabis consumption.

The survey data sources are divided between those to use as benchmarks and those to use as projectors (Table 1). The choice of benchmark series stems from the likelihood that these data sources have more accurate results (Rotermann and Macdonald 2018) or from an inability to use them as projectors because of their intermittent nature. The projector series candidates are grouped according to collection methodology and collector, reflecting quality or bias differences stemming from these differences (Rotermann and Macdonald 2018).

4.1.1 Benchmark estimates

The 2002 Canadian Community Health Survey – Mental Health and Well-being (CCHS-MHW) and the 2012 Canadian Community Health Survey – Mental Health (CCHS-MH) are assumed to have the least error because of their data collection method, their relatively larger sample sizes and the context within which their questions were posed (Rotermann and Macdonald 2018). The HPS, CADS, NADS and GSS present data that are interspersed enough that using them amounts to an additional benchmarking exercise. They were considered as a group for inclusion as benchmarks for the 1980s and 1990s. However, series created using them as benchmarks generate results significantly at odds with other data sources, ultimately leading to their omission from the time series construction exercise. The prevalence estimates from the CCHS-MHW and the CCHS-MH form the benchmark levels to which movements through time are affixed.

4.1.2 Projector estimates

To construct time series projectors, the survey data are grouped by collection methodology and collection source to limit the extent to which bias from different methodologies can influence trend estimates for cannabis consumption prevalence. Here, four groups are used:

• Group 1: data collected by Statistics Canada (CTUMS and CTADS)
• Group 2: the CAMH estimates (OSDUHS and Monitor)
• Group 3: early estimates (Le Dain Commission and academic studies)
• Group 4: CAS and CADUMS.

For the CAMH estimates, the age groups do not align exactly with the age groups available from later surveys. The same is true of independent studies from earlier periods. This feature of the data is difficult to control for. Here, the CAMH data and the data from earlier periods are used as instrumental variables rather than as direct observations. For the CAMH data, combinations of different age groups, weighted by population, are used. Based on examinations with other data points, they exhibit a strong correlation, supporting their use as instruments. For earlier-period surveys, the results are assumed to be acceptable, as it is not possible to compare overlapping periods.

Analysis of the time series properties of data from these groups shows clearly that the trends in data from Group 1 and Group 2 are consistent, but that data from Group 4 can be at odds with other sources. It is unclear how data from Group 3 compare with other sources, as there is no overlapping data from which comparisons can be made.

Therefore, time series projectors are created as follows. The same approach is used for all age groupings back to 1977. First, data from CTUMS and CTADS are used to interpolate for 2014. Based on the growth rate between the benchmarks from 2002 to 2012, the implied projector data point is projected for 2002, and then the missing value is interpolated for 2013. The CAMH data are used to chain the series back to 1977 (or the earliest year available for older age groupings).

For earlier periods, more judgment is applied to promote time series consistency because fewer data points are available. For 15- to 17-year-olds, the projector continues to backcast by linking data from CAMH to that from Smart, Fejer and White (1972).  Beginning in 1970, the Le Dain Commission data are incorporated. The direct question on cannabis consumption prevalence is used as is for this age group, and the information from the graphical question is used to back-chain the series to the earliest year possible. This produces a projector that, after benchmarking, fits well with observed information. For 18- to 24-year-olds, the growth rate between the study by Smart and Fejer (1971) and the 1977 data point based on the CAMH data is used to create the historical projection. From this projected point, the Le Dain Commission survey data are used to back-chain the prevalence estimate. This produces a benchmarked estimate that is higher than the published estimates but maintains the relative position of the early data points compared with those of later periods. For 15- to 17-year-olds and 18- to 24-year-olds, the time series are assumed to be constant from the earliest data point back to 1960. This creates a low level of early total consumption that helps in applying time series filters. For 25- to 44-year-olds, there are more limited data available. Prevalence estimates for this age group are projected back to 1971 based on the growth rate between CAMH and Smart and Fejer (1971). The level from Lanphier and Phillips (1971) is then used as is for 1970. For 45- to 64-year-olds, the prevalence series are assumed to begin in 1977 and are based on back-chaining to the earliest data point from the CAMH.

4.1.3 Benchmarked estimates

Charts 1 through 4 show the results of the benchmarking exercise, along with the source data estimates and the results of including the extended group of HPS, CADS, NADS and GSS benchmarks. The results present three salient points.

First, there is sufficient information on cannabis consumption in Canada to construct long-run estimates of the benchmarked prevalence of consumption. These estimates can be constructed for different age groups, and, as the benchmarking exercise shows, not all data sources should be used in constructing the time series. It is beneficial to omit certain data to improve the projection and benchmarking results. It is also important to note that the data are not of consistent quality, and data from further in the past should be viewed as less exact than modern estimates. Moreover, the benchmarked estimates do not correspond to the values derived directly from surveys (except for the benchmark CCHS-MH and CCHS-MHW data points), and this discrepancy can be larger for historical periods. 

Second, using the extended set of benchmarks leads to benchmarked series that are too low in historical periods. They are too low compared with existing survey estimates and compared with results from the United States for the two younger age groups (Appendix A).

Third, prevalence estimates from CAS and CADUMS present a noticeably different time path from other data sources. For 15- to 24-year-olds, the difference is less pronounced and manifests itself as a stronger decline in prevalence than is shown in other data sources. For older age groups, the CAS and CADUMS data show a declining prevalence rather than a rising prevalence. Given their differences from other survey data (see Rotermann and Macdonald 2018) and their contradictory information, CAS and CADUMS are viewed as aberrant observations^{Note 4} and are not used to assess changes in prevalence through time.