The Price of Cannabis in Canada

The Scale of Local Cultivation and Production

Decreases in cannabis prices have been attributed to the adoption of new cultivation techniques that have increased both the size and speed of crop yields. Specifically, hydroponic cultivation, an intensified form of agriculture that brings nutrients and minerals directly to the plant through a water solution (Roberto, 2005), has allowed for a year-round growing season, more fruitful crop yields, and faster turnover rates than would otherwise be harvested from seasonal grows through outdoor cultivation sites (Vanhove et al., 2012). In the United States (US) and Australia, national drug price data have shown that the widespread adoption of hydroponic cultivation techniques has coincided with drops in cannabis prices (e.g. Caulkins & Padman, 1993; Clements, 2004; Clements & Zhao, 2009; Rhodes et al., 1994).^{Footnote 6} This was shown by Clements and Zhao (2009), who found a substantial drop in cannabis prices, as compared to other illegal products in Australia between 1990 and 1999, attributing the decrease to the adoption of hydroponic cultivation techniques and a reduction in penalties associated with cultivation.

In Canada, increases in the domestic production of cannabis have been associated with the adoption of hydroponic cultivation techniques. This was demonstrated by Bouchard and Dion (2009), who found a relationship between the number of hydroponic shops and the number of cannabis cultivation offences from 1977 to 2006 in Quebec. As the number of hydroponic shops increased, so did the number of individuals arrested for cannabis cultivation. Interviews with local growers and analysis of police-reported statistics led the authors to argue that increases in hydroponic cultivation resulted in a drop in cannabis prices by reducing the barriers to market entry, increasing competition, and impacting the availability of cannabis.

Law Enforcement

Studies have shown that increases in law enforcement activity have only modestly, if at all, influenced cannabis prices. Reuter and Kleiman (1986), examining the impact of a federal government crackdown on cannabis distribution over a five-year period, showed that increases in cannabis seizures and arrests had only a modest effect on cannabis prices. Comparing the number of cannabis seizures and arrests made by the Federal Bureau of Investigation from 1980 to 1984 to price and use data, the authors illustrated that, despite fluctuation in interdictions, cannabis prices remained relatively stable and had no impact on overall use.^{Footnote 7} Similar results were found by Kleiman and Davenport (2012), who concluded that despite increases in the number of arrests for cannabis traffickers from 1980 to 2012, cannabis prices have declined substantially over the same period.

The lack of a relationship between law enforcement interdiction efforts and cannabis prices may be attributed to the unique characteristics of the cannabis market. Increases in law enforcement interdictions are expected to raise prices by causing suppliers to operate inefficiently, referred to as the ‘structural consequences of product illegality' (Reuter, 1983). From this perspective, law enforcement activity may have less of an impact on the cannabis market, where production is largely domestic and competition is abundant (Bouchard, 2007). This perspective is supported by Kleiman and Davenport (2012), who argue that the scale and scope of the cannabis market makes it challenging to effectively target, requiring that law enforcement concentrate their resources on export-import and high-level domestic transactions, which only account for a small component of the final retail price. Although increases in enforcement may not increase prices due to the scale and size of the market, the authors suggest that a substantial reduction or cessation in enforcement may lead to decreased prices.

Distribution Costs

Distribution costs stemming from drug importation and/or exportation also influence drug prices. Prices tend to increase with the distance from drug sources, such as points of importation or domestic and/or international cultivation, a pattern typically referred to as price dispersion (Caulkins, 1995; Caulkins & Baker, 2010). Research from the US has found that retail prices for cannabis increase with distance from northern California, the US region most associated with domestic production (Caulkins & Padman, 1993; Caulkins, 1995; Rhodes et al., 1994). Retail prices for cannabis also increase with distance travelled north from the Mexican border, a source country associated with supplying US cannabis at a price gradient of roughly US$325 to US$475 per pound per 1,000 miles travelled (Caulkins & Bond, 2012). Similar dynamics appear to impact prices in Canadian markets. Research suggests that consumers in British Columbia, Ontario, and Quebec (provinces with significant cannabis production) enjoy lower retail prices than do consumers in ‘import' provinces, and especially those that are further from cannabis sources (Boucher, Lawrence & Maslov, 2013).

Quantity Discounts

Even within local drug markets, prices vary across market levels. Bulk purchases tend to fetch lower per unit prices, as compared to smaller, retail transactions. Higher retail prices are a by-product of price markups that accrue as drugs move along the distribution chain (Caulkins & Padman, 1993; Caulkins & Reuter, 1998; Clements, 2006; DeSimone, 2006). The largest markups occur between wholesalers and retailers, although there is evidence of declining price margins for drugs (Storti & De Grauwe, 2009). Using wholesale and retail price points supplied by member states, the United Nations Office on Drugs and Crime (UNODC) (2010) reported retail price markups of 110 and 129 percent for cannabis resin and herb, respectively. The size of price markups was also found to be subject to geographic variation, with countries in Africa reporting higher markups than those in Europe.

Despite disparities in prices, most purchases tend to be smaller in scale, and conducted at the retail level. For instance, Moeller and Pedersen (2014) found that, among a sample of cannabis transactions video-taped by Copenhagen police between 2003 to 2004, the average purchase size was three grams of cannabis, with 88 percent of cannabis consumers (n = 847) purchasing three grams or less. Caulkins and Pacula (2006), based on the 2001 US National Household Survey on Drug Abuse, found that the median purchase size of loose cannabis resin was slightly more than a quarter ounce. More recent estimates put the average amount of cannabis acquired at nearly one half ounce (Kilmer et al., 2014). This discrepancy in purchase size may be attributed to sampling characteristics, with Moeller and Pederson (2014) sampling transactions that occurred outdoors and during the daytime, situational features that may decrease the likelihood and costs of detection by purchasing smaller quantities, whereas Caulkins and Pacula (2006), and Kilmer et al. (2014) relied on more representative surveys of users and likely captured transactions that also occurred in less risky contexts.

Drug Potency

The price of cannabis also varies according to its potency, whether true (e.g. Δ⁹-tetrahydrocannabinol (THC) content) or perceived. Within the literature, quality is often used synonymously with potency, measured by THC content. Greater domestic production and adoption of hydroponic cultivation has resulted in more potent cannabis and larger variation in the potency of cannabis available due to selective cross-breeding of different strains (Burgdorf, Kilmer, & Pacula, 2011; Potter, Clark, & Brown, 2008). But apart from a unique dataset that matched price data with measures of THC concentration for a series of cannabis purchases (see Ben Lahkdar, 2009; Ben Lahkdar et al., 2016), it is much more common for price data to be matched with self-reported assessments of cannabis potency, or quality. The caveat with this approach is that neither law enforcement officials nor cannabis users are capable of accurately assessing the true quality of cannabis that is seized or purchased. Therefore, what is typically controlled for in the calculation of price estimates is a subjective measure of perceived quality. Despite not representing the ‘true' quality of cannabis, self-reports of perceived quality have been more strongly associated with cannabis prices than actual THC content (Ben Lahkdar, 2009; Ben Lahkdar et al., 2016; Cole et al., 2008; Goudie et al., 2007; Sifaneck et al., 2007) and, thus, have been suggested to be a better predictor for estimating cannabis prices.^{Footnote 8}

Characteristics of the Transaction

Transaction-level characteristics have been associated with retail prices. Some studies have found transactions that occur in private locations to be associated with higher self-reported prices (Caulkins & Pacula, 2006), whereas others have found that they are associated with lower self-reported prices (Pacula et al., 2010). Using data from a 2001 general population survey, Caulkins and Pacula (2006) examined how transaction characteristics, such as purchase location (e.g. private dwelling, inside public building, outside, in other location, or close to home) and from whom they purchased (e.g. friend, family, or stranger), influenced the self-reported price for cannabis. The findings showed that there was no relationship between cannabis prices and from whom the purchase was made, and, surprisingly, that prices were higher for transactions made in private dwellings, where transactions costs were expected to be lower because of decreased risks of detection.

Conducting a similar analysis and controlling for the location of purchase (e.g. buyer's neighborhood or other locale), the nature of the exchange (e.g. direct contact with a dealer or third-party handoff) and the nature of the relationship between the buyer and dealer (e.g. regular, occasional, new source), Pacula et al. (2010) modelled how transaction characteristics influenced prices for cash transactions of cannabis purchases among an arrestee population. Contrary to Caulkins and Pacula (2006), the authors found that transactions in private dwellings were associated with lower prices, although no relationship was found between the self-reported price and whether the transaction occurred in the buyer's neighborhood. As noted by Pacula et al. (2010), the divergent results may be attributed to the two different samples (a household versus an arrestee population) and that Caulkins and Pacula (2006) were unable to control for regional market differences. However, both studies stress caution in interpreting the results, as they were unable to control for the potency of cannabis, thus potentially introducing omitted variable bias.

Cannabis Policies

Studies have found that the adoption of medical marijuana laws has been associated with price changes. Using 1990 to 2011 price data collected from High Times, an online venue for users to self-report prices paid for cannabis, Anderson et al. (2013) evaluated how changes to state-level cannabis laws impacted the price of illegal cannabis in the US. Findings showed that the legalization of medical cannabis led to increases in supply of high potency cannabis and a corresponding reduction in cannabis prices. However, price decreases were only observed in the fourth full year following regulation, with no effect on the price of low potency cannabis, thus suggesting a lag effect. This relationship was attributed to the tendency of medical dispensaries toward producing higher potency strains and that cannabis intended for the medical market was being diverted to the illicit market (Anderson et al., 2013). This research is consistent with work by Pacula et al. (2010), who found that self-reported prices for cannabis among the arrestee population were higher in states with medical cannabis laws. Anderson et al. (2013) argues that this increase in price is due to the introduction of a more potent product to the market, although Pacula et al. (2010) were unable to control for the effect of potency on price in their models. Past research on the determinants of cannabis prices is diverse, but broadly consistent in finding that changes to prices are influenced by both endogenous and exogenous factors. Endogenous factors include changes to internal supply and demand, either due to new innovative cultivation techniques or more ample product, both of which have a residual effect on cannabis prices. Exogenous factors, such as changes in the enforcement of medical cannabis laws, has the corresponding effect of increasing or decreasing the risks associated with producing and trafficking cannabis, along with the availability of cannabis on the market. Assessments of cannabis markets should take into account the full set of factors that could influence prices, above and beyond shifts in enforcement.

Official Sources

Official sources capture recorded drug transactions made by undercover law enforcement agents. In the US, the main official sources for drug price data, administered by the Drug Enforcement Administration (DEA), include: the System to Retrieve Information for Drug Evidence (STRIDE); the Illicit Drug Wholesale/Retail Price Report; and the Illegal Drug Price/Purity Reports (IDPPR). STRIDE is a database of drug seizure data, with submissions from federal, state, and local law enforcement agencies, established to ensure such data are readily available to researchers and policymakers to address issues pertaining to drug price and availability.^{Footnote 9} The Illicit Drug Wholesale/Retail Price Report and the IDPPR obtain data on cannabis prices and THC content from records of undercover buys and subsequent laboratory analysis.^{Footnote 10} Caulkins and Bond (2012) found strong correlation in cannabis prices captured by official data, particularly between the IDPPR and STRIDE datasets. The level of granularity embedded within these datasets has been invaluable for analyzing price variation across drug markets, although when compared to the US, such data are not as readily available in other countries. In Canada, the RCMP has made efforts to collect national drug price data through initiatives such as the Illicit Drug Price List. However, publicly available data has been much broader in scope, resulting in the publication of macro-level summary statistics rather than the micro-level data needed to measure regional price variations.

The foremost problem with official data concerns the general absence of potency information. Official data, such as STRIDE, do not contain information on cannabis potency, whereas other sources, such as IDPPR, contain information on cannabis potency, but only publish information on the annual average THC percentage, and does not distinguish between potency at the retail and wholesale levels of the market. Official data are also comprised of records obtained from cannabis seizures that tend to be highly concentrated in particular regions and across few cities (also Pacula et al., 2001, 2010). In addition, the absence of a systematic data collection instrument further confounds official sources. Because data is not collected with the underlying goal of ensuring a representative sample of drug purchases, official data contains bias stemming from law enforcement knowledge and prioritization of producers, suppliers, and/or dealers from whom purchases are made. Law enforcement must also arrange ‘the buy', which typically consists of large wholesale purchases (Caulkins, 2007a), although most cannabis purchases, by comparison, consist of small quantities between known acquaintances (Caulkins & Pacula, 2006). Such limits have important implications for understanding cannabis prices, as transactions between acquaintances or in personal settings are much more likely to escape the attention of the police.

Self-report Data

Complementing official data are self-report data collected through national surveys, local research initiatives, and/or crowdsourcing websites. Comprehensive national initiatives to collect self-report drug use data have been undertaken primarily in the US, Australia, and New Zealand. Analysis by Davenport and Caulkins (2016) found that prices obtained from seizure data tend to be lower than those obtained from self-reports and, according to midpoint estimates of the price ranges in both official and self-report data, that inconsistencies between the two sources could theoretically be as large as 10 to 20 percent.

In the US, the National Survey on Drug Use and Health (NSDUH) and the Arrestee Drug Abuse Monitoring (ADAM) program both collect data on cannabis transactions and prices. The NSDUH is a nationally representative survey of the household population aged 12 years and older that has been conducted annually since 2002, replacing the National Household Survey on Drug Abuse (NHSDA).^{Footnote 11} A cannabis purchasing module was added in 2001 to collect data on participants' most recent cannabis purchases. Questions concerned details on the location in which the transaction occurred, relationship with the seller, and the quantity and price of the purchase. ADAM, administered by the US Department of Justice, surveys the arrestee population regarding their drug purchases, characteristics of the transaction, including location, and dollar amount paid for cannabis. Unfortunately, this program is no longer funded and the latest report from this program was released in 2013 (ADAM II).^{Footnote 12}

Other national initiatives to survey drug prices include Australia's Illicit Drug Reporting System (IDRS) and New Zealand's National Drug Survey. The IDRS, administered by the Australia's National Drug and Alcohol Research Center, aims to monitor patterns in the price, purity, and availability of drugs, including cannabis, so as to identify trends in Australia's illegal drug markets. The initiative triangulates data collected from: 1) annual surveys of intravenous drug users, which include questions concerning the price of users' most recent cannabis purchase within the previous six months; 2) interviews with professionals who have regular contact with individuals in the cannabis market; and 3) official and self-report data, including drug seizure data and national household surveys of drug use. In contrast, New Zealand's National Drug Survey obtains a representative sample of the population aged 13 to 45 years to be interviewed on their drug purchases. Using computer-assisted telephone interviews, respondents that answer positively to having tried cannabis in the past year are asked about the method of procurement (e.g. obtained for free, purchased, or grew themselves), the frequency of purchases, the quantity purchased, and, for each quantity, the amount they would expect to pay (Wilkins et al., 2005). In Canada, by comparison, self-report surveys, including the Canadian Tobacco, Alcohol, and Drugs Survey (CTADS)^{Footnote 13} and the Canadian Community Health Survey-Mental Health (CCHS-MH), capture prevalence of cannabis users and frequency of cannabis consumption, but do not provide information on price, quantities purchased, or quantities consumed.

More focused efforts to obtain cannabis transaction data include convenience samples through online surveys, such as the Global Drug Survey (GDS), and more local efforts, such as those by Pacula, Jacobson, and Maksabedian (2016). The GDS is an independent research company based out of London that produces reports for different organizations. The GDS consists of an anonymous online survey which is promoted through media sources and is administered annually in the middle of November for approximately six weeks. For 2014 and 2015, the GDS reported receiving over 100,000 responses.^{Footnote 14} Pacula et al. (2016), conducting a more local effort, relied on survey data collected by the research firm GfK, which grouped 2,009 English-speaking adults over the age of 18 across Colorado, Washington, Oregon, and New Mexico. Each respondent was asked about their purchasing behaviour, including prices paid for cannabis.

A second source of self-report price data concerns crowdsourcing websites. This shift to online sources has come at the result of the emergence of websites, such as the Price of Weed and High Times, which allow individuals to anonymously submit price data about their cannabis purchases, including the self-perceived quality, quantity, and location of purchase. This method of data collection affords a number of advantages over traditional surveys, perhaps the most obvious of which is the ability to access otherwise hard to reach populations on a truly massive scale. However, a significant drawback is its reliance on a convenience sample and the inability to distinguish the number of transactions reported per person.

The Price of Weed is a dedicated website for users to submit price data on their most recent cannabis transaction. The website provides a systematic venue for users to report their purchase prices, providing a drop-down list of possible locations, quantities (e.g. ranging from one gram to one ounce), and quality (e.g. low, medium, high). In contrast, High Times invites users to self-report the price, strain, and location of their cannabis purchases by email or via Twitter. The data is then published in monthly Trans High Market Quotation (THMQ) reports that contain a year-to-date US price index, average per ounce prices for the top five reported strains, price observations for many US states, and prices of international submissions.

Price of Weed data has been used either as a sole data source or to augment traditional data sources to estimate cannabis prices (Boucher, Lawrence, & Maslov, 2013; Davis, Geisler, & Nichols, 2016; Malivert & Hall, 2013; Parliamentary Budget Officer, 2016; Werb et al., 2012; Zook, Graham, & Stephens, 2011) and has been found to share some degree of correlation with official and other self-reported datasets (Caulkins & Bond, 2012). One of the most comprehensive studies to estimate illicit cannabis prices in Canada relies on data from Price of Weed. Boucher et al. (2013) produced monthly estimates by city and province using more than 5,700 self-reported observations submitted by users for the period spanning September 2010 to November 2012, finding that the non-weighted median national price was $7.39 per gram, irrespective of quality. Apart from seasonal and regional variation, prices were relatively stable over the measurement period. Users in Newfoundland and Labrador reported some of the highest prices for cannabis ($10.00 per gram), whereas users in Quebec reported the lowest cannabis prices ($5.63 per gram). The relatively higher prices in Newfoundland and Labrador were attributed to the strong local demand, the small scale of local production, and the costs incurred due to transportation to remote local communities.

Self-reported data provide a means to capture users not found in official sources, although the prices captured include much more random variation (Caulkins & Bond, 2012). Besides the usual concerns of reliability, in addition to the previously discussed measurement issues related to quantity and quality of purchase, self-reported drug price data has repeatedly been criticized for capturing the prices paid by users for their most recent purchase, which may not necessarily reflect their typical purchase. However, past research finds no difference in the dollar amount of users' most recent purchase, when compared to the average value of their other recent and past purchases (see Bond et al., 2014).

Online Drug Markets

More recently, research has started to consider online drug markets as a source for drug prices. Online drug markets, henceforth referred to as cryptomarkets (Martin, 2014), have surfaced due in part to the emergence of the ‘dark web'.^{Footnote 15} The dark web refers to content hosted on ‘darknets' or peer-to-peer anonymity networks leveraging the infrastructure of the Internet and which are accessible only through encrypted Internet browsers, such as The Onion Router (Tor) (Owen & Savage, 2015). Cryptomarkets follow a design resembling that of legitimate marketplaces, such as Amazon™ or eBay™, such that vendors advertise entire inventories of products to prospective buyers, with product listings denoting prices by quantity, shipping information, and customer reviews. Cannabis represents a sizeable proportion of all product listings on cryptomarkets. Cannabis originating from Canada was found to represent at least 6 percent of all known product listings advertised on the Silk Road, one of the original online cryptomarkets (Broséus et al., 2016; Christin, 2013; Martin, 2014).

Cryptomarkets facilitate direct links between producers and consumers, bypassing the wholesale level of the drug market (Martin, 2014) and attracting consumers through lower advertised prices, wider selection variety, and better quality (Barratt, Ferris, & Winstock 2014). An estimated 7 to 18 percent of cannabis users have consumed drugs ordered from cryptomarkets such as Silk Road (Barratt et al., 2014). Despite high levels of participation, studies have found that the collective revenues of cryptomarket vendors represent a meager share of the global drug market (Kruithof et al., 2016; Soska & Christin, 2015). Revenue estimates are useful for gauging the size and scope of the online market for drugs, but are not all that useful for policy assessment compared to drug price estimates. However, there are no current price estimates for cannabis or other drugs purchased from cryptomarkets.

The available research suggests that the factors influencing drug prices advertised on cryptomarkets are likely to be similar to those that influence prices in traditional drug markets. Variations in the advertised price for drugs appear to be largely attributed to the costs incurred through domestic and/or international shipping, although risk premiums associated with package seizure may also be factored into the retail price (Christin, 2013). Price variation also stems from the quantity of drug(s) sold, as wholesale purchases garner lower per gram prices (Aldridge & Décary-Hétu, 2016). Although most cannabis price listings found online are for quantities under $100, an estimated 10 percent of listings exceed the thousand-dollar threshold often associated with wholesale transactions (Kruithof et al., 2016). Prices are also affected by what are referred to as ‘holding prices' or temporary increases in prices when product inventory is low or completely out of stock, which may skew price estimates upward (Soska & Christin, 2015).

Compared to more conventional data sources, price listings sourced from cryptomarkets provide current price data for standardized quantities of drugs that are reported directly by drug suppliers. These data also capture international drug purchases, an aspect of drug markets not traditionally accounted for within official data or self-report surveys (for an exception, see Barratt et al., 2014). But, as with official data sources, the drawback of drug price data sourced from cryptomarkets is the biased view toward the supply-side view of drug markets, with very little information regarding drug users themselves (Aldridge & Décary-Hétu, 2016). Furthermore, the manner in which cryptomarkets coordinate purchases between drug users and suppliers is in no way reflective of a typical drug purchase involving known acquaintances in personal settings.

Summary

Each of the data sources described provides complementary insight into the drug market. Official data sources reflect contact between law enforcement, users, and suppliers, whereas self-report surveys reflect prevalence rates of drug use and prices paid by users. Data from online cryptomarkets capture product listings and prices set by suppliers. While it is most common for price estimates to be calculated using a single data source, triangulation of price data allows for a cross-comparison of drug prices derived from different methods of reporting, which ultimately yields more reliable estimates. Comparing drug prices across sources also enables an assessment of the endogenous and exogenous factors that influence prices across different facets of drug markets (e.g. factors that influence price in traditional drug markets as compared to cryptomarkets). Analysis of this sort would also provide knowledge of the factors that have a ubiquitous influence on the economics of drug markets and which are simply characteristic of particular facets of drug markets.

Participation Elasticity

Most studies of cannabis price elasticity calculate participation estimates. This trend may be attributed to a lack of data on the amount of cannabis consumed by those who use it. Studies of participation elasticity have primarily been conducted in the US and Australia, often relying on national surveys that ask users about whether they have used cannabis within a set time frame (e.g. past-month or past-year). Studies in the US have examined participation elasticity across different sub-populations, including youth, college students, and arrestees, whereas studies in Australia have primarily relied on general population studies, subdividing estimates across adult and youth respondents. Here, participation elasticity estimates on US samples are reviewed, before turning to estimates on Australian samples.

Pacula et al. (2001) represents one of the first studies to estimate participation elasticity of cannabis demand among youth. Relying on a nationally representative school-based survey, Monitoring the Future (MTF) from 1982 to 1998, the authors calculated cannabis usage prevalence rates among high school seniors who had used cannabis in the past-month and past-year. Cannabis prices were obtained from the DEA's Illicit Drug Wholesale/Retail Price Report, which provided the retail price of cannabis across states. Elasticity was estimated through a repeated cross-sectional analysis, which controlled for students' perception of harm stemming from regular cannabis use and the potency of cannabis, measured by THC level. Among high school students, participation elasticity ranged from -0.002 to -0.69 for monthly participation and -0.06 to -0.47 for annual participation. Given concerns over the possible endogeneity of the covariates^{Footnote 17}, specifically concern over whether students' perceptions of the harm of cannabis use were correlated with cannabis prices, Pacula et al.'s (2001) preferred participation elasticity estimate for past-month and past-year use was -0.30. Based on these results, the authors concluded that a 10 percent drop in prices would increase the number of youth reporting monthly and annual use by 3 percent.

Jacobson (2005) conducted a similar study using MTF data to estimate cannabis usage prevalence among youth aged 15 to 19 years and data from the NHSDA to estimate cannabis usage prevalence among the general population. In contrast to Pacula et al. (2001), price data was also obtained from user self-reports on High Times and law enforcement data from STRIDE.^{Footnote 18} Despite relying on an alternative data source, results were consistent with Pacula et al. (2001), showing a negative relationship between prices and prevalence of cannabis use, with price increases leading to drops in the overall number of users. The findings also highlighted a feedback loop, in which denser markets—areas with larger youth cohorts—were associated with lower prices, which in turn led to increased use (also see Caulkins, 2007b).

Studies by Williams, Pacula, Chaloupka, and Wechsler (2004, 2006) have also found prevalence of cannabis use among college students to be sensitive to changes in prices. Data on cannabis use among college students was obtained from pooled cross-sectional surveys from three waves (1993; 1997; 1999) of the College Alcohol Study (CAS), a nationally representative study of college students aged 18 to 24 years, administered by the Harvard School of Public Health. Cannabis price data was obtained from the IDPPR. Controlling for students' alcohol use, the maximum fine for possession of small amounts of cannabis, and state fixed effects, Williams et al. (2004) estimated college students' past-month participation elasticity to be -0.24, with a similar result for past-year participation elasticity (-0.20), suggesting a 10 percent increase in price would lead to a drop in past-month users of nearly 2.5 percent and a 2 percent drop among past-year users. In a follow-up study controlling for cocaine use, Williams et al. (2006) found participation elasticity to be slightly more conservative, estimating an annual cannabis participation of -0.19 among college students. However, running additional analyses, and dividing their sample into younger (aged 18 to 20) and older college students (aged 21 to 24), the authors found that older college students were more responsive to changes in prices (-0.26) as compared to younger students (-0.16).

Estimates of participation elasticity among the general US population have generated similar results (DeSimone & Farrelly, 2003; Rhodes, Johnson, Han, McMullen, & Hozik, 2001). DeSimone and Farrelly (2003) estimated annual participation elasticity and frequency of use across pooled cross-sectional data from the NHSDA between 1990 and 1997. Similar to Pacula et al. (2001), data on cannabis prices was obtained from the DEA's IDPPR. Although the authors found that cannabis participation was negatively associated with prices, the relationship was very weak, with results sensitive to the inclusion of additional covariates. Other researchers have claimed that this weak relationship is not surprising given the lack of variation in their price data (capturing a shorter period than earlier work) (Pacula, 2010). However, Rhodes et al. (2001), who also relied on NHSDA data (1988 to 1997), found prices to be significantly associated with cannabis use among youth, but not adults. The divergent results may be partially explained by the different sources of price data, with Rhodes et al. (2001) representing one of the few studies to use STRIDE data to estimate cannabis prices. Rhodes et al. (2001) also represents one of the only studies to examine participation elasticity among heavy users. Examining cannabis use among adult arrestees, the authors used the ADAM^{Footnote 19} survey for the years 1989 to 1998. Results showed that hard-core users, measured as arrestees who used cannabis more than ten times a month, were more sensitive to price changes (-2.79), when compared to moderate users (-2.65). These results suggested that a 1 percent increase in the price of cannabis would reduce the proportion of hard-core cannabis use among adult arrestees by nearly 3 percent.^{Footnote 20}

Whereas multiple sources have been used to generate estimates of use to calculate participation elasticity of cannabis demand in the US (e.g. ADAM, CAS, MTF, NHSDA), estimates of participation elasticity of cannabis demand in Australia rely primarily on the National Drug Strategy Household Survey (NDSHS). The NDSHS is a general population survey that has been administered every two to three years since 1985, aiming to capture a representative sample of Australians aged 14 years and older (aged 12 years and older for 2004 surveys and onwards). The survey is designed to provide data on attitudes and behaviour associated with licit and illegal drug use among the noninstitutionalized civilian population in Australia. Important for price elasticity estimates, questions on cannabis use over the past year have allowed researchers to examine the prevalence of cannabis use among Australians.

One of the first studies to calculate participation elasticity of cannabis demand in Australia aggregated NDSHS survey responses to create a longitudinal time series in the annual prevalence of cannabis use (Clements & Daryal, 2003). Their estimation of cannabis prices relied on the assumption that such prices were constant over time (from 1988 to 1995) and focused on changes in the relative price between cannabis and alcohol. The authors found a participation elasticity of -0.50. This study is unique among Australian studies in that the authors aggregated the data to create longitudinal estimates and did not use observed cannabis prices. In contrast, most other studies in Australia estimate participation elasticity using price series data derived from undercover purchases made by state-level law enforcement and the IDRS.

Cameron and Williams (2001) used pooled cross-sectional NDSHS data from 1988 to 1995 to examine price elasticity, comparing individual-level changes in cannabis use relative to cannabis prices derived from official sources. Despite a different approach, results were similar to Clements and Daryal (2003), with a slightly higher price elasticity of -0.89. However, using the same data source, Williams et al. (2004) controlled for state-level effects and found more conservative estimates of -0.18. By including such controls in their models, the authors were able to reduce the likelihood that differences in drug use across states would be wrongly attributed to price, rather than state-specific factors, such as differences in attitudes toward cannabis or variation in law enforcement.

Other studies of price elasticity in Australia have examined how polysubstance users— individuals who use cannabis in conjunction with other drugs—respond to changes in cannabis prices. Williams and Mahmoudi (2004) found that individuals who used both cannabis and alcohol together were more sensitive to changes in price. Using pooled cross-sectional NDSHS data for 1988 to 1998 and official price data, the authors found that annual prevalence rates of cannabis use were negatively related to price, but that polysubstance users who reported both cannabis and alcohol were most affected by price changes. Similar results were found by Ramful and Zhao (2009), who concluded that cannabis users who also used cocaine and heroin were more likely to be influenced by price changes than just cannabis users alone. Results across both studies suggest that individuals who consume more than one drug are more likely to be influenced by price changes, stemming from a substitution effect of one drug for another.^{Footnote 21}

Consistent with US studies examining the participation elasticity of cannabis demand, Australian studies have demonstrated that individuals in their early twenties are more sensitive to changes in cannabis prices. Using pooled NDSHS data from 1988 to 1998, Williams (2004) demonstrated that younger cannabis users (aged 20 to 24 years) were more sensitive to price changes as compared to older age cohorts (aged 25 years and above). This is consistent with studies that have examined the decision to ‘start using cannabis', which have found that lower prices are associated with an earlier age of first use. Using responses to the question “about what age were you when you first used marijuana/cannabis?” drawn from the NDSHS, van Ours and Williams (2007) found that, among a sample of respondents aged 14 to 22 years, being a Native born Australian, having lower education, as well as price predicted initiation into cannabis use, with a price elasticity of -0.50 to -0.70.

Full Price Elasticity

Full price elasticity estimates aim to examine the degree to which changes in cannabis prices influence the total amount of cannabis consumed. Few general population surveys ask cannabis users about consumption, so researchers have had to leverage other data sources. Here, four studies that attempt to estimate the full price elasticity of cannabis demand are outlined.^{Footnote 22} The first two apply behavioural simulation approaches, examining how current users would respond to hypothetical price changes. The latter two rely on transaction data across samples of current users. All studies find that users are sensitive to changes in price, with increases in prices decreasing the total quantity of cannabis consumed.

Nisbet and Vakil's (1972) research represents one of the first studies to examine the full price elasticity of cannabis demand, calculating both participation and full price elasticity estimates among a sample of college students. The authors calculated estimates from self-report questionnaires that were mailed to UCLA students, who were asked to report their prevalence and frequency of cannabis use over the past month, as well as how a price change would influence their cannabis use. Based on 926 returned questionnaires, estimates of participation elasticity fell within the range of -0.7 to -1.0, while full price elasticity was estimated at a slightly higher range of -1.01 to -1.51, suggesting that demand for cannabis was quite sensitive to price changes.

Collins, Vincent, Yu, Liu, and Epstein (2014) found similar results among a sample of 59 regular cannabis users. In contrast to the sample obtained by Nisbet and Vakil (1972), the authors recruited regular cannabis users, defined as those using cannabis at least three times per week, between the ages of 18 to 25 years in the metropolitan area of Buffalo, New York. Participants were asked to complete a computerized purchasing task where they were to report their hypothetical high grade cannabis use across 16 increasing price points that ranged from US$0 to US$160 per joint.^{Footnote 23} Participants were shown photos of ‘average sized' cannabis joints, which were defined as weighing approximately half a gram. The authors found that demand for high-grade cannabis was inelastic across prices that ranged from US$0 to US$13 a joint, whereas demand was elastic for higher priced cannabis, ranging from US$15 to US$160 per joint, suggesting individuals who used cannabis more frequently are more sensitive to fluctuations in price. This is consistent with previous studies that have suggested heavier users are more likely to have less disposable income to spend on cannabis (Pacula & Lundberg, 2014). Overall elasticity of demand was estimated to be -1.75. Although the estimate is slightly higher than that obtained by Nisbet and Vakil (1972), this may be due to the authors' sample of regular users as compared to a general sample of college students.

More recently, studies have turned to transaction data to calculate the full price elasticity of cannabis demand, obtained by interviewing users about their most recent purchase (Ben Lakhdar et al., 2016) and through online crowdsourcing platforms (Davis et al., 2016). In order to calculate full price elasticity, Ben Lakhdar et al. (2016) interviewed a sample of 268 heavy cannabis users contacted through urban centers and techno-festivals in France. Participants were asked about their most recent cannabis purchase, including the price and the perceived quality of the drug (e.g. measured on a scale of 1 to 10). In addition, interviewers asked respondents if they could purchase a small sample of the drug for toxicological analysis to assess the true potency, as measured by THC content. The authors found that a 10 percent decrease in prices would increase near-daily users' consumption by upwards of 2 percent. The study was one of the first to demonstrate that potency has a positive effect on a user's consumption, with higher potency associated with larger transactions; however, including potency in the models did not impact price elasticity estimates. The authors noted that their reliance on cross-sectional price observations limited them to providing short-term estimates, as they were unable to control for historical price observations.

Davis et al. (2016) found slightly more conservative estimates of full price elasticity using cannabis transactions reported to the Price of Weed crowdsourcing website. The authors relied on 23,611 cannabis transactions reported from September 2010 to August 2011, which provided information on the quantity, quality, and price of the purchase. While this design represents an advantage over most studies examining cannabis price elasticity, providing a national sample of self-reported cannabis prices, the main limit of this approach is that the sample is not random and each transaction cannot be assumed to be independent (e.g. one person may submit multiple transactions to the website). To mitigate this limitation, the authors control for clustering by estimating a weighted Ordered Least Squares regression (OLS) with cluster robust standard errors. Findings showed that the price elasticity for cannabis across the US ranges from -0.67 to -0.79.

Importantly, Davis et al. (2016) also represents one of the few studies to include supply-side instruments in their models.^{Footnote 24} In their models, Davis et al. (2016) integrate electricity rates, distance to Mexico City, and medical cannabis laws, all of which are endogenous factors found to influence cannabis prices: electricity rates account for a large proportion of total cost of domestically produced cannabis; distance to Mexico City accounts for a sizeable cost of internationally produced cannabis, particularly given that prior estimates put cannabis produced in Mexico at 40 to 67 percent of all cannabis consumed in the US (Kilmer et al. (2010)^{Footnote 25}; and medical cannabis laws were included due to their relevance for producers, often allowing for home cultivation and, thus, lowering costs associated with supply. The key assumption is that electricity rates, distance to Mexico City, and medical cannabis laws are not correlated with the error term in the demand equation, but do help explain cannabis prices. When including supply-side instruments, the authors found estimates of price elasticity obtained through OLS regression with IVs to be similar or slightly larger than estimates obtained through standard OLS regression in absolute value, ranging from -0.60 to -0.79 as compared to the more consistent -0.69 estimate across models.^{Footnote 26}

Summary

A review of the literature suggests that participation and full price elasticity estimates vary with respect to the heterogeneity observed across different populations from which samples are drawn. In the US, participation elasticity estimates range from -0.20 to -0.30 for youth and student populations, with findings generally suggesting that demand for cannabis among youth is more sensitive to changes in price compared to adults. Estimates controlling for cannabis consumption further suggest that participation elasticity is stronger among heavy users (-2.79) compared to more moderate users (-2.65) in a sample of US arrestees. Findings from Australian studies also indicate that participation elasticity is more sensitive among youth compared to adults, with estimates ranging from -0.50 to -0.89. Results from samples drawn from the general population by comparison, indicate that participation elasticity ranges from -0.50 to -0.70. However, by controlling for heterogeneity across regional drug markets using state-level fixed effects, participation elasticity estimates (-0.18) become more consistent with the more conservative US estimates. Full price elasticity, in contrast, ranges from -1.01 to -1.51 among US college samples and may be as high as -1.75 for the general population. Research from France also supports findings from the US, with estimates in the range of -1.7 to -2.1, although findings rely on a sample of heavy cannabis users. In  comparison, an analysis using a sample drawn from nationally crowdsourced data found much more conservative price elasticity estimates, ranging from -0.67 to -0.79 across the continental US, results which may be attributed to the larger variation in self-reported prices compared to prices derived from official data sources. These ranges are consistent with findings from studies that suggest price elasticity estimates tend to range by one half, results that hold true across various contexts (e.g. Clements, 2008; Kilmer et al. 2010).^{Footnote 27}

Summary

Overall, the descriptive statistics and trends indicate that cannabis prices vary by region, quality, and quantity, findings that are consistent with cannabis prices specifically and drug markets more generally. Findings show that the typical purchase size for cannabis in Canada is 14.2 grams (one half ounce) and that the majority of price entries (over 90%) reported purchases of medium or high quality, a finding also consistent with crowdsourced price data from the US (Davis et al., 2016). Bivariate analysis indicate that the prices paid in Ontario, Quebec, and British Columbia have a significant and positive relationship with quality, such that higher quality cannabis commands higher prices in these regional markets. When aggregated to the national level, price was statistically and positively related to quality. In addition, prices are typically lower in provinces associated with higher cannabis production, compared to provinces in which the majority of cannabis consumed is not locally produced, but imported. With the exception of some small degree of fluctuation and seasonal variation, prices have remained relatively stable across the five years for which data are available. Nonetheless, findings show that prices are generally lower than those reported by Boucher et al. (2013), which is consistent with a main observation from the drug market literature—cannabis prices are decreasing.

Quantity

The quantity of cannabis purchased for each transaction was used as the outcome variable in the analysis. The average transaction size for the sample was 17.04 grams (SD: 11.03) (Table 7). Once again, this is consistent with US self-reported transaction sizes by Kilmer et al. (2014), finding transactions to be roughly one half ounce over a decade period, from 2000 to 2010, but slightly higher than the average transaction size reported to Price of Weed in the US (13.92 grams) (Davis et al., 2016).

The main relationship of interest is the degree to which cannabis prices influences the quantity purchased for each transaction. Price captures the quantity adjusted price per gram paid by the consumer. The average price of cannabis was $7.32 per gram (SD: $3.76), and was highest for high quality at $7.69 per gram (SD: $3.63), followed by low quality cannabis (average: $7.26; SD: 5.57), and medium quality cannabis (average: 7.14; SD: 3.81). Given that price is not the only factor that plays a role in the quantity of cannabis consumed by users, product and transaction characteristics, as well as demographics are included as controls in the models. Product characteristics include the self-reported quality of the drug, transaction characteristics, the season of purchase, and demographics including education and cannabis offences rates.

The majority of transactions reported to the Price of Weed were classified as medium quality (55%), followed by high quality (41%), with only a fraction of users classifying their purchase as low quality (4%). Given that quality captures users' perception of cannabis potency, the measure may capture other components than just the perceived THC content. While studies have yet to assess the relationship between users' subjective perceptions and the true potency of cannabis (e.g. THC content), they have shown that users' perceived potency plays a role in the price paid (e.g. Ben Lakhdar, 2009; Ben Lakhdar et al., 2016; Cole et al., 2008; Goudie et al., 2007; Sifaneck et al., 2007). However, there are mixed results on whether a drug's potency (whether real or perceived) has an impact on demand (Ben Lakhdar et al., 2016; Davis et al., 2016). Davis et al. (2016) found that users who perceive cannabis to be of high quality are more likely to purchase greater amounts, whereas Ben Lakhdar et al. (2016) found no such relationship.

To capture transaction characteristics, the season the user purchased cannabis was included as a covariate. This measure was a categorical variable that captured Winter, Spring, Summer, and Fall. Most transactions occurred in the Spring (34%), followed by Summer (25%), Fall (23%), and Winter (18%).

Consistent with previous studies, education was included as proxy to control for users' income (e.g. Ben Lakhdar et al., 2016; Davis et al., 2016; DeSimone & Farrelly, 2003). Education measures the percent of the population that has acquired a bachelor's degree or higher in the month and province that the transaction took place. Data on education-level was obtained from the Labour Force Survey, a nationally representative survey of the civilian non-institutionalized population aged 15 years of age and older. The survey asks respondents about their highest level of educational attainment at the time of survey. Previous studies have suggested that individuals with more disposable income may be less sensitive to changes in prices (Pacula & Lundberg, 2014) and that cannabis prices are influenced more by individual characteristics than product characteristics (Ben Lakhdar et al., 2016).

Cannabis offences measure the rate of police-reported cannabis offences per 100,000 in the province where the transaction took place. Cannabis offences include possession, trafficking, and production of cannabis, as detected and reported by police. Data was obtained from the Police-Reported Crime Statistics in Canada publications for the years 2011 (Brennan, 2012), 2012 (Perreault, 2013), 2013 (Boyce, Cotter, & Perreault, 2014), 2014 (Boyce, 2015), and 2015 (Allen, 2016). The reports publish provincial cannabis offence rates for each year relying on data from the Uniform Crime Reporting (UCR) survey. The UCR reports each criminal incident recorded in the information systems of participating Canadian police surveys.^{Footnote 51} For each criminal incident, the UCR only counts the most serious violation and, thus, likely represents an underestimate of all cannabis offences. Previous studies have shown that variation in prosecution for cannabis use, as measured through decriminalization (Cameron & Williams, 2001) and the probability of arrest for possession (DeSimone & Farrelly, 2003), is associated with cannabis demand/use—with some exceptions (Davis et al., 2016).

In addition, to control for endogenous factors that may influence cannabis prices, electricity rates were included as a supply-side instrument. Electricity rates account for a large proportion of the total costs of domestically produced cannabis. Climate, security concerns, as well as greater control over the quality of the product, leads many producers to cultivate cannabis indoors. However, indoor growing, which generally requires high intensity lighting equipment for prolonged periods, creates substantial costs to producers. For instance, previous studies showed that the average cannabis grow-op consisted of approximately 36 lights, with each light using 1,000 watts per hour, and are on for an average 14 hours a day over the growing cycle (Diplock & Plecas, 2011). Electricity costs have been found to range from 42 percent for growers with 600 plants (large scale growers) to 76 percent of the total costs for growers with 23,000 plants (industrial growers) (Vanhove, Surmont, Van Damme, & De Ruyver, 2014). When broken down by pound, Caulkins (2010) found that electricity for lighting accounts for approximately US$75 of the $200 to $400 cost of indoor production for each pound of cannabis.

The electricity rate measures the average one-month residential bill in each province for 1,000 kilowatts of electricity. Data on electricity rates was obtained from Manitoba Hydro, who publishes annual estimates of electricity rates across various cities in Canada.^{Footnote 52} Information on electricity rates is obtained from consulting electricity suppliers within local municipalities. Because rates are only published for select cities across provinces (e.g. in Alberta electricity rates are only available for Calgary and Edmonton), city-level estimates are used to generate province-level averages. OLS estimates including electricity rates exclude cannabis transactions reported from Prince Edward Island, as electricity rates are not provided for any cities within this province (n = 96 observations).

A total of five models are estimated to examine how different sets of covariates influence price elasticity of cannabis demand. Model 1 is our baseline, regressing cannabis prices on the size of the transaction. Model 2 introduces product characteristics into the equation, examining how the quality of the purchase influences the quantity demanded. Model 3 looks at transaction characteristics, including the season the cannabis was purchased as a covariate. Model 4 adds in demographic characteristics, including education rate, as well as the rate of cannabis offences in the province in which the transaction occurred. Lastly, Model 5 includes electricity rates as a supply side instrument to control for an exogenous factor that may influence cannabis prices, but not demand. All models are weighted using inverse probability weighting. This method weights the observations by the number of transactions made within each province, capturing the inverse of the probability that each province is included in the current sample.