Data Shadows of an Underground Economy: Volunteered Geographic Information and the Economic Geographies of Marijuana By Matthew Zook*, Mark Graham** and Monica Stephens***

* Geography, University of Kentucky ** Oxford Internet Institute, Oxford University *** Geography, University of Arizona

This version from 8/30/2011

A draft report from the Floating Sheep Working Paper series (FSWP001)

Underground economies, by their nature, are extremely challenging to analyze and study. They tend to be characterized by a lack of empirical data, unresponsive and secretive research participants, and difficult research environments. While these challenges remain, the dissemination of information technologies and emerging practices and technologies of volunteered geographic information offer productive avenues to study hidden economies such as the trade in illegal drugs. Recreational drug consumption is a widespread practice with significant health and policy issues; but basic data about production, distribution and retailing remain difficult to obtain, are often problematic and rarely comprehensive (NDIC, 2009a, 2009b). Without detailed studies and reliable data – particularly at a finer scale than the national level – it is challenging to develop both clear understandings of the spatial dimensions of these economies and the effect of governmental policies. In response to these issues, this paper analyzes the retail geography of one underground drug economy, marijuana, through the use of a new and unique database of anonymous usersubmitted reports on purchase price. While offering insight on the specific retail geographies of marijuana purchases in the United States this research also highlight the potential and shortcomings of user-generated content in studying other underground, hidden, and hard to measure economies. The paper begins by reviewing existing scholarship within economic geography on underground, informal and non-capitalist economies to highlight the theoretical and empirical issue of interest within the discipline. The review then turns to the relatively unstudied spatialities of illegal drug economies with specific focus on marijuana. Building upon the significant lacunae that this review reveals, the paper draws upon a set of over 16,000 spatially-referenced, retail purchases of marijuana to cartographically and statistically analyze price differences across the United States. The analysis demonstrates both that the marijuana economy conforms to standard economic theory in terms of supply and transportation effects as well as confirming that state level regime of legalization have a negative and significant effect on price. In contrast rates of arrests for possession or sales do not produce statistically significant effects on the price of marijuana, but these demand side effects are more complex to interpret Although these findings are unsurprising, they demonstrate the effectiveness of combining volunteered geographic information with studies of economic practices that would otherwise be impossible to conduct. By measuring a facet of the underground economy never before studied with a national-level dataset, comprehensive at each scale, the analysis of user-generated data undoubtedly allows for unprecedented insights into a highly significant component of everyday economic and social life. And by so doing, shows a productive path for future work in economic geography.

Studies of Underground Economies Underground economies1 are not just niche or tangential components of contemporary economic activities. In most countries, they comprise a highly significant proportion of economic activity and are deeply integrated into more formal sectors and practices (Daniels 2004). While underground and informal economies are difficult to measure and define, size estimates vary from 12 percent (in developed nations) to 39 percent of the Gross Domestic Product (Schneider & Enste, 2000). It is also important to note that informal or underground economic activities comprise more than just illegal activities. Undeclared work often takes the form of mutual aid with non-profit motives (Williams 2004; Gibson-Graham 1996). Informal or underground economies often provide empowering networks, spaces, and resources for disruptions of hegemonic capitalist practices. Alternate economic spaces and informal work allow for gifting, generosity, and a range of other economic practices that remain uncommodified, uncommodifiable, and outside of formal networks of trade (Leyshon et. al. 2003). In other words, despite Marx and Engels’ (1848) famous assertion that capitalism must “nestle everywhere, settle everywhere, establish connexions everywhere,” many examples exist of people sharing, giving, working together, and transacting in ways outside of formal economic frameworks. Thus, studies of unpaid informal economic activity could arguably open up discursive space to (and thereby legitimising) informal work as alternatives or parallels to market capitalism (Gibson-Graham 1996). Moreover, studies of underground economies need also recognize that many of these activities are driven by market forces but for reasons of culture, practice or law are not included in the measured economy. Informal vendors and workers range from roadside food stalls to construction workers paid in cash to home-based beauty salons are market transactions even without the formal processes of taxes or paper receipts. Illuminating the market based underground economies could be highly beneficial by uncovering facets of the economy such as child labour, slavery and human trafficking that are able to persist in part precisely because they are able to remain unmeasurable and uncodified by traditional data collection methods. Conversely, codifying and measuring informal and previously hidden practices could also bring negative effects by allowing these practices to be that much more easily embedded into the vicissitudes of global markets. Because of their importance, many researchers have tried to study the dynamics of informal, underground or illegal economic activities. Williams and Round (2007), for instance, employed an ambitious survey of six hundred households in order to measure informal economies in the Ukraine. However, studies like the one conducted by Williams and Round are inevitably 1

Underground economies have traditionally been challenging to delineate and define. Felge’s (2002) taxonomy of illegal, unreported and informal economies is useful in this sense. However, for the purposes of this paper, we choose the employ the term “underground economies” as a blanket phrase to encompass all economic activities, practices and sectors that typically go unmeasured and unrecorded in traditional economic surveys.

confronted by a range of limitations. First, is the need for ambitious, time-consuming, and expensive data collection methods such as interviews and surveys. This limitation is by no means unique to work on underground economies, but is accentuated by the fact that it is especially challenging to crosscheck and triangulate sources that engage in illegal or illicit practices. Furthermore, the contexts of underground economies frequently mean that only certain kinds of data can be collected. Williams (2005), for instance, in his study of the Ukrainian informal economy, had to focus on perceptions instead of first-hand accounts. While researcher led surveys and interviews are extremely valuable in providing intensive knowledge on the specifics of the study targets, they cannot provide the extensive scope that allows one to see more general patterns (Sayer, 1992). Given the importance of underground economies and the methodologically intensive approach often utilized in their study, we argue that it is useful to explore novel forms of data collection to better understand traditionally unmeasured scales of these economies. The phenomenon of volunteered geographic information (VGI) in which ordinary citizens make reports on local conditions holds great promise as means of aggregating data on a wide range of activities. However, relying upon volunteered data on underground economies raises important ethical considerations. Users may be unaware that they are the subjects of inquiry or their data may be reappropriated for purposes not disclosed to the user (Elwood, 2008). While an individual may contribute small amounts of seemingly innocuous and trivial information to the internet, when mixed with locational information, can cause personal harm, or effectuate sanctions based upon the information (Graham, 2005). As the Internet shifts away from an oligopticon towards a government controlled panopticon (Thrift & French, 2002), web-based research on underground activities has the potential to empower, exploit or disenfranchise the participants in the activities under study. With these concerns in mind, this paper now turns to one significant facet of the underground economy: the sale and distribution of marijuana. We examine work already carried out on the economic geographies of the substance before exploring the possibilities for user-generated content in the mapping and measuring of marijuana. Mapping and Measuring Marijuana with Volunteered Geographic Information It is estimated that marijuana has one of the largest annual market values of all traded goods and services in the underground economy and may account for almost one percent of the world’s GDP (Pollard 2005). Four percent of the world’s population use the substance annually and over ten percent of North Americans are reported to have smoked marijuana some point in their lives (UNODC 2006, 2008). In addition to relatively high levels of consumption, the United States is also a major producer of marijuana. Estimates from 1998 suggest that marijuana is the fourth

biggest cash crop in the country and is a particularly important source of regional income for parts of northern California and Appalachia (NORML, 1998). Despite the social and economic importance of the marijuana economy, academic research is hampered by the same data issues that substantially hinder research on other informal economies. Moreover much of the academic work on marijuana economies focuses on micro-level processes such as elasticities of demand (Nisbel and Vakil, 1972; Saffer and Chaloupka, 1999) and relies on relatively small survey datasets rather than local or state level studies of economic structure or demand. Precisely because of the underground nature of the marijuana market, studies of this market and its geographies tend towards informative, descriptive reviews (see Rengert, 1996; NDIC, 2009a) that provide important insight but do not lend themselves to methodologically extensive modeling approaches. Even when market data is available it is rarely available at the sub-national level. For example, the Price and Purity of Illicit Drugs report (Fries et al, 2008) prepared for the U.S. Office of National Drug Control Policy provides national level figures on the retail price of marijuana (approximately $16 per gram in 2007 dollars) but little geographical specificity on price differences. This can largely be attributed to sample size issues as the System to Retrieve Information from Drug Evidence (STRIDE) database used contains only 5,634 observations for marijuana sales from 1981 to 2007. Other estimates for prices in the United States are characterised by a similar degree of nonspecificity (e.g. the UNODC estimates that prices in the US vary from $10–15 per gram) (UNODC 2008). The lack of spatialized data about prices means that there are few insights into the contours of marijuana’s economic geographies. To date the limited geographically oriented research has primarily focused on single metropolitan areas or even specific neighborhoods (Vilalta, 2010; Jansen, 1991) Due to the lack of any effective fine-scaled data collection methods, official statistics on prices tend to be extremely vague both monetarily and geographically. Echoing the Fries et al (2008) study, the European Monitoring Centre for Drugs and Drug Addiction (2008), using extensive surveys, provides no better data than the observation that “prices of both herbal cannabis and cannabis resin varied from EUR 2 to EUR 14 per gram, with a majority of European countries reporting prices in the range EUR 4–10 for both products.” The European studies similarly do not provide any meaningful understandings of the economic geography of marijuana at a sub-national scale. Given these shortcomings it is evident that little is known systematically about the economic geographies of marijuana despite its socio-economic importance. This paper argues, however, that these gaps can now be potentially addressed via volunteered geographic information. More specifically, this analysis seeks authoritative answers to the following research questions: (1) what are the sub-national variations in the price of marijuana in the United States? and (2) What factors can explain these variations? More broadly, this research seeks to address whether

volunteered geographic information, created by thousands of decentralized users rather than one central governmental agency, is capable of providing a meaningful measure of economic activity, be it underground or formal. Before outlining the data collection methods and results in more detail, it is necessary to first outline both some of the characteristics and limitations of volunteered geographic information. Volunteered Geographic Information as a Data Source for Research There are two broad traditions in the use of volunteered-geographic information and usergenerated content. The first is the use of such content in targeted projects with distinct humanitarian, health or other policy objectives and the second is leveraging user contributions in academic research in heretofore unprecedented ways. Three projects that are highly visible and transformative examples of the first tradition include Wikipedia, OpenStreetMap, and Ushahidi. The Wikipedia project is arguably the most visible use of user-generated content with 15 million users contributing content and edits to create millions of articles in hundreds of languages: a task that would have been cost prohibitive using traditional publishing structures. OpenStreetMap has had a similarly transformative effect in the creation of geographical data. By enabling users to upload and edit spatial data about anywhere on Earth, the platform has become home to one of the most comprehensive and copyright free maps ever created. Finally, the Ushahidi project developed a platform for crisis situations in which anyone with a mobile phone or Internet connection can share time-critical information (e.g. locations of floods, lootings, refugees, etc) with disaster response teams (Zook et al, 2010). The platform is now regularly deployed in almost every significant crisis situation and has had dozens of installations since its inception in 2007. These three projects illustrate both the enormous scale of work that can be accomplished, and the ways in which user-generated content can enable projects that would have never before been possible. The second arena of leveraging volunteered content and effort is already well established within certain fields. For instance, the U.S. space agency NASA utilized volunteers referred to as “clickworkers” to help classify carters on the Martian surface. While generally a skilled task limited to trained geologists, NASA developed a simplified identification algorithm – including multiple classifications of the same crater by different users– in which online visitors could participate. The results indicated that “the automatically-computed consensus of a large number of clickworkers is virtually indistinguishable from the inputs of a geologist with years of experience in identifying Mars craters.” (Benkler, 2007) More recently, NASA has continued to leverage this volunteer labor via the Galaxy Zoo project which seeks to classify tens of millions of galaxies imaged by the Hubble telescope. It is important to note that user-generated content and participation are inevitably characterized by a number of important lacunae. Despite claims by people like Clay Shirky (2008, 2011),

online participation varies by income, age, language. User generated content is inevitably produced by only a small proportion of the world’s population such as the tiny percentage of users who are responsible for most of the content within Wikipedia (Ortega et al. 2009). A similar pattern of participation has been noted in OpenStreetMap (Haklay et al. 2008). Furthermore, crowdsourced platforms are generally characterised by not only a relative lack of participation, but also strong geographic bias particularly between developed and developing countries. Most contributions tend to happen from (and about) hubs in the global information economy. One of the starkest examples of this is the fact that more user-generated content is created about Tokyo than the entire continent of Africa (Graham and Zook, 2011). Although the demographics of the volunteer labor force is less of concern for crater or galaxy classification, it is of significant interest when studying social and economic processes on earth. Therefore it is important to remain mindful of the constraints to the creation of crowdsourced data even as one draws upon thousands of user reports that allow for analysis on a range of unprecedented topics; including the geographies of retail market pricing of marijuana in the United States. Crowdsourcing in the Underground Economy Obtaining data on underground economy activities has by definition never been a straightforward process. In the case of marijuana and other illegal drugs economies, most of the available data derives from local and federal law enforcement agencies, particularly the U.S. Department of Justice, charged with combating the trade. As a result the data reflects the moments at which the underground economy intersects with law enforcement and is defined in terms of arrests, eradication of plants and facilities, etc. Little data on the daily, mundane operation of the marijuana economy is available in a systematic manner.2 The ability of the Internet to network disparate actors in an easy and anonymous manner offers the means to crowdsource information about the marijuana economy from the perspective of the final consumer. While a number of pro-marijuana web sites provide information for consumers including suggestions of where to buy marijuana3, the website PriceofWeed.com stands out as a user friendly implementation that makes sharing and searching retail pricing information extremely easy. As noted on the homepage, "We want to crowdsource the street value of marijuana from the most accurate source possible: you, the consumer. Help by anonymously submitting data on the latest transaction you've made" (Price of weed, 2011). The website appeared in early September 2010 and prominently features on its home page a data input form inviting users to submit their own retail marijuana purchases including information on city location (pre-populated based on the geo-location of a user’s IP address), the quantity and quality purchases as well as the purchase price in dollars (see Figure 1). Central to this input form is the submit button labeled “Submit Anonymously” further emphasized by text stating “you are 2

For example, the analysis of drug dealing by Steven Levitt (Levitt and Dubner 2006) is based on data from one gang. 3 For example, www.webehigh.com and www.dopestats.com.

anonymous”. The effort is clearly to reassure users that the act of self-reporting an illegal purchase of marijuana will not result in any harm to the informant.4 Figure 1, Location and Price Input Form at Price of Weed

Directly beneath the data input form is a Google Maps mashup showing the average price of marijuana at the state and province level for Canada and the United States (See Figure 2). Below the map, the user is provided with the average price of an ounce of marijuana in the user’s state level location across a range of quality levels. In addition, the website provides metrics on users’ perceptions of law enforcement attitudes towards marijuana (ranging from lightly enforced to heavily enforced) as well as the social acceptance of marijuana at the state level. Finally, and most importantly for this research, the site posts the individual, anonymized crowdsourced reports on specific marijuana purchases including the amount, quality, price and the city level location (See Figure 3). Figure 2, State/Province Level Map Mashup of Average Price

4

While it is technologically possible to track individual IP addresses of a website’s visitors, the dataset provided by the Price of Weed website to the authors contained no such information and there is little reason to suspect that it is being collected. Moreover there are a number of significant barriers to linking an IP address to a specific individual. First, is the issue of collecting IP addresses tied to submitted reports to the Price of Weed website. To do this would require the cooperation of either the site operator or its hosting company; neither of which has much incentive to cooperate without a legal order to do so. Second, is the issue of connecting specific individuals to any one IP address. Dynamic IP address assignment and Wifi hotspots often mean that an IP addresses will be shared by a number of individuals rather than a single one. The third issue is whether the submission of a price report is an illegal or actionable offense in itself and whether law enforcement agencies would seek to track the reports. For these reasons, it seems extremely unlikely that participation in the Price of Web website would result in any negative impact to the average user.

Figure 3, Price of Weed Pricing Information

The goal and structure of the website also provides a response to the important ethical issue of how user volunteered geographic information is used (Goodchild, 2007; Elwood, 2006). While some social media sites have been criticized for stealth data gathering, PriceofWeed.com explicitly informs users of the reuse of their submissions and highlights this by making all previous submissions and maps derived from them readily available. This openness is also reflected in the willingness of the operators of PriceofWeed.com to share the data with this research project when they were contacted via email in March 2011. The operators of the website provided the authors with two separate data files containing (1) 22,365 price reports from September 2010 to March 2011 and (2) 12,873 user generated social ratings of law

enforcement attitudes towards marijuana and the level of social acceptance. Table 1 outlines the variables included within each database. Due to structure of the data collection it is not possible to link entries within the two databases directly, e.g., connecting a specific price report to an assessment of law enforcement attitudes and social acceptance. Despite this small limitation, the two databases provide a unique and spatially referenced source for price and perception within the underground marijuana economy. Table 1, Database Variables Price   Submissions   id

 

Social  Ratings   id

Variable  Description   IDs  are  NOT  consistent  between   databases  and  do  not  allow  joins  

country region city timedate               Law   Enforcement

Country  location   State/Province  location   City  location   Date  and  time  of  submission   Price  in  Dollars   Amount  in  Ounces   Quality  (High,  Medium,  Low)   User  assessment  on  law  enforcement   regarding  marijuana  (1=  lightly   enforced,  5=heavily  enforced)  

 

country region city Timedate   price amount quality    

 

 

 

 

 

 

   

 

 

 

Social   Acceptance

 

 

User  assessment  on  social   acceptance  of  marijuana  (1=   accepting,  5=very  intolerant)  

While the Price of Weed website accepts reports from any location in the world, the vast majority of reports are from the United States (76.3 %) and Canada (10.0%). Given this large sample size of reports, this research limits its cartographical and statistical analysis to the United States. Mapping the Price of Pot As a preliminary step to the analysis of data from Priceofweed.com, it was necessary to clean the data, remove outliers, and determine inconsistencies in the user-generated data that could be maliciously or unintentionally inaccurate. The initial file obtained from Price of Weed included 17,054 original records for the United States. As users contribute the quantity of marijuana purchased in grams and fractions of an ounce, all entries were converted into prices for one ounce so that a consistent measurement of price could be established. It was then important to remove all records where individuals were recorded paying $20 or less per ounce of marijuana

(as this is an impossibly low threshold) or $800 per ounce (an impossibly high threshold)5. The mean value was then calculated at $317 per ounce and all records that were more than two standard deviations above or below the mean were removed. The remaining 16,549 records were then geocoded according to the city and state combinations volunteered by users. This produced a point file of 16,502 price records associated with locations in 2,397 cities throughout the United States, including Puerto Rico, Alaska and Hawaii. To ensure the accuracy of the price information submitted by users, the analysis relied on the redundancy of locally specific price information. If only one user contributes price information for a given area, he or she may inadvertently introduce error into the data set. Research into the quality of VGI has identified that at any given spatial-unit, user-generated content can be characterized as being high quality data after five redundant entries (Haklay, Basiouka, Antoniou, & Ather, 2010). Based on these findings we experimented with slightly different thresholds beginning with the removal of all entries with only one record in a given city. This reduced the database to 14,817 entries, with 11,860 records in cities with 5 or more entries and 9,386 in cities with 10 or more entries for all levels of user-submitted marijuana quality ratings. Table 2, Distribution of Observations by Quality Type Quality Average Type n Price High 9,955 $377.02 Medium 5,353 $245.14 Low 1,194 $138.12 Sixty percent of the cleaned pricing data is classified as high quality marijuana while medium quality accounts for another 32 percent and low quality represents less than 8 percent of the reports. Whether this distribution reflects the current consumption ratios of marijuana is unknown. Therefore to take advantage of sample size and reduce possible variability due to quality differences this analysis is limited to reports on high quality marijuana purchases. As illustrated by the map on the Price of Weed website (see Figure 2) there is substantial geographical variation in price reports within the United States with generally lower prices in the West and higher prices in the Northeast and South. This is borne out by the average state price for high quality marijuana outlined in Table 1 with Oregon at the low end ($256 per ounce) and Delaware at the high end ($450 per ounce). These prices are consistent albeit about twenty percent lower than the average price of $16 per gram in 2007 (which corresponds to $477 per ounce in 2010 dollars) reported by Fries et al (2008) and suggest that the downward trend in prices noted by the Purity Report continues. 5

We are also aware of a bias towards higher prices when lower quantities are purchased as those purchases to not achieve a bulk discount, although we were unable to account for this in the analysis. We did assume the conversion employed by drug retailers of 28 grams per ounce, rather than the standard conversion rate of 28.35 grams per ounce.

Table 3, Distribution of High Quality Observations by State State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware D.C. Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas

n 56 34 170 56 1370 287 124 26 71 575 209 38 45 419 139 90 157

Ave. Price 405.8 328.1 353.9 396.0 323.0 301.3 426.2 450.0 460.7 361.8 412.2 369.2 314.2 419.8 385.0 414.3 388.9

State Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York N. Carolina

n 74 76 57 162 368 342 217 34 178 94 67 65 58 198 56 876 254

Ave. Price 366.9 446.3 360.0 436.3 416.3 368.8 396.3 393.8 413.6 282.2 361.7 301.6 407.6 412.4 351.3 416.9 417.9

State N. Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island S. Carolina S. Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

N 21 272 68 351 400 71 98 22 124 511 75 61 223 359 35 206 16

Ave. Price 407.5 372.2 419.2 255.8 414.3 419.3 399.0 415.5 415.6 421.3 337.9 393.6 411.9 293.0 392.8 408.6 400.6

Limiting price differential to the state level, however, does not take advantage of the nuanced city level geography that is contained within the database. To understand the variation in marijuana prices at the local level, a continuous surface was created by interpolated from neighboring observations with known prices. A statistical interpolation technique known as kriging was used to identify the average variance among price differences at the local level, in this case through a spherical semivariogram model. This was calculated based on the average variance of all point pairs in 3000-meter intervals. To obtain a price for each location, an interpolated value was estimated as a weighted average of prices from its twelve neighboring points. Map 1 (below) includes the original data that was used for a kriging interpolation of all other maps in this paper.6 Map 2 utilizes the price records from all cities with thirty or more records (6,387 records) to create a surface that is the generalized. This is based on the twelve cities with thirty or more values most proximate to areas that do not meet that threshold. Map 3 utilizes the price records from all cities with twenty or more records (7,594 records in the continental US) to 6

The interpolated results presented in the following maps are based on user generated data reports regardless of quality levels. This is due to the author’s desire to take maximum advantage of the data. When the analysis is limited to reports on high quality pricing the resulting cost surfaces are very similar.

create a surface that is the generalized. The unknown values are similarly determined by the twelve most proximate cities with 20 or more values. Map 4 similarly determines values using all 14,718 values in the continental United States for an interpolation. While increasing the number of points utilized for the interpolation may decrease the accuracy of the data, it also creates more variation in pricing at a finer resolution. Decreasing the number of points increases the accuracy of the data while holding the general pricing trend across the United States consistent. Due to their non-contiguity with the lower forty-eight states – as well as issues with the number of points in Hawaii and the point to area ratio in Alaska – these two states are not included in this cartographic analysis. Map 1, The cities for which we have data, Classifications = 30+, 20-30, 10-20, 2-8, 1

Map 2, Interpolated map for cites where n > 30

Map 3, Interpolated map for cites where n > 20

Map 4, Interpolated map for cites where n > 10

Map 5, Interpolated map for cites where n > 2

Analyzing the Price of Pot While the cartographic analysis provides a useful and geographically nuanced visualization of price differences it says little about the factors behind these variances. Multivariate regression analysis allows the statistical testing of the direction and strength of the relationship between the retail price of marijuana and other factors. However, because the price point data are based on city names selected by users rather than clearly defined geographic units such counties or latitude/longitude coordinates, accurately selecting corresponding secondary data is problematic. This is compounded by the issue that much of the other data related to the marijuana economy, e.g., arrest rates or eradication efforts, are only disaggregated to the state level. These limitations, combined with the variety of state level regulatory regimes (as of March 2010, 16 states had some form of legalized medical marijuana statutes), have led this research to select the state as the most appropriate scale at which to conduct an analysis. Selection of Independent Variables Explanatory variables for differences in price were selected based on four standard theoretical explanations including supply, demand, transportation and cultural milieu. However, given the underground nature of the marijuana economy, there are few if any direct measures of any of these factors available. Nevertheless there are meaningful proxies that allow this analysis to explore the causal relations behind differences in retail price for marijuana in the United States. Table 1 provides a brief description of the variables as well as an indication of the source of the data. Table 4, Independent Variable Descriptions and Sources Variable Name Dependent Variable - Price Rank in Plant Eradication Percentage Youth Marijuana Possession Arrests per capita PoW Reports per capita Distance to Humboldt County, CA Marijuana Sales Arrests per capita Social Acceptance Law Enforcement Medical Marijuana State

Description Price per ounce of high quality marijuana based on cleaned crowdsourced data; Aggregated to the state or city level State ranking in number of marijuana plants eradicated by law enforcement (1=largest number eradicated; 50=smallest number eradicated) Percentage of population that is age 15 to 24 Marijuana arrests for possession per 100,000 pop, 2002

Source Price of Weed

Number of Price of Weed price reports per 100,000 pop (2010)

Price of Weed & U.S. Census Calculated by authors

Great circle distance to centroid of Humboldt County, CA to the centroid of each city; State level distance is based on the average distance of cities located within a state that have pricing reports Marijuana arrests for sales per 100,000 pop, 2002 Average user assessment on the social acceptance of marijuana (1= accepting, 5=very intolerant); Aggregated to the state or city level Average user assessment of law enforcement attitude regarding marijuana (1-5; 1= lightly enforced, 5=heavily enforced); Aggregated to the state or city level Indicates a state with medical marijuana programs as of March 2011 (16 states in all)

NDIC (2009a) U.S. Census Gettman (2004)

Gettman (2004) Price of Weed Price of Weed MPP (2011) and news reports

Determining the supply side to the marijuana economy is difficult due to the fact that the “amount of marijuana available for distribution in the United States is unknown” (NDIC, 2009a, 1). Law enforcement reports suggest that both domestic production and imports (primarily from Canada and Mexico) are key sources for the supply of marijuana but firm figures are not available (NDIC, 2009b, Hudson, 2003). One supply side related measure that is available the number of marijuana plants eradicated by law enforcement agency. This measure, however, is problematic as there is “significant variability in or nonexistence of data regarding the number of cannabis plants not eradicated during eradication seasons, cannabis eradication effectiveness, and plant-yield estimates.” (NDIC, 2009b, no page #). Nevertheless it is clear that certain states – most notably California which accounts for more than 70 percent of the eradicated plants – have larger marijuana production systems than others. Therefore to measure the supply side, this analysis uses the state’s rank of in number of eradicated plants to capture differences in production levels without using the overly specific and likely misleading counts of eradicated plants. It is assumed that an increased local supply of marijuana would result in lower retail prices. The demand side is likewise challenging as there are no customer surveys or market studies to draw upon. Instead, this analysis uses a trio of measures that represent different aspects of demand. The first is the percentage of the population that is between 15 and 24 years of age as this demographic has the highest level of marijuana consumption (NDIC, 2009a). The second is the number of arrests for marijuana possession per capita in a state. Because both of these variables are subject to measurement issues, e.g., does a state have more arrests because of demand or more aggressive law enforcement, this analysis also uses the number of Price of Weed reports per capita to gauge demand for marijuana. Because the website is equally accessible from all states there is no theoretical reason to presume that a user’s decision to submit a price report would geographically biased. Anticipating the direction of the effect of demand, however, is more complicated. While standard economic theory predicts higher demand to result in higher prices, larger numbers of marijuana users per capita could lead to lower prices as producers develop better distribution networks and establish price equilibrium at a lower dollar amount. Given the underground nature of this economy and its fundamental dependence upon clandestine distribution systems for sales, the latter effect on prices seems more likely. Transportation costs are a long standing factor in production and pricing within economic geography and underground economies are no different. In fact, given the risk associated with moving illegal products, one would expect that the friction of distance would be even more pronounced within the marijuana economy. Again, the lack of information about marijuana production and distribution makes it challenging to find a meaningful metric for transportation costs. However, figures provided by the National Drug Intelligence Center (NDIC) show that California far outstrips the rest of the country in plant eradication. Moreover, numerous reports

have identified the “emerald triangle” of Mendocino, Trinity and particular Humboldt County in California as key and long standing centers for marijuana production and export to the rest of the country (Regan, 2009; PBS Newshour, 2010). NDIC (2009b) highlight the distribution pattern of the so called “Corridor B” through which marijuana produced in California is moved along Interstates 15, 80, 70 and 40 and which account for “31 percent of all marijuana seizures on interstates from 2008 through October 2009”. Therefore to provide a straightforward but meaningful measure of distribution costs within the marijuana economy, this analysis uses the great circle distance from Humboldt County, CA to each location with a price report. The working assumption is that that increases in distance from Humboldt County, CA would result in higher prices. The final theoretical effect to test is the cultural and legal milieu in which this underground economy operates. Presumably places that are more accepting of marijuana use – both in terms of social norms and law enforcement – would have lower prices as marijuana transactions would entail less risk. The first measure of these milieu effects is based on the number of arrests for marijuana sales per capita with the assumption lower arrest rates are reflected by lower prices. Two other measures are derived from the user assessment of law enforcement attitudes and social acceptance reported to the Price of Weed website. While these measures cannot be directly tied to a specific price report, the aggregated average for each state measures marijuana users’ perception of the social acceptance of marijuana and the aggressiveness of law enforcement towards the marijuana economy. Less tolerance for marijuana either from society at large or law enforcement is assumed to translate into higher prices. The final measure of cultural milieu is a binary variable that indicates whether a state has some form of legalized medical marijuana use which would be reflected in lower prices. Findings of the Models Based on the independent variables outlined above, this analysis ran a series of state level models7 with the average price for an ounce of marijuana as the dependent variable (See Table 1). Overall the independent variables are consistent in terms of sign with changing significance levels as variables were introduced. The supply side variable of rank in plant eradication was positive and significant throughout all models: indicating that states with less marijuana production (corresponding to a higher ranked value) had higher retail prices. The demand side variables exhibited much less consistency with per capita arrest rate for possession showing no statistical significance. The percentage population between 15 and 24 years old was positive and significant in only a single model. The results for these variables provide little support for any demand side effect. However utilizing the metric of number of Price of Weed reports per capita 7

While the level of the state is the most appropriate level of analysis, multivariate analysis at the city level shows similar results. While city level data for age, arrest statistics, plant eradication and per capita Price of Weed reports are not available, multivariate regression using the remaining independent variables show the same effects for distance from Humboldt and medical marijuana as seen in the state level models. A model with these just these two independent variables and limited to 174 cities with more than 10 reports had an adjusted r-squared of 0.55.

is consistently negative in sign and significant except in Model 4 where this variable has a high degree of collinearity with other variables. This suggests that within the economy of marijuana, a larger base of consumers helps to lower prices, perhaps by helping to develop the distribution networks required for this underground activity. The transportation effect on marijuana prices is the strongest explanatory factor in this model and is consistently significant. The farther away a market is from Humboldt County, the higher the retail price. While the NDIC (2009a; 2009b) reports make clear that significant amounts of marijuana are imported from both Mexico and Canada, these model results strongly support the idea that northern California is the key node in marijuana production for the U.S. market and more significant than imports. This finding should be taken with care, however, as there are two factors which may be inflating the Humboldt County effect. First, the regression portion of this analysis is limited to high quality marijuana and Mexican imports have generally been classified as lower grade with smaller amounts of THC (the main pharmacological element within marijuana) (Fries et al, 2008). Thus, marijuana imported from Mexico is less likely to be captured within the dataset. The second issue is that much of the marijuana production within Canada has been centered within British Columbia and is of high quality (NDIC, 2009a, Fries et al, 2008). Thus, while these imports are likely captured by the reports which make up this dataset, the distance from British Columbia to U.S. states is roughly analogous to Humboldt County and may also contribute to the strength of this variable. Irrespective of the specific reasons, the results strongly support a west to east price gradient. Within the milieu effect, states that have legalized medical marijuana have statistically significant lower prices than states without these provisions. It appears evident that these legal changes have served to lower prices within the marijuana economy. The rest of the social and legal milieu variables, however, prove to be insignificant in the models. Arrests for marijuana sales and the crowdsourced measures of social acceptance and law enforcement do not reach any statistical threshold. Moreover the two metrics derived from Price of Weed data are positively and strongly correlated with one another (Pearson correlation = 0.807) as well as negatively correlated with the per capita number of Price of Wee reports (Pearson correlation = -0.674). As a result there is a high level of collinearity between these variables within Model 4. For that reason the last two variables are dropped in the final model which results in acceptable collinearity rates for all independent variables. Table 5, State Level Model Results

*/**/*** = significant at the 95 / 99 / 99.9 percent interval Conclusions The importance of underground activities in almost all facets of contemporary economies cannot be understated, and yet it remains that we know little about them. This paper both offers a detailed analysis of the retail geographies of marijuana, and a demonstration that despite the many limitations of volunteered geographic information in academic research, data collected in a decentralized manner can nonetheless allow for unprecedented insights into aspects of the economy that would otherwise be unmeasurable. As Castells (2001, 6) notes, “Neither utopia nor dystopia, the Internet is the expression of ourselves” and this paper demonstrates that this includes underground economies. This paper has uncovered important conclusions that would have previously not been possible to observe using traditional data collection methods. The sample of thousands of user-generated reports on marijuana prices has allowed us unprecedented insight into one aspect of the underground economy that is much discussed, but remarkably understudied. The data reveal four key significant factors associated with marijuana prices. The most unsurprising of these findings is that the production of marijuana is a significant factor in explaining marijuana prices. Simple laws of supply and demand are undoubtedly at work here. Further supporting the link between the supply and price of marijuana is the finding that prices increase with distance from Humboldt County, California. This finding suggests not only that supply-side factors place a large role in determining the final price of marijuana, but also that a significantly amount of high quality marijuana sold in the US is domestically cultivated as opposed to being imported (the fact that distance from Mexico and distance from Canada does not how up as a significant factor in prices further supports this argument). Alternative datasets would clearly be needed to conclusively

make conclusions about the origins and sources of marijuana, but these findings do provide directions for hypotheses that future research could employ. A related observation is that all of the sixteen states that allow medical marijuana to be consumed have lower prices than those that do not. This finding appears to indicate that by allowing medical marijuana to be legally sold, states remove a certain amount of demand from the underground economy, thus driving down prices. In other words, the findings unsurprisingly support a speculation that many people currently making use of marijuana for medicinal purposes previously resorted to illegal means to obtain their supplies. However, an alternate explanation would be that because most medical marijuana initiatives were passed through referendum, it is likely that there is a higher rate of tolerance, and presumably a higher rate of marijuana usage in those places. The work presented here also demonstrates the need for additional research in order to both triangulate our findings and further explore trends and patterns that we uncovered in our analysis. With existing data, it is unlikely that crowdsourced data will prove useful or effective for additional work on the retail geographies of marijuana beyond an analysis of temporal changes in price. While user-generated content has underpinned a range of projects and studies, it takes enormous initial effort (or simple good fortune) to encourage thousands of geographically dispersed people to freely contribute data. Most projects that have successfully harnessed user-generated content also provide additional benefits such as being able to view aggregate data (priceofweed.com itself provides this functionality) or playing games as a method of generated content (as in the case of Galaxyzoo). Although this paper is a study of just one data platform, there are myriad other sources of user generated content that could be employed for academic research. Harvesting data from general purpose sites like Twitter; or more single-purpose sites dedicated to particular practices (like illegal labor and hacking) provide myriad further opportunities for research on underground economies. As such, it is important that future work ask what other aspects of underground economies could and should be studied by employing data aggregated through user-generated platforms. Could other illegal or illicit objects (e.g. other drugs and firearms), and services (e.g. prostitution) be mapped and measured from online sources? What sort of insights can these data allow? What insights will never be possible? And how should privacy concerns be best addressed, especially within the contexts of illegal or underground practices? An often overlooked fact is that the geographies of user-generated content tend to be highly clustered and uneven. It is thus important to exercise caution when extrapolating from user-generated online data to offline trends and patterns. By measuring a facet of the underground economy never before studied at such a scale, the analysis of user-generated data undoubtedly allows for unprecedented insights into a highly

significant component of everyday economic and social life. More worryingly however, usersubmitted data also allow for the uncovering of patterns and trends that could spark negative policy and legal reactions and repercussions on the very people who freely contributed information. It seems unlikely that mapping and measuring the prices of marijuana could have significant undesirable repercussions on contributors of information; but ethical concerns undoubtedly come to the fore when using crowdsourced data on other aspects of the underground economy beyond the original purpose of their genesis. The increasing information shadows to economic activities raise the spectre of state and corporate control into previously unregulated practices of informal exchange (e.g. bartering and gifting practices). The analysis of usergenerated content also raises the possibility of surveillance into a range of undocumented work practices (e.g. undocumented farm work). Yet from a more optimistic perspective, there have already been examples of user-generated content being employed to benefit people that occupy spaces outside of, or on the fringes of, formal economic practices. What this paper most acutely demonstrates is that irrespective of questions about how volunteered geographic information should be used, the data shadows of almost all economic activities are becoming more pervasive and encompassing and will inevitably be employed in an increasing amount of future research in economic geography. While volunteered geographic information can be effectively used to study far more visible (and legal) economic practices (user-generated content now exists to track and map everything from gas prices to the cost of coffee), it is in underground economies that such work has the potential to most rapidly expand our understandings of previously hidden economic practices and relationships. As we have demonstrated, important facets of underground economies are often challenging or even impossible to study using traditional methods, but can now be explored and examined in order to fill in the gaps of our understandings of contemporary economic geographies.

Bibliography (not finalized) Benkler, 2007. The Wealth of Networks Castells, Manuel (2001) Internet Galaxy (Oxford: Oxford University Press). Castells, Manuel, 1998. The End of Millenium. Blackwell (Daniels 2004) (Urban challenges: the formal and informal economies in mega-cities). Elwood, S.A., 2006. Critical Issues in Participatory GIS: Deconstructions, Reconstructions, and New Research Directions. Transactions in GIS 10 (5), 693-708. Elwood, S. (2008). Volunteered geographic information: future research directions motivated by critical, participatory, and feminist GIS. GeoJournal, 72(3), 173-183. European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) 2008 The State of the Drugs Problem in Europe http://www.emcdda.europa.eu/attachements.cfm/att_64227_EN_EMCDDA_AR08_en.pdf (Felge 2002) - Defining and estimating underground and informal economies: The new institutional economics approach Fries, Arthur , Robert W. Anthony, Andrew Cseko, Jr., Carl C. Gaither, Eric Schulman. 2008. The Price and Purity of Illicit Drugs: 1981 – 2007. Institute for Defense Analysis Report prepared for the U.S. Office of National Drug Control Policy. (IDA Paper P-4369). Available at http://www.whitehousedrugpolicy.gov/publications/price_purity/price_purity07.pdf Gettman, Jon. 2004. NORML 2002 Almanac of Marijuana Arrest Statistics. http://norml.org/pdf_files/state_arrests_2004/NORML_US_Marijuana_Arrests.pdf (Gibson-Graham 1996) The End of Capitalism (as We Knew it) Goodchild, M. 2007. Citizens as Sensors. Graham, M. and M. Zook. 2011. Visualizing the Global Cyberscape: Mapping User Generated Placemarks. Journal of Urban Technology 18(1): 115-132. Graham, S. (2005). SoftwareSorted Geographies. Progress in Human Geography, 29(5), 562-580. Leyshon A, Lee R and Williams C (eds) (2003) Alternative Economic Spaces. London: Sage Publication Haklay, M. M., Basiouka, S., Antoniou, V., & Ather, A. (2010). How Many Volunteers Does it Take to Map an Area Well? The Validity of Linus Law to Volunteered Geographic Information. Cartographic Journal, The, 47(4), 315-322. Hudson, Ray. 2003. Marijuana Availability in the United States and Its Associated Territories, Report Prepared by the Federal Research Division, Library of Congress under an Interagency Agreement with the National Guard Bureau Counterdrug Office (NGB-CD). http://www.loc.gov/rr/frd/pdf-files/MarAvail.pdf

Jansen, ACM. 1991. Cannabis in Amsterdam: A geography of hashish and marihuana. D. Coutinho (Muiderberg) Levitt, S. and S. Dubner. Freakeconomics. 2006. William Morrow. Marijuana Policy Project (MPP). 2011. In the States. Available at http://www.mpp.org/states/ National Drug Intelligence Center (NDIC). 2009a. Domestic Cannabis Cultivation Assessment. U.S. Department of Justice (2009-L0848-001A). Available at http://www.justice.gov/ndic/pubs37/37035/37035p.pdf National Drug Intelligence Center (NDIC). 2009b. National Drug Threat Assessment 2010. U.S. Department of Justice (2010-Q0317-001). Available at http://www.justice.gov/ndic/pubs38/38661/ Nisbet, Charles T. and Firouz Vakil. Some Estimates of Price and Expenditure Elasticities of Demand for Marijuana Among U.C.L.A. Students The Review of Economics and Statistics Vol. 54, No. 4 (Nov., 1972), pp. 473-475 NORML. 1998 Report on US Domestic Marijuana Production http://norml.org/index.cfm?Group_ID=4444 PBS News Hour. 2010. California to Let Voters Decide on Marijuana Legalization, Aired 13 Oct 2010. http://www.pbs.org/newshour/bb/law/july-dec10/marijuana_10-13.html Pollard, Niklas. 2005 http://www.boston.com/news/world/europe/articles/2005/06/30/un_report_puts_worlds_illici t_drug_trade_at_estimated_321b/ Regan, Trish. 2009. Pot growers thrive in Northern California. January 22. CNBC News. Available at http://www.msnbc.msn.com/id/28354324/ Rengert, George F. 1996. The geography of illegal drugs. Boulder, Colo. : Westview Press. Saffer, Henry and Chaloupka, Frank. 1999. THE DEMAND for ILLICIT DRUGS. Economic Inquiry. 37(3). 1465-7295 (Samers 2005 The Myopia of “Diverse Economies”, or a Critique of the “Informal Economy”) Sayer, Andrew (1992): Method in Social Science. Routledge. London. Schneider, F., & Enste, D. H. (2000). Shadow economies: Size, causes, and consequences. Journal of Economic Literature, 38(1), 77-114. doi: 10.1257/jel.38.1.77 Thrift, N., & French, S. (2002). The automatic production of space. Transactions of the Institute of British Geographers, 27(3), 309-335. UNODC 2006 World Drug Report 2006 http://www.unodc.org/pdf/WDR_2006/wdr2006_chap2_biggest_market.pdf

UNODC 2008 World Drug Report 2008 http://www.unodc.org/documents/wdr/WDR_2008/WDR_2008_eng_web.pdf

Vialta, Carlos J. 2010. The spatial dynamics and socioeconomic correlates of drug arrests in Mexico city Applied Geography 30 (2010) 263–270 (Williams 2004) (Geographical Variations in the Nature of Undeclared Work) (Williams 2005) Evaluating the magnitude of the shadow economy: a direct survey approach). (Williams and Round 2007) (Re-thinking the Nature of the Informal Economy: Some Lessons from Ukraine) Zook, M., Graham, M., Shelton, T. and S. Gorman. 2010. Volunteered Geographic Information and Crowdsourcing Disaster relief: A Case Study of the Haitian Earthquake. World Health and Medical Policy 2(2): 7-33.