Journal of Quantitative Criminology

, Volume 23, Issue 3, pp 221–241 | Cite as

A Capture–Recapture Model to Estimate the Size of Criminal Populations and the Risks of Detection in a Marijuana Cultivation Industry

Original Paper

Abstract

Originally developed in biology, capture-recapture methodologies have increasingly been integrated into the study of human populations to provide estimates of the size of “hidden populations.” This paper explores the validity of one capture-recapture model—Zelterman’s (1988) truncated Poisson estimator—used to estimate the size of the marijuana cultivation industry in Quebec, Canada. The capture–recapture analysis draws on arrest data to estimate the number of marijuana growers “at risk of being arrested” for a period of five years (1998–2002). Estimates are provided for growers involved in two different techniques: (1) soil-based growing, and (2) hydroponics. In addition, the study develops an original method to estimate the prevalence of cultivation sites “at risk of detection.” A first set of findings shows that the cultivation industry is substantial; the estimated prevalence of growers compares to estimates of marijuana dealers in the province. Capture–recapture estimates are also used to compare the risks of being arrested for different types of offenders. Results indicate that hydroponic growers—those involved in large scale and sophisticated sites—face lower enforcement-related risks than growers involved in smaller enterprises. The significance of these findings is discussed in the context of the widespread development, both in Europe and in North America, of a successful domestic production-driven, rather than importation-driven, marijuana trade.

Keywords

Marijuana cultivation Risks of arrest Risks of detection Size of criminal populations Capture–recapture methodologies 

Introduction

The organizational structure of the marijuana trade has undergone a series of transformations since the end of the 1970s. Whereas marijuana used to be mainly imported from foreign sources to industrialized nations such as Canada or the United States, a large share of marijuana supply is now produced domestically. Available estimates indicate that over 50% of the available marijuana in the United States is of domestic origins, making the country its own main marijuana supplier (Chalsma and Boyum 1994; National Drug Intelligence Center [NDIC] 2004). The proportion of domestically produced marijuana may be even more substantial in Canadian markets, as the country is assumed by police sources to be self-sufficient (Royal Canadian Mounted Police [RCMP] 2004). The RCMP (2004) reported that close to 95% of marijuana seized in Canada in 2003 was of domestic origins. Although the cultivation industry now appears well established in Canada, arrest data indicate that major developments are still very recent. Marijuana cultivation offenses in Canada increased substantially in the late 1980s and especially in the 1990s (Plecas et al. 2002). The number of cultivation offenses increased from 668 in 1980, to 1334 in 1990, and to 9041 recorded offenses in 2000 when it nearly surpassed the number of (10,686) drug selling offenses (Canadian Center for Justice Statistics 2001).

Such seemingly rapid and extensive changes in the organization of an illegal industry are uncommon, and raise important questions about our understanding of how the marijuana trade is structured. A first set of questions concerns the size of the cultivation industry, not only as measured in dollar turnover rates (Wilkins et al. 2002a; Caulkins and Reuter 1998; Rhodes et al. 2000) but in terms of opportunities for new and existing offenders to enter the illegal trade. In an importation-driven industry, a small population of offenders can supply large markets through regular but relatively infrequent importation of large drug shipments from foreign sources. In a production-driven industry, a large number of producers and growers should supply the same large market through the addition of multiple small marijuana cultivation sites. How many active offenders have found a niche in this new industry, and how do these populations compare to the prevalence of marijuana dealers?

A second set of questions concerns the reasons why offenders have gradually chosen cultivation over importation as a preferred source of marijuana supply. A popular assumption among drug market researchers is that the substitution has been stimulated in response to law enforcement interventions (Chin et al. 2000; Weisheit 1992; Reuter et al. 1988). Domestic marijuana production is considered by these sources as an adaptive and innovative strategy triggered by increases in risks of arrest and risks of detection among importers. Two waves of adaptations are usually identified. The first one dates back to the 1970s, when several major police operations were conducted at the U.S.-Mexican border, which may have increased the risks of detection for imported marijuana (Reuter and Kleiman 1986; Weisheit 1992). Because marijuana is bulky and is sold at low prices, adaptive strategies available to cocaine and heroin smugglers (such as decreasing the size of each drug shipment but multiplying its frequency) are less efficient than producing it domestically (Reuter et al. 1988; Weisheit 1992).

Law enforcement pressure may also have contributed to the development of a second wave of adaptations, from outdoor (soil-based) to indoor (soil-based) marijuana cultivation. According to the limited existing research on marijuana cultivation, large-scale eradication programs were used by the DEA to discourage the early marijuana cultivation industry in the 1980s, situated in outdoor, rural settings (Potter et al. 1990; Hafley and Tewksbury 1995; Weisheit 1992). As suggested by Reuter et al. (1988), growers of outdoor crops likely realized how vulnerable they were to law enforcement and to other risks (thieves, animals, plant diseases), and at least some of them moved to less conspicuous indoor cultivation, which could also be conducted in urban settings. The trend from outdoor towards indoor cultivation has also been noticed in Europe (Jansen 2002; Hough et al. 2003), in New Zealand (Wilkins and Casswell 2003), and in Canada (Plecas et al. 2002).

This risk/adaptation hypothesis provides a reasonable account of the development of the marijuana cultivation industry. But in the absence of macro-level research on the size and structure of the industry, or on the levels of risks that law enforcement agencies are able to achieve, it remains difficult to evaluate its empirical value. Displacement effects are assumed to have occurred, but no studies estimated how these populations of growers are distributed among different techniques, or how they migrate from one technique to another. Whereas substance abuse studies have integrated prevalence measures into their research and found innovative ways of analyzing shifts and trends in consumption (e.g. Caulkins et al. 2004; Rhodes 1993; Brecht and Wickens 1993), no research has been conducted on the supply side of drug markets. Besides, such prevalence measures are needed to monitor the effectiveness of law enforcement efforts. With what is known to date, it remains difficult to answer seemingly simple questions: How risky is growing marijuana compared to selling it? And how are risks distributed between different techniques and locations (indoor, outdoor) and different sizes of operations? Part of the problem is technical: most members belonging to an active population of offenders are never arrested and thus are not available for census, and most cannot be reached through available survey methods used to asses the prevalence of drug use.

Capture–recapture methods have been shown to provide valid measures for various hidden populations. These methods build on the relative importance of a recurring pattern in an observed population (e.g. re-entry into treatment, re-arrest) to infer the proportion of this population that is active, but unobserved in the data. Capture–recapture methods have originally been developed in biology to estimate the size of animal populations (Seber 1973; Schwarz and Seber 1999), but have increasingly been integrated into the study of human populations mainly through epidemiological research. Capture–recapture models have been used extensively in the field of substance abuse to estimate the prevalence of drug users susceptible to treatment in a variety of communities (Bohning et al. 2004; Calkins and Atkan 2000; Hser 1993; Choi and Comiskey 2003; Hickman et al. 1999; Smit et al. 1997; Brecht and Wickens 1993). Following the pioneering analyses of Willmer (1970) and Greene and Stollmack (1981) on general populations of offenders, researchers also used capture-recapture methods to estimate populations of burglars (Riccio and Finkelstein 1985), car thieves (Collins and Wilson 1990), prostitutes (Rossmo and Routledge 1990) and their clients (Roberts and Brewer 2006), illegal gun owners (van der Heijden et al. 2003), and drug dealers (Bouchard and Tremblay 2005). Despite these few promising attempts, the methodology is still not well-known and its applications to criminology are just starting to be explored.

The present study proposes the use of a capture-recapture model––Zelterman’s truncated Poisson estimator––to analyze the size of a marijuana cultivation industry situated in a Canadian province (Quebec). Prevalence estimates cover both the number of active growers, and the number of cultivation sites they operate. These separate estimates allow for a comparison of the risks of arrest for growers, and the risks of detection for the cultivation sites, which are types of risks that bear different meanings and consequences for offenders. Estimates are differentiated across three cultivation techniques: (1) outdoor, (2) indoor, and (3) hydroponics. Zelterman’s model has been used in a handful of studies in the criminological literature but the model’s value and limitations are not well understood (Bouchard and Tremblay 2005; Collins and Wilson 1990; Choi and Comiskey 2003; Smit et al. 1997, 2002; Bohning et al. 2004). In the current study, the model is used on marijuana cultivation data for an extended time period (5 years). Confidence intervals are calculated and goodness-of-fit statistics are provided to assess the relative fit of the model to the data. Finally, a method is proposed for deriving the number of cultivation sites at risk of detection from the estimated population of growers.

Data and estimation model

Data

Arrest data for marijuana cultivation offenses are needed for the capture–recapture analysis on growers. Assuming a satisfactory prevalence estimate of marijuana growers, a combination of fieldwork data on co-offending patterns for individual cultivation sites and of police data on the size and distribution of cultivation-related seizures can be used to derive the prevalence of cultivation sites.

Arrest data

A population of offenders at risk of being arrested can be estimated from a distribution of arrested offenders for a given offense. The MIP data set (Module d’Informations Policières) which comprises all crime-related incidents in Quebec was used for this purpose. Arrests for which the first or second charge (or most serious charges) was for marijuana cultivation were considered as “marijuana cultivation arrests.” These arrests were extracted for the 1997–2003 period. Only those offenders for whom a date of arrest was precisely recorded were selected for the analysis. Because the objective is to uncover the number of re-arrested offenders, “duplicates” had to be avoided. Manual validation of every re-arrest resulted in the elimination of 3.5% of cases. For the seven-year period being studied, there was at least one marijuana cultivation charge in a total of 10,647 cases. These arrests involved 10,204 different offenders, 410 (4.0%) of whom were recidivists, that is, were arrested at least twice for a marijuana cultivation offense between 1997 and 2003. No offenders were arrested more than four times during the seven-year period. Police data classify cultivation offenses according to whether the plants are grown in soil (79% of all arrests) or hydroponically (or soil-less, comprising 21% of all arrests). Soil-based cultivation can be carried out either in outdoor or indoor settings, a distinction that is only made in seizure data. Hydroponic cultivation is always carried out indoors. Unless indicated otherwise, “indoor” will refer to the soil-based technique, and “hydroponic” to soil-less cultivation. The proportions of arrested females (15%), and the mean age of arrested offenders (34 years old) are similar for both cultivation methods, and similar to findings reported elsewhere (Plecas et al. 2002).

Seizure data

A second data set was needed to derive the prevalence of cultivation sites, comprising all the seizures made by the Quebec Provincial Police [QPP] for the years 2000 and 2001 (N = 3212). Seizures were classified in three categories: outdoor (65%), indoor (30%), and hydroponics (5%). Compared to the arrest data set, hydroponic seizures were underrepresented. Seizures with no arrests are common to outdoor cultivation cases (only 13.9% of seizures lead to an arrest), which increases the proportion of arrested offenders for hydroponic methods compared to other methods. In addition, all police organizations in the province participate to the MIP data set, whereas the data set on seizures lacks most indoor seizures made by Montreal’s police department (SPVM). Hence, the risks of detection faced by both types of indoor growers (hydroponics and soil-based) should be underestimated.1

The seizure data set was useful in providing a distribution of sizes for the cultivation industry as a whole.2 Similar to prevalence studies that distinguish between heavy and light drug users in order to account for large differences in drug use patterns (e.g. Everingham et al. 1995; Pudney et al. 2006), the estimation procedure that is developed in the current paper distinguishes between small non commercial sites (20 plants or less), small commercial sites (21–100 plants), and large commercial sites (101 + plants). As presented in Table 1, adding the three cultivation techniques creates nine types of cultivation sites. Descriptive seizure data show that size distribution differs by cultivation technique. Only a small percentage of sites operated for personal use are found for indoor and hydroponic sites, whereas more than a third of outdoor sites qualify as non commercial.3 Another finding is that, as the level of sophistication of the technique increases, so does the median number of plants grown.4 Overall, outdoor sites usually contain between 35 and 40 plants, the size increases to a little more than a 100 plants for indoor sites, and to nearly 350 plants for hydroponic ones.5
Table 1

Number of seizure cases and median number of plants grown per type of marijuana cultivation site, Quebec, 2000–2001

Type of cultivation sitea

Median size(# of plants)

Mean annual # of casesb

Percent of cases

Outdoor

    Small

9

358

36.8

    Medium

45

355

36.5

    Large

228.5

259

26.6

    Total

36.0

972

100

Indoor

    Small

7

75.5

19.8

    Medium

51

99.5

26.1

    Large

360.5

206

54.1

    Total

119.5

381

100

Hydroponic

    Small

18

1.5

1.9

    Medium

59

13.5

17.2

    Large

485

63.5

80.9

    Total

345.0

78.5

100

a Small: 20 plants or less; Medium: 21–100 plants; Large 101 + plants

b Cases in which the number of plants was not specified were removed from the analysis (N = 349, or 174.5/year)

Fieldwork data

Information on the co-offending patterns for cultivation sites of different sizes was also needed to derive the prevalence of cultivation sites. Such data were retrieved from a convenience sample of 20 interviewed growers who were active in the Quebec industry between 1993 and 2005. I interviewed nine growers between 2004 and 2006. Growers were recruited in a variety of informal ways,6 and interviews were conducted in cafés and pubs in Montreal and Quebec City. Information was gathered on many topics, including details on the dynamics of their career in cannabis cultivation, and a variety of issues touching the social and economic world of cannabis cultivation. Only information on key variables regarding co-offending patterns was analyzed for the purpose of this study.

I also obtained access to the written accounts of 17 interviews conducted by undergraduate students in a criminology class at Université de Montréal between 1998 and 2003. Students were asked by the professor to meet with a “successful delinquent”. Eleven interviews were added to the sample because they contained precise information on both of the key parameters used in this study: the number of plants grown for a specific crop, and the number of co-offenders involved from start to finish. The total sample of 20 growers were 95% males, they had a mean age of 27 years, were involved on cultivation sites of 118 plants (mean), and described the cultivation patterns of 34 crops (10 outdoor, 13 indoor, and 11 hydroponics). Only crops for which growers reported different parameters (e.g. change in the number of plants, co-offenders involved, or cultivation technique) were added to the fieldwork data set. Compared to arrest data, the interviewed sample comprised more males and younger growers who were involved on smaller cultivation sites. However, the correlation between the number of plants grown and the number of co-offenders involved is very high (r = 0.88, P < 0.001) and the range of sizes found in the sample of interviewed growers is wide (from 1 to 1,800 plants grown). Thus, simple OLS regression models should provide valid estimates of the number of co-offenders necessary for the median sizes derived from seizure data.

Estimation model

I start by presenting the estimation model that will be used to estimate the prevalence of marijuana growers. I will proceed with the results derived from that first model before considering the second model developed in this study to estimate the prevalence of cultivation sites.

Zelterman’s truncated Poisson estimator

Truncated Poisson methods such as Zelterman’s estimator (Z) provide the necessary framework for estimating hidden populations of offenders. If data on known arrests and re-arrests follow the Poisson distribution specified by Z’s model, the missing cell in the distribution should be estimated correctly, that is, the number of offenders with zero arrests. For data to follow a general Poisson distribution, a number of assumptions must be respected: (1) the population under study must be closed; (2) the population has to be homogenous; (3) the probability for an individual to be observed and re-observed must be held constant during the observation period (the independence assumption).

Such assumptions when using data on criminal populations may not be respected. The first and second assumptions pose obvious difficulties. Offenders tend to go in and out of offending at different periods of their lives; some are more active than others, and they may trigger different probabilities of arrest and re-arrest. Moreover, arrested offenders may modify their behaviour after an arrest, and police may also be tempted to over-target them following an arrest, leaving the third assumption unsatisfied.

Zelterman (1988) derived a truncated Poisson estimator designed to be robust to departures from these assumptions (see also Collins and Wilson 1990; Smit et al. 1997; 2002). It is given by

$$ {\text{Z = N / (1 - e}}^{{{\text{( - 2*n2/n1)}}}} {\text{)}} $$
(1)
where Z is the total population, N is the total number of individuals arrested with a marijuana cultivation charge, n1 is the number of individuals arrested once, and n2 is the number of individuals arrested twice in a given time period.

Zelterman’s Poisson estimator has a number of attractive features for estimating criminal populations. First, it can minimize the impact of population heterogeneity in arrest risks by eliminating the minority of high-rate offenders with multiple arrests. In fact, Eq. 1 shows that only those offenders arrested once (n1) or twice (n2) are considered for establishing the arrest rate parameter. Zelterman (1988) and other researchers who derived similar models (Chao 1989), base their approach on the rationale that estimation models should be complex enough to be meaningful, but simple enough to contain only the parameters that are necessary, and close to the quantity to be estimated: “Observations that are close to the object of interest should, intuitively, have more bearing on it” (Zelterman 1988, p. 227). In other words, using information on those offenders who are not arrested very often should be more meaningful in assessing the prevalence of non arrested offenders. The trade-off is that Zelterman’s estimator will generally provide less information, and more conservative estimates than more complex models that consider the full range of arrestees and their different arrest rates (e.g. an heterogenous Poisson model), or models that consider a series of covariates in fitting an estimation curve (e.g. a Poisson-based regression model). For example, compared to the 30,298 index offenders estimated by Greene and Stollmack’s (1981) heterogenous Poisson model for D.C. in 1975, the Z model derives an estimate of 29,842 offenders (a 2% underestimate). Compared to the 62,722 illegal gun possession offenders estimated by van der Heijden et al.’s (2003) Poisson-based regression model, the Z model derives a 50,866 offender estimate (a 23% underestimate). The arrest risks derived from Z will increase accordingly, but the effects are usually not dramatic: in the latter example, the estimated risks increase from 4% to 5% of illegal gun owners arrested annually.

Another advantage of the model is it can be used on only one sample (as with arrest data), whereas other capture–recapture approaches require three or more samples to derive estimates. The use of multiple samples is warranted in many situations. For example, one interested in estimating the total population of drug users should not strictly consider treatment data (entry and re-entry into treatment), but also possibly arrest and hospital records (Hser 1993). Using only arrest data confines the interpretation to prevalence estimates of offenders “at risk of being arrested,” which only concerns a minority of illegal drug users, but the majority of their suppliers (Bouchard and Tremblay 2005).

The Z estimator, however, assumes that the hidden population of interest is a “closed” population. The likelihood of severe departures from this assumption is minimized by the paper’s analysis of re-arrest distributions at an aggregate level (arrests and re-arrests at the provincial level) rather than at a city or neighborhood level. This procedure does not account for the fact that some offenders may go in and out of the criminally active population, but it reduces the possibility of offenders being excluded from the sample simply because they moved to another city or neighborhood. There is also some indication from the literature that, in most cases, using closed population models on open populations is not a major sin. Kendall’s (1999) simulations showed that animal prevalence estimates from closed population models are only minimally affected when movements in the population occur randomly, and when there is no sign of a massive emigration or immigration during the period under study. If the period under study is short enough, criminal population movements are unlikely to be swift and massive enough to have an impact on the prevalence estimates derived from closed population models such as Zelterman’s.

In order to estimate yearly variations in populations of active offenders at risk of being arrested, the analysis used a moving average which, at all times, included three years. For example, to estimate the population of growers at risk of being arrested in 1998, the arrestee population from 1997 to 1999 was pooled. For the year 1999, the year 1997 was dropped and 2000 was added. The strategy of using a three-year unit of estimation has clear advantages. First, it gives growers a reasonable length of time to get re-arrested and to start another cultivation site. Most arrested and convicted growers in Canada are not sentenced to incarceration, but when they are (in less than 20% of cases), sentence duration is less than 6 months (Plecas et al. 2002). Second, capture–recapture methods require some minimal level of re-arrests to function, and using only one year would not generate enough re-arrests for the estimator to be used. Recidivism in marijuana cultivation is slower because it requires some level of organization. Especially for indoor ventures, it can take a few weeks to find a (new) cultivation site, to convince other interested co-offenders, or to gather the necessary start-up capital. This is indicated in the data by the higher proportion of re-arrested offenders for soil-based growing (3.1%) than for hydroponics cultivation (1.3%). The option of using more than three years was also considered. However, this would have violated other assumptions, basically that growers remain in the criminally active population for as long as four or five years, which would have exaggerated the average career length of non-recidivist offenders.

Model evaluation

An important criterion in the choice of a model is that it provides a good fit to the data, that is, the distribution of arrests and re-arrests estimated by the model should resemble the observed distribution derived from the data. Table 2 presents the theoretical and observed distributions of soil-based and hydroponic growers arrested for the years 1997–2003.7 Neyman’s chi-square test was used to evaluate the fit.
Table 2

Theoretical (T) and Observed (O) arrest distributions for a marijuana cultivation offense, and goodness-of-fit values for zelterman’s estimator

# of arrests

1998

1999

2000

2001

2002

Soil-based

T

O

T

O

T

O

T

O

T

O

1

2643.1

2641

3206.9

3206

3814.5

3814

4051.2

4051

4306.8

4303

2

55

55

70

70

93

93

105

105

1161

116

3

0.8

3

1

1

1.5

2

1.8

2

2.1

6

4

0

0

0

1

0

0

0

0

0

0

X2

1.33

 

1.00

 

0.13

 

0.02

 

2.54

 

Hydroponic

1

616

616

808

808

1081

1081

1245

1245

1219

1219

2

8

8

8

8

14

14

17

17

17

17

3+

0

0

0

0

0

0

0

0

0

0

X2

0.00

 

0.00

 

0.00

 

0.00

 

0.00

 

Notes: Neyman’s X2 is given by \( {\text{X}}^{{\text{2}}} {\text{ = }}{\sum\limits_{{\text{j - 1}}}^{\text{n}} {\frac{{{\text{(n}}_{{\text{j}}} {\text{ - $ \hat{u} $ }}_{{\text{j}}} {\text{)}}^{{\text{2}}} }} {{{\text{n}}_{{\text{j}}} }}} } \)

Apparent from examining the distributions and the low or null chi-square values is that Zelterman’s estimator provides a near perfect fit to the data. The estimator performs well with the data because the distributions analyzed have very few cases of three or more arrests, parameters that are not captured by the model (see Eq. 1 supra). The smaller the proportion of multiple arrests (3+), the better the fit. Although goodness-of-fit tests are not truly decisive criteria in the choice of a model (Coull and Agresti 1999), such a good performance supports the use of the Z estimator and other similar models (e.g. Chao 1989) in this particular case. It should also be noted that Zelterman’s estimator has been shown to perform well as an estimator of hidden populations, even in circumstances where the model fit is weaker than what is the case in the present analysis (Bohning et al. 2004; Collins and Wilson 1990; Wilson and Collins 1992). Other criteria in model evaluation include an estimation of the intervals of confidence generated by the model. Those estimated in the present analysis (using Zelterman 1988) are narrow enough to be considered meaningful, especially for distributions with a larger proportion of re-arrests (such as the distributions for soil-based growers, see Table 2). The wider confidence intervals calculated for hydroponic growers illustrate the fact that capture–recapture estimates become more volatile when used on distributions that depend on lower proportions of recaptures. On the other hand, Zelterman (1988) designed the estimator to be robust especially with this type of data (Choi and Comiskey 2003).

Results

Table 3 presents the estimated populations of growers at risk of being arrested between 1998 and 2002 for the two categories found in arrest data: soil-based growers (indoor and outdoor), and hydroponic growers. Prevalence estimates were derived using Eq. 1 for each arrest distributions presented in Table 3. The pooled Z estimates were also divided by three to produce an annual population of marijuana growers in Quebec that can be compared.
Table 3

Estimated populations of soil-based and hydroponic growers, Quebec, 1998–2002

 

1998

1999

2000

2001

2002

Soil-based growers

Z estimatea

66,159

76,717

82,126

82,307

84,305

C.Ib

(61,845–71,121)

(72,561–81,379)

(78,382–86,245)

(78,781–86,163)

(80,904–88,004)

Arrests (N)

0c

63,460

73,439

78,217

78,149

79,880

1

2,641

3,206

3,814

4,051

4,303

2

55

70

93

105

116

3

3

1

2

2

6

4

0

1

0

0

0

Annual population at risk of arrestd

22,053

25,572

27,375

27,436

28,102

Hydroponic growers

Z estimatea

24,337

41,617

42,825

46,845

44,935

C.I.b

(18,635–35,068)

(33,686–54,434)

(36,466–51,870)

(40,597–55,183)

(38,933–53,126)

Arrests (N)

0c

23,713

40,801

41,730

45,583

43,699

1

616

808

1,081

1,245

1,219

2

8

8

14

17

17

3+

0

0

0

0

0

Annual population at risk of arrestd

8,112

13,872

14,275

15,615

14,978

a As estimated by Eq. 1, for three years (e.g. 1998 = 1997–1999)

b Confidence intervals, as estimated using Zelterman (1988: 228, Eq. 7)

c Z estimate––number of arrested offenders

d Z estimate/3, to reflect a moving average population for each year

Based on the arrest and re-arrest distributions presented in Table 3, the model estimates that there were almost twice as many soil-based growers (28,102) as hydroponic growers (14,978) in 2002. Simplifying the logic of the model, these estimates are derived by combining two pieces of information: the total number of growers arrested (N in Eq. 1), and the proportion of offenders re-arrested for a specific offense (−2*n2/n1). For example, for the 2001–2003 period (or 2002 in Table 3), there were 3.6 times as many soil-based arrests as hydroponic arrests (4,553 vs. 1,553 total arrests), but also twice as many soil-based growers re-arrested (2.76% vs. 1.38% re-arrests). The model assumes that this lower proportion of hydroponic growers re-arrested translates into lower arrest risks for these growers as a whole. As a result, instead of assuming that there are 3.6 times as many soil-based growers as hydroponic growers, the model estimates that the difference is almost twice as small (1.9).

Adding the annual estimates for hydroponic and soil-based cultivation in Table 3, the Z estimates reveal that an annual average of 30,000–45,000 offenders were active and at risk of being arrested for a marijuana cultivation offense between 1998 and 2002 in Quebec. This estimate of the size of the industry is an interesting finding in itself. One can first appreciate the meaning of such an estimate through a comparison with the number of marijuana users in the province. The 2004 Canadian Addiction Survey indicated that 15.8% of the Quebec population aged 15 and older had used marijuana in the year preceding the survey; this amounted to 990,531 estimated users. Assuming that all growers are at least occasional users, this number would imply that between 3% and 4.5% of self-reported past-year marijuana users would also be marijuana growers. Unfortunately, the Canadian survey does not ask respondents whether or not they grow marijuana, but the figures presented in this paper are comparable to findings reported elsewhere. For example, a survey conducted in New Zealand indicated that 3.4% of marijuana users interviewed had grown “most to all” the marijuana they consumed (Field and Casswell 1999). Another survey conducted in Amsterdam indicated that 17 out of 214 (8%) marijuana users were also growers at the time of the interview (Cohen and Kaal 2001).8 The prevalence of marijuana growers in Quebec derived from capture-recapture analysis appears to be reasonably lower than the level found in an important producing region like Amsterdam, and interestingly similar to New Zealand—a country which also shows comparable prevalence rates of cannabis use.

Perhaps more informative is a comparison of these estimates with similar estimates of the prevalence of marijuana dealers in this region. Using the same model, Bouchard and Tremblay (2005) estimated that close to 45,000 offenders were at risk of being arrested for a marijuana dealing offense in 1998 in Quebec, making the marijuana dealing and cultivation trades roughly similar in size. This finding is important, because it offers a representation of market structure that is different from the pyramidal view that characterizes the distribution chain of most importation-driven illegal industries. A lengthy distribution chain that widens slowly down to consumers becomes less important here. One can hypothesize that the mere prevalence of growers makes it possible that most users can get supplies directly from growers, or they remain only one handshake away from access to a crop. A classical chain distribution probably remains to distribute the marijuana coming out of the largest cultivation sites. However, as far as size is concerned, these findings suggest that the cultivation industry is now a significant criminal opportunity for offenders in Quebec.

The estimates presented in Table 3 also show that the prevalence of marijuana growers at risk of being arrested increased during the period under study: from 30,000 growers in 1998 to 45,000 in 2002. The rise is consistent with trends in drug use in Quebec. The 2004 Canadian Addiction Survey indicated that the number of individuals who had consumed marijuana in the year preceding the survey increased by 23% in Quebec since 1998 (compare Daveluy et al. 2000, and Canadian Center for Substance Abuse 2004). The increasing trend is also consistent with government reports asserting that Canada is exporting increasingly larger amounts of marijuana to the United States in recent years (NDIC 2004). But perhaps more importantly, the growth almost exactly parallels the rise in the number of legal “grow shops” selling cultivation equipment and supplies to plant growers. The number of grow shops increased from 50 to 82 between 1998 and 2002 (Bouchard 2007), a 64% increase that compares to the 50% increase in the prevalence of marijuana growers illustrated in Table 3. It would take a few more years of observation to determine a real trend for the marijuana industry and to be able to interpret it properly, but the estimates suggest that the development of the industry may have reached a plateau in recent years, after a period of rapid growth. Incidentally, the prevalence of grow shops also reached a plateau after 2003 in Quebec (Bouchard 2007).

Criminal population estimates can be used as a relevant denominator to calculate a different type of arrest risk than what is typically presented in studies interested in these issues. Instead of considering “how many arrests for how many crimes” (Blumstein et al. 1986), criminal population estimates allow consideration of the following measure: how many arrests for how many offenders at risk? Such a measure is especially important for drug market crimes for which the total number of crimes (e.g. the total number of drug transactions conducted in a year) is not necessarily as meaningful a risk measure than for most predatory crimes (Bouchard and Tremblay 2005).

Table 4 presents the annual risks of being arrested for a cultivation offense according to the technique used by growers. Three observations stem from this analysis. First, the risks of being arrested for a cultivation offense are quite low (between 2 and 5%), at least compared to similarly derived risks for drug dealers (3–7%) in Quebec (Bouchard and Tremblay 2005). Despite the large increase in resources invested in marijuana eradication programs in Quebec, still very few growers are apprehended. Second, risks are more than twice as low for hydroponic growers (2–3%) compared to offenders involved in soil-based methods (4–5%).9 This finding is surprising, because the police are not likely to know, before intervening on a site located indoors, which specific technique (soil-based or hydroponics) is used by the growers.10 In addition, data on seizures indicate that hydroponic cultivation sites are, on average, four to five times as large as soil-based sites (Table 1). This finding suggests that hydroponic growers might be more successful at avoiding detection than other indoor growers. Given the average size of their cultivation sites (median of 350 plants), they certainly have more to lose than other types of growers, and are more likely to have the financial capital necessary to invest in protection devices.
Table 4

Annual prevalence of soil-based and hydroponic growers and the risks of being arrested for a marijuana cultivation offense, Quebec, 1998–2002

 

1998

1999

2000

2001

2002

Soil-based

Prevalence of growers at riska

22,053

25,572

27,375

27,436

28,102

Number of arrestsb

920

1118

1332

1422

1518

Annual risks of arrest (%)c

4.2

4.4

4.9

5.2

5.4

Hydroponic

Prevalence of growers at riska

8112

13,872

14,275

15,615

14,978

Number of arrestsb

211

275

370

426

418

Annual risks of arrest (%)c

2.6

2.0

2.6

2.7

2.8

a Z estimate, from Table 3

b Mean number of arrests for three-year period (e.g. 1998 = mean number of arrests for 1997–1999)

c Number of arrests/prevalence of growers at risk

Lastly, Table 4 illustrates the importance of analyzing the trends in arrest and re-arrest patterns, instead of just the absolute number of offenders arrested for a specific offense. The third and sixth rows of Table 4 present the number of arrests for soil-based and hydroponic cultivation, respectively. Both trends suggest that marijuana cultivation may have become more risky during that time period: The number of arrests for soil-based cultivation increased by 65% while the number of arrests for hydroponic cultivation almost doubled, from 211 to 418 arrests. However, the capture–recapture analysis suggests a very different picture. The risks for hydroponic growers remained mostly unchanged during that period, while the population of soil-based growers rose enough to absorb most of the parallel rise in arrests.

The prevalence of cultivation sites and the risks of detection

Risks of arrest for marijuana growers should be distinguished from risks of detection for cultivation sites. Estimating the risks of detection is important because losses for these offenders can be significant, especially for indoor and hydroponic cultivation sites. Furthermore, because a significant proportion of outdoor growers are never arrested, seizures often represent the only costs incurred by these offenders. What proportion of cultivation sites are detected by law enforcement?

The strategy used to estimate the number of cultivation sites begins by adjusting the prevalence estimates of soil-based growers (see Table 6, Appendix A). Adjustments are necessary for two reasons. First, the prevalence of cultivation is derived from the prevalence of growers “at risk of being arrested.” Not all growers may be at risk of being arrested. For example, seizure data show that only 14% of outdoor cases lead to at least one arrest. As expected, the proportion is much higher for indoor (76%), and hydroponic cases (95%). Adjusting the prevalence of soil-based growers to take these cases into account, the estimates for soil-based growers increase from 28,102 to 39,733 in 2002 (Table 6, Appendix A). Second, arrest data do not distinguish between outdoor soil-based growers from those who grow in indoor settings. This distinction is important for an analysis of the risks of detection: outdoor sites are likely to be more vulnerable to detection than indoor cultivation sites located in private houses or apartments. Fortunately, seizure data make that distinction. For example, it is estimated that among the 39,733 estimated soil-based growers for 2002, 37% are outdoor, and 63% are indoor growers (Table 6, Appendix A).

Prevalence estimates of marijuana cultivation sites can be derived from these adjusted populations of growers. The calculation could be labeled as a “division of labor” approach, as it uses information on the specific co-offending patterns of different types of cultivation sites, allowing for the possibility of economies of scale for large commercial sites. From Eq. 1 supra is derived a prevalence of cultivation sites, which is given by

$$ {\text{S = }}\sum {\text{(Z}}_{{\text{i}}} {\text{/c}}_{{\text{i}}} {\text{)}}\lambda _{{{\text{i,n}}}} $$
(2)
where S is the annual number of cultivation sites at risk of detection, Z is the prevalence of growers of type i, c is the number of co-offenders working on a median size of type i, and λ represents the proportion of seizures for type i and of sizes n.
As illustrated in Table 5, prevalence estimates for nine types of cultivation sites can be derived from Eq. 2. The Z estimated prevalence of growers for 2002 were taken for the calculations (column 2). Median size (column 3) and the distribution of seizures according to size and type of cultivation site (column 5) were retrieved from seizure data (see also Table 1). The number of co-offenders for a given median size (column 4) has been estimated by linear regression analysis, using fieldwork data. Regressing the number of plants harvested by growers on the number of co-offenders involved in the cultivation site from start to finish provided a satisfactory and interesting measure for the purpose of the current study.11 All three cultivation techniques required a minimum of three co-offenders to set up even a small cultivation site, but the larger hydroponic sites benefited from economies of scale. Both large outdoor and hydroponic sites require the collaboration of 5–6 co-offenders, but hydroponic sites are more than twice as large (Table 5).
Table 5

Annual prevalence and risks of detection by type of cultivation sites, Quebec, 2000–2001

Type of cultivation site

Prevalence of growersa

Median size (# of plants)

Co-off/med sizeb

Percent of Cases

Prevalence of cultivation sitesc

Mean annual # of cases

Risk of detection

Outdoor

    Small

14,644

9

2.9

36.8%

1,858

358

19.3%

    Medium

14,644

45

3.3

36.5%

1620

355

21.9%

    Large

14,644

228.5

5.5

26.6%

708

259

36.6%

Indoor

    Small

25,089

7

3.0

19.8%

1,656

75.5

4.6%

    Medium

25,089

51

3.4

26.1%

1,926

99.5

5.2%

    Large

25,089

360.5

5.9

54.1%

2,301

206

9.0%

Hydroponic

    Small

14,978

18

3.1

1.9%

92

1.5

1.6%

    Medium

14,978

59

3.4

17.2%

758

13.5

1.8%

    Large

14,978

485

5.8

80.9%

2,089

63.5

3.0%

Total

    

13,008

1431.5d

11.0%

a Adjusted prevalence figures for outdoor and indoor growers for 2002 (from Table 6, Appendix A)

b As estimated through OLS regression, using fieldwork data. See footnote 11

c As estimated by Eq. 2

d The mean number of seizures for 2000–2001 is 1606. However, the number of plants was not specified in 10.8% of cases (or 174.5 per year). Adding these cases increase the risks of detection by 1.3%, or from 11.0% to 12.3%

Columns 2–5 are used in the calculation of S (column 6). For example, to estimate the number of indoor sites of more than 100 plants, I divide 25,089 by 5.9 co-offenders for a median size of 360.5 plants, and then multiply by 0.54—the seizure rate for outdoor sites of more than 100 plants to obtain a prevalence of 2,301 of such cultivation sites. Overall, Table 5 reveals that the 54,711 estimated marijuana growers in 2002 were active in about 13,008 cultivation sites in Quebec (mean of 4.2 co-offenders per site). Whereas the prevalence of outdoor and hydroponic growers was found to be similar (column 2), the smaller size of outdoor sites increases their overall prevalence compared to hydroponic sites (4,172 vs. 2,925). However, the larger size and higher productivity of hydroponic sites gives them a larger share of the total marijuana production in the industry.

Risks of detection were calculated for the different types of cultivation sites (Table 5, last column) by dividing the mean annual number of seizures (column 7) by the estimated prevalence of cultivation sites (column 6). Because the majority of cultivation cases start from a seizure but not all cases lead to an arrest, risks of detection were expected to be larger than the arrest risks calculated earlier (Table 4). Table 5 shows that the mean risks of detection for outdoor and indoor sites were 19–37% and 5–9% respectively, compared to risks of arrest of the order of 4–5% for growers involved in such sites. More similar risks of arrest and risks of detection were found in the hydroponic industry, perhaps a consequence of the high rates of arrest following a seizure (95%).

These findings confirmed a general impression from drug market commentators that outdoor sites are more vulnerable to detection than indoor greenhouses.12 Interestingly, the magnitude of the estimated risks of detection is in line with other research on “rates of interception” at the import or production levels. Using some of the best data and estimation procedures available, Rhodes et al. (2000) suggested that US authorities were seizing between 10 and 15% of heroin, and 20–25% of cocaine entering the US. Wilkins et al. (2002b) recently estimated that risks of detection for outdoor marijuana production sites vary between 26 and 32% in New Zealand, acknowledging that these rates appear higher than those being recorded overseas. Based on the analysis undertaken for this paper, it is estimated that between 19 and 37% of outdoor cultivation sites are eventually detected.

Contrary to the aforementioned studies, however, the present analysis offers the advantage of estimating risks of detection according to the size of the cultivation site. By simultaneously presenting data on the size of a criminal enterprise and on the risks of detection, the proposition that risks increase with the size of illegal ventures can be tested. The relationship between risks and size has often been explored in different contexts (most notably in Reuter 1983, 1985), yet, scarcely measured empirically. In the case of marijuana cultivation, one could hypothesize that larger sites are more vulnerable to seizures because of the increased number of co-offenders who participate (the group hazard hypothesis), because of the heightened risk of system failures for large indoor sites (e.g. floods, fire), and because of the mere “visibility” that comes with larger sizes, especially for outdoor sites.

Not unexpectedly, Table 5 shows that risks of detection do increase with the size of cultivation sites. This finding holds, however, only when one considers each cultivation technique individually. Outdoor sites that are maintained for growers’ own marijuana consumption receive less police attention (19%) than small-scale outdoor commercial sites for which risks of detection increase to 22%. Large-scale outdoor sites appear to be the most vulnerable to detection as close to 37% are being seized on an annual basis. Risks increase at a less pronounced rate for indoor sites. From 4.6% for non-commercial indoor ventures, risks of detection slightly decrease to 5.2% for mid-size operations, and then to 9% for large indoor commercial sites. Risks also appear to increase with size for hydroponic sites, but the risk levels are so low that the absolute change remains trivial. Risks of detection vary between 2 and 3%, between three and four times less than the figures estimated for indoor sites. Recall that the median number of plants grown for large-scale hydroponic facilities was 485 plants in 2000–2001, compared to 360.5 plants for indoor sites. The relative impunity of hydroponic sites, compared with much smaller outdoor sites, could encourage growers to invest in even larger ventures, if they can.

Were enforcement-related risks underestimated in the hydroponic marijuana industry? Additional analyses conducted to evaluate the sensitivity of the estimates to a change in some of the parameters used for the estimation did not alter the findings in any fashion.13 Risk differences between size and cultivation techniques are too important to be explained as measurement artifacts.

Discussion and conclusion

Successful criminal innovations are those that either increase criminal gains or decrease the risks of being arrested, or both at the same time (Lacoste and Tremblay 2003). The proposition that the massive development of a criminal innovation is due to an increase in risks for a substitute criminal opportunity is plausible (e.g. the risk/adaptation hypothesis), but it may be difficult to demonstrate. This is partially due to a lack of methodological tools for estimating market or method-specific populations of offenders, and for tracking their evolution. Also, there is a lack of comparable measures of displacement inducing factors, for example, differential pay-offs per crime, or a systematic analysis of the differential enforcement-related risks faced by different types of marijuana growers, involved in different cultivating practices, and working at different sizes of operations.

This paper illustrates the feasibility of using Zelterman’s (1988) truncated Poisson estimator for determining population and risk estimates for a variety of offenders involved in the marijuana cultivation industry. A first set of findings suggests that the size of the industry, as measured by the number of growers, is substantial. Estimates indicate that about 45,000 growers were at risk being arrested in 2002 in Quebec, which equals the number of marijuana dealers in the province (Bouchard and Tremblay 2005). Method-specific estimates indicate that there are more growers working in indoor settings than outdoor settings. Conversely, outdoor cultivation sites are still found in large proportions, partly because they are typically smaller in size. It also appears that soil-based methods are not disappearing to the benefit of hydroponics, despite the enhanced productivity associated with the more sophisticated techniques.

An interesting spin-off provided by method-specific population estimates is the possibility of comparing the aggregate level of risks faced during the course of one year by marijuana growers involved in different cultivation techniques. Method-specific comparisons indicate certain inequalities in the distribution of risks. Risks of arrest are lowest for hydroponic growers compared to growers involved in soil-based methods. This finding persists when considering the risks of detection. As expected, risks of detection do increase with size, but only within a particular technique. Hydroponic facilities, even the largest ones, have a 2–3% risk of detection, compared to risks of 5–9% and 19–37% for indoor and outdoor sites respectively. Hydroponic growers appear to be better than other growers at protecting themselves and their cultivation sites from police detection.

These findings are subject to the validity of the prevalence estimates derived from Zelterman’s (1988) truncated Poisson estimator. These estimates are based on a number of assumptions that are likely to be violated and the impact of the violations is not always easy to assess. A first unresolved issue concerns the independence assumption, which asserts that the occurrence of one event (e.g. an arrest) will have no incidence on the probability of subsequent events. In other words, the independence assumption excludes the possibility of behavioral change that could reduce the risks for a particular offender (e.g. through a deterrent effect) or increase them (e.g. the police targeting past arrested offenders). Both scenarios are possible but none are expected to be important enough to have an impact on the prevalence estimates. Drug offenders typically show very high rates of recidivism following incarceration or probation (Spohn and Holleran 2002), but such rates could not be calculated for the purpose of the current study. With regard to police targeting previously arrested offenders, it should be noted that only one offender was re-arrested more than three times during a seven-year span. Another issue with the independence assumption is that information on the arrest of one offender should not be viewed as providing information on the arrest of another offender. However, arrest data show that a significant proportion of growers are arrested with at least one other co-offender. The violation of this assumption is almost inevitable when using arrest data, but I am not aware of any study that tried to assess its effect on the population estimates. The effects of violating this particular assumption when estimating criminal populations should be assessed in future research.

Second, the study was designed to limit departures from the closed population assumption (no entries and exits during the capture–recapture experiment) by concentrating on a large geographical area (Quebec) and on a reasonably short time frame (three years), given the average career length of marijuana growers and the mean sentence meted out for those who are convicted. Open population models have the disadvantages of being much more complex and data intensive than closed population models, and the results are not always satisfactory (e.g. the Jolly-Seber model in Hser 1993; Brecht and Wickens 1993). However, these models offer a more realistic representation of the dynamics both in and out of criminal careers, and perhaps more precise estimates of the size of criminal populations. A possible hypothesis is that a Markov chain model could better depict the patterns in and out of outdoor marijuana cultivation—a seasonal criminal offense with high probabilities of arrest between August and October, and much lower the remainder of the year.

Finally, the homogeneity assumption (the population is homogenous with regard to arrest probabilities) may not be a major issue in this study. First, Zelterman’s truncated Poisson model assumes that the capture probabilities of those offenders most rarely arrested follow a distinct pattern that should be estimated separately. Second, analyzing subgroups of growers who were homogenous relative to the cultivation technique that they used (soil-based versus hydroponic) also reduces heterogeneity. However, the distribution of risks within these populations may still vary widely. Although variations in the level of risks within a group may not constitute a violation of the Poisson distribution if risks are non zero (van der Heijden et al. 2003), researchers should pay attention to the distribution of risks—especially relative to the different roles assumed by offenders. An aggregate measure of risks like the one estimated in this study is useful for uncovering important macro level patterns, for example, that hydroponic growers almost completely avoid police detection compared to other types of growers. But one wonders whether or not a 5% risk level is meaningful for an important proportion of soil-based growers, or whether risks are more lopsidedly distributed, with low-level employees receiving the bulk of police attention.

One of the main findings of this study is that most hydroponic growers may receive very important financial incentives from their activities (given the size of the average cultivation site, their higher productivity rates, and the higher prices at which they can sell their product14) while systematically avoiding police detection. Hydroponic growers’ dual advantage (low risks, high gains) over others may have important implications for the future developments of the marijuana industry. The success of hydroponics as a criminal innovation might induce other growers or would-be growers (including offenders involved in other crimes) to adopt this particular technique. The higher the prevalence of hydroponic cultivation sites, the more efficient offenders are, and the higher the organizational capacity of the marijuana industry. If detection of hydroponic sites continues to pose difficulties to police agencies, an increase in the prevalence of hydroponic growers will make the bulk of arrests continue to fall on the most vulnerable and less organized outdoor cultivation sites, further augmenting inequalities in the distribution of risks in the marijuana cultivation industry.

Whether or not the hydroponic technique “catches on” and continues to diffuse and be adopted by an increasing number of offenders does not strictly depend, however, on a cost/benefit analysis. Weisheit (1991, 1992) and Hough et al. (2003) demonstrated that marijuana cultivation, for a large number of growers, is not just about money but also about the “love of the plant.” Many growers are part of what may be described as a marijuana subculture, cultivating marijuana for “intangible rewards”, but their market significance is likely to be minor in terms of the total amount of marijuana produced. Perhaps more decisive at the macro level, hydroponic cultivation implies relatively large sizes to be advantageous. Offenders face limited opportunities for marijuana cultivation in indoor settings, and large-scale marijuana cultivation requires the common pooling of more co-offenders; first to produce, supervise, and run the installations, but also to distribute very large quantities of marijuana to consumers. The fact that outdoor cultivation provides new adopters with a smoother entry into the trade, along with the possible influence of demand-related factors (low prices, taste?), has probably prevented a faster or more significant shift to hydroponic cultivation in recent years.

Footnotes

  1. 1.

    The SPVM is assumed to intervene on a similar number of indoor and hydroponic cases than the rest of the province. For example, a newspaper article reported that the SPVM discovered 28 hydroponic greenhouses in 1998 (Breton 2000), whereas our QPP data indicate that they discovered 31 sites for that same year. This underestimate of indoor and hydroponic seizures will be taken into account when analyzing the risks of detection later in the paper.

  2. 2.

    Using data on seizures may not reflect the distribution of sizes for the industry as a whole but only those at risk of being detected, a convenient bias for an analysis concerned with providing estimates for this type of sample.

  3. 3.

    The very low proportion of non commercial sites found in the sample is interesting. Qualitative studies like those of Weisheit (1992), or Hough et al. (2003) almost exclusively interviewed small-time growers which gave the impression that they represented the majority of growers. Conversely, by relying on police data, the current study probably overemphasizes larger cases, but nonetheless demonstrates that they are far from scant, at least in the region under study.

  4. 4.

    It is likely that these figures on the number of plants per cultivation site are inflated, because police typically treat all types of plants equally: plants of low quality, or baby plants, are counted even though only a variable amount of these will reach maturity. The inflation rate is unknown, but is not a major problem for the purpose of this study as it is likely to be constant for all types of cultivation sites. However, the inflated figures would be problematic for a different study that wanted to estimate the quantity of marijuana produced in the province. Such an estimate would also be inflated.

  5. 5.

    The median is used because the distribution of sizes is highly skewed. The mean number of plants seized is 128, 372, and 816 plants for outdoor, indoor, and hydroponic sites, respectively.

  6. 6.

    One respondent was referred by a colleague criminologist who was supervising him while he served the end of a federal sentence in a halfway house in Montreal after being found guilty of cannabis cultivation. Two other respondents were referred by a criminology student after a seminar I taught on cannabis cultivation. The six others were referred to me by mutual acquaintances after learning about the research.

  7. 7.

    The theoretical distributions can easily be estimated by creating a spreadsheet similar to the Poisson distribution calculator available on Carnegie Mellon University Department of Biology’s website: www.bio.cmu.edu/courses/03438/PBC97Poisson/PoissonCalc.xl. The arrest rate parameter necessary for such calculation must first be estimated by Eq. 1, i.e. by dividing the total number of arrests by Z, the estimated prevalence of growers.

  8. 8.

    The survey also included the cities of San Francisco and Bremen, Germany, but findings were either unclear, or the amount of users interviewed insufficient to reach any conclusions. For example, only one user out of 262 in San Francisco reported growing marijuana at the time of interview, but more than 79 said they had done so in their lifetime. In Bremen, 4% of respondents said they grew marijuana at the time of the interview, but the sample is simply too small (N = 50) to make any inference about the prevalence of growing among users in this city.

  9. 9.

    Note that the risks for soil-based growers remain higher overall, even when the adjustment procedure is taken into account (Table 4, Appendix A).

  10. 10.

    Because very few outdoor soil-based growers are likely to be at risk given the low percentage of outdoor seizure that lead to an arrest (14%), the figures presented for soil-based cultivation mostly concern indoor growers.

  11. 11.

    I used a simple linear regression model of the form C = a + b*p, where C is the number of co-offenders per site and p is the number of plants grown per site. The regression coeffients are presented as following: for outdoor sites (n = 10): C = 2.805 + 0.0116*p; for indoor sites (n = 13): C = 2.955 + 0.0082*p; for hydroponic sites (n = 11): C = 2.981 + 0.0057*p.

  12. 12.

    The finding holds even when increasing the number of seizure cases by a factor of two––to compensate for the absence of most cases from Montreal.

  13. 13.

    One rather extreme scenario is to assume economies of scale for outdoor sites instead of hydroponic sites. For example, increasing the number of offenders necessary to produce 485 hydroponic plants from 5.8 to 7.8 produces an increase in risks from 3% to 4%, whereas a similar inverse operation for outdoor sites (reducing crew size from 5.5 to 3.5 co-offenders) decreases the risks of detection from 37% to 23%. In the end, the difference in risks between large outdoor and hydroponic sites (23% vs. 4%) remains very important.

  14. 14.

    Hydroponically produced marijuana is typically stronger in the drug’s psychoactive content (Tetrahydrocannabinol, or THC) than soil-produced herb. Interviewed growers report that it is also sold at higher prices, and hydroponic growers can cultivate more crops per year than with any other method.

Notes

Acknowlegements

I would like to thank Pierre Tremblay for his decisive comments on an earlier version of this paper. Peter Reuter, Therese Brown, Carlo Morselli, Maurice Cusson, Mathieu Charest, Julien Piednoir, Sue-Ming Yang and three anonymous reviewers also provided useful suggestions. Finally, I am grateful for the contributions of Marteen Cruijff, Paul Fugère, Chloé Leclerc, Maïa Leduc, and Barbara Wegrzycka in analyzing some of the data presented in the paper.

References

  1. Blumstein A, Cohen J, Roth JA, Visher CA (1986) Criminal careers and “Career Criminals”, vol 1. National Academy Press, Washington, DCGoogle Scholar
  2. Bohning D, Suppawattanabodee B, Kusolvisitkul W, Viwatwongkasem C (2004) Estimating the number of drug users in Bangkok 2001: a capture–recapture approach using repeated entries in one list. Eur J Epidemiol 19:1075–1083CrossRefGoogle Scholar
  3. Bouchard M, Tremblay P (2005) Risks of arrest across markets: a capture–recapture analysis of ‘hidden’ dealer and user populations. J Drug Issues 34:733–754Google Scholar
  4. Bouchard M (2007) Growers and facilitators: analyzing the development of the cannabis cultivation industry Unpublished working paper. Department of criminology and criminal justice, University of MarylandGoogle Scholar
  5. Brecht M-L, Wickens TD (1993) Application of multiple-capture methods for estimating drug use prevalence. J Drug Issues 23:229–250Google Scholar
  6. Breton P (2000) Les serres hydroponiques sont plus nombreuses. La Presse: E3Google Scholar
  7. Calkins RF, Atkan GB (2000) Estimation of heroin prevalence in Michigan using capture-recapture and heroin problem index methods. J Drug Issues 30:187–204Google Scholar
  8. Canadian Center for Justice Statistics (2001) Uniform Crime Report Program. Statistic Canada, OttawaGoogle Scholar
  9. Canadian Centre on Substance Abuse (2004) Canadian Addiction Survey. Canadian Centre on Substance Abuse, OttawaGoogle Scholar
  10. Caulkins JP, Behrens DA, Knoll C, Tragler G, Zuba D (2004) Markov chain modeling of initiation and demand: the case of the US cocaine epidemic. Health Care Manag Sci 7:319–329CrossRefGoogle Scholar
  11. Caulkins JP, Reuter P (1998) What price data tell us about drug markets?. J Drug Issues 28:593–612Google Scholar
  12. Chalsma AL, Boyum D (1994) Marijuana situation assessment. Office of National Drug Control Policy, Washington, DCGoogle Scholar
  13. Chao A (1989) Estimating population size for sparse data in capture–recapture experiments. Biometrics 45:427–438CrossRefGoogle Scholar
  14. Chin V, Dandurand Y, Plecas D, Segger T (2000) The criminal justice response to marihuana growing operations in BC. Department of Criminology and Criminal Justice, University College of the Fraser Valley, AbbotsfordGoogle Scholar
  15. Choi Y, Comiskey C (2003) Methods for providing the first prevalence estimates of opiate use in Western Australia. Int J Drug Policy 14:297–305CrossRefGoogle Scholar
  16. Cohen PDA, Kaal HL (2001) The irrelevance of drug policy: patterns and careers of experienced marijuana use in the populations of Amsterdam, San Francisco and Bremen. Cedro/Uva, Dutch Ministry of Health, Welfare, and SportsGoogle Scholar
  17. Collins MF, Wilson RM (1990) Automobile theft: estimating the size of the criminal population. J Quant Criminol 6:395–409CrossRefGoogle Scholar
  18. Coull BA, Agresti A (1999) The use of mixed logit models to reflect heterogeneity in capture–recapture studies. Biometrics 55:294–301CrossRefGoogle Scholar
  19. Daveluy C, Pica L, Audet N, Courtemanche R, Lapointe F (2000) Enquête Sociale et de Santé 1998. Institut de la Statistique du Québec, QuébecGoogle Scholar
  20. Everingham SMS, Rydell PC, Caulkins JP, (1995) Cocaine consumption in the United States: estimating past trends and future scenarios. Socio-Econ Plan Sci 29:305–314CrossRefGoogle Scholar
  21. Field A, Casswell S (1999) Drugs in New Zealand: a National Survey 1998. Alcohol and Public Health Research Unit, University of AucklandGoogle Scholar
  22. Greene MA, Stollmack S (1981) Estimating the number of criminals In Fox JA (ed), Models in quantitative criminology. Academic Press, New YorkGoogle Scholar
  23. Hafley SR, Tewksbury R (1995) The rural Kentucky marijuana industry: organization and community involvement. Deviant Behav 16:201–221 CrossRefGoogle Scholar
  24. Hickman M, Cox S, Harvey J, Howes S, Farrell M, Frischer M, Stimson G, Taylor C, Tilling K (1999) Estimating the prevalence of problem drug use in inner London: a discussion of three capture–recapture studies. Addiction 94:1653–1662CrossRefGoogle Scholar
  25. Hough M, Warburton H, Few B, May T, Man L-H, Witton J, Turnbull P (2003) A growing market: the domestic cultivation of marijuana. Joseph Rowntree Foundation, YorkGoogle Scholar
  26. Hser Y (1993) Population estimation of illicit drug users in Los Angeles county. J Drug Issues 23:323–334Google Scholar
  27. Kendall LW (1999) Robustness of closed capture–recapture methods to violations of the closure assumption. Ecology 80:2517–2525Google Scholar
  28. Jansen ACM (2002) The economics of marijuana cultivation in Europe Paper presented at the 2nd European Conference on drug trafficking and law enforcement, Paris, September 2002Google Scholar
  29. Lacoste J, Tremblay P (2003) Crime and innovation: a procedural analysis of patterns in check forgery. In: Smith MJ, Cornish DB (eds) Theory for practice in situational crime Prevention. Criminal Justice Press, Monsey, NY, pp 169–196Google Scholar
  30. National Drug Intelligence Center (2004) National Drug Threat Assessment 2004, National Drug Intelligence Center, US Department of JusticeGoogle Scholar
  31. Plecas D, Dandurand Y, Chin V, Segger T (2002) Marihuana growing operations in British Columbia: an Empirical Survey (1997–2000). Department of Criminology and Criminal Justice, University of the Fraser Valley, AbbotsfordGoogle Scholar
  32. Potter G, Gaines L, Holbrook B (1990) Blowing smoke: an evaluation of marijuana eradication in Kentucky. Am J Police 9:97–116Google Scholar
  33. Pudney S, Badillo C, Bryan M, Burton J, Conti G, Iacovou M (2006) Estimating the size of the UK illicit drug market. In: Singleton N, Murray R, Tinsley L (eds) Measuring different aspects of problem use: methodological developments. Home office online report 16/06, pp 4—120, available at: http://wwwhomeofficegovuk/rds/pdfs06/rdsolr1606pdfGoogle Scholar
  34. Reuter P (1983) Disorganized crime: the economics of the visible hand. MIT Press, Cambridge, MassGoogle Scholar
  35. Reuter P (1985) The organization of illegal markets: an economic analysis. National Institute of Justice US Department of Justice, Washington, DCGoogle Scholar
  36. Reuter P, Crawford J, Cave J (1988) Sealing the borders: the effects of increased military participation in drug interdiction. Rand, Santa MonicaGoogle Scholar
  37. Reuter P, Kleinman MAR (1986) Risks and prices: an economic analysis of drug enforcement. In: Tonry M, Morris N (eds) Crime and justice: an annual review of research, vol 7. University of Chicago Press, ChicagoGoogle Scholar
  38. Rhodes W (1993) Synthetic estimation applied to the prevalence of drug use. J Drug Issues 23:297–321Google Scholar
  39. Rhodes W, Layne M, Johnston P, Hozik L (2000) What America’s users spend on illegal drugs 1988–1998. Office of National Drug Control Policy, Washington, DCGoogle Scholar
  40. Riccio LJ, Flinkestein R (1985) Using police arrest data to estimate the number of burglars operating in a suburban county. J Crim Just 13:65–73CrossRefGoogle Scholar
  41. Roberts JM, Brewer DD (2006) Estimating the prevalence of male clients of prostitute women in Vancouver with a simple capture–recapture method. J Roy Stat Soc Seri A 169:1–12CrossRefGoogle Scholar
  42. Rossmo DK, Routledge R (1990) Estimating the size of criminal populations. J Quant Criminol 6:293–314Google Scholar
  43. Royal Canadian Mounted Police (2004) Drug Situation in Canada––2003. Criminal Intelligence Directorate. Royal Canadian Mounted Police, OttawaGoogle Scholar
  44. Schwarz CJ, Seber GAF (1999) Estimating animal abundance: Review III. Stat Sci 14:427–456CrossRefGoogle Scholar
  45. Seber GAF (1973) The estimation of animal abundance. Griffin, LondonGoogle Scholar
  46. Smit F, Reinking D, Reijerse M (2002) Estimating the number of people eligible for health service use. Eval Prog Plan 25:101–105CrossRefGoogle Scholar
  47. Smit F, Toet J, van der Heijden P (1997) Estimating the number of opiate users in Rotterdam using statistical models for incomplete count data In European Monitoring Centre for Drugs and Drug Addiction (EMCDDA), 1997 Methodological Pilot Study of Local Prevalence Estimates. EMCDDA, LisbonGoogle Scholar
  48. Spohn C, Holleran D (2002) The effect of imprisonment on recidivism rates of felony offenders; a focus on drug offenders. Criminology 40:329–357CrossRefGoogle Scholar
  49. van der Heijden P, Cruyff M, van Houwelingen H (2003) Estimating the size of a criminal population from police records using the truncated Poisson regression model. Stat Neerl 57:289–304CrossRefGoogle Scholar
  50. Weisheit RA (1991) The intangible rewards from crime: the case of domestic marijuana cultivation. Crime Delinquen 37:506–527CrossRefGoogle Scholar
  51. Weisheit RA (1992) Domestic marijuana: a neglected industry. Greenwood Press, New YorkGoogle Scholar
  52. Wilkins C, Battha K, Casswell S (2002a) A demand-side estimate of the financial turnover of the marijuana black market in New Zealand. Drug Alcohol Rev 21:145–151CrossRefGoogle Scholar
  53. Wilkins C, Battha K, Casswell S (2002b) The effectiveness of cannabis crop eradication operations in New Zealand. Drug Alcohol Rev 21:369–374CrossRefGoogle Scholar
  54. Wilkins C, Casswell S (2003) Organized crime in marijuana cultivation in New Zealand: An economic analysis. Contemp Drug Prob 30:757–777Google Scholar
  55. Willmer MAP (1970) Crime and information theory. University Press, EdinburghGoogle Scholar
  56. Wilson RM, Collins MF (1992) Capture–recapture estimation with samples of size one using frequency data. Biometrika 79:543–554CrossRefGoogle Scholar
  57. Zelterman D (1988) Robust estimation in truncated discrete distributions with applications to capture-recapture experiments. J Stat Plan Infer 18:225–237CrossRefGoogle Scholar

Copyright information

© Springer Science+Business Media, LLC 2007

Authors and Affiliations

  1. 1.Department of Criminology and Criminal JusticeUniversity of MarylandCollege ParkUSA

Personalised recommendations