Estimation of Cocaine Availability
1996–1999
2. The STAR Model
The STAR model incorporates various cocaine availability estimates into a cohesive, connected model. The model hinges on the notion of a transition of cocaine from one stage – estimate of drug (or drug precursor) availability, distributed within a specific geographic region – to the next. The transition is a computational link between stages that converts drug (or drug precursor) availability at one stage to availability at another stage, and includes reductions (seizures, losses, etc.). Table 2 details stages and transitions between stages (including reductions), and lists data sources utilized in STAR. Although the table presents stages in numerical order, the model is not necessarily applied sequentially from stage 1 to stage 9. For example, the model could just as easily begin at stage 9 and work back – adding in reductions – to a potential production number. Alternatively, the model could begin with event-based data 2 and work backward or forward. The important point is that the model is flexible and not bound to any specific ordering of stages.
The model is comprised of nine stages and eight transitions. Stages 1 through 4 are production stages within the growing areas, and Stages 5 through 8 track cocaine HCl from Andean labs to the streets of the U.S. and non-U.S. destinations. Figure 1 depicts the geographical areas involved in each stage.
Stage 1 begins with net coca cultivation, in each growing area, from the previous growing year. The transition to Stage 2 – net cultivation in the current year – is the net change, calculated as: the amounts from Stage 1, plus new growth, minus eradication and field abandonment. The eradication figures used in this research are the "effective" eradication figures calculated from the estimation of coca crop cultivation.
Stage 3 is net cultivation converted into net leaf amounts, via calculations performed in Transition 2/3. These calculations utilize leaf yield estimates and compensate for leaf losses, including both licit consumption and leaf seizures. Stage 4 is the amount of coca base available from the net leaf, calculated by using the leaf-to-base conversion factors assumed in Transition 3/4. Stage 5 represents cocaine availability at the HCl labs, and Transition 4/5 links cocaine base from growing regions to the HCl labs. Stage 6 describes cocaine HCl availability at South American departure points.
At Stage 7 the flow branches into two parts. The transition from 6 to 7A is the amount of cocaine exported from South America that moves toward non-U.S./LTAM markets (e.g. Europe and Canada). The transition from 6 to 7B is the amount of cocaine exported from South America that moves toward the U.S. markets.
Figure 1: Geographic Areas of STAR Stages
Stage 8 is cocaine available at U.S. border regions. Stage 8 uses the Border Allocation Model to apportion cocaine amounts to U.S. border entry regions, by conveyance types 3 . Transition 7B/8 subtracts border seizures.
Stage 9 is the amount of cocaine available for consumption in various consumption regions within the country. Transition 8/9 incorporates the Domestic Allocation Model to describe cocaine movement from border entry regions to consumption regions 4 and accounts for domestic seizures.
The STAR model applies sequential transitions through a serious of matrix operations 5 . This matrix formulation has several advantages: algebraic conciseness, ability to project assumptions at any stage on predicted flows at subsequent stages, and ability to gauge transition probabilities connecting flows, as well as flow amounts. The model was programmed using the matrix programming language of SAS/IML (SAS Institute, 1990), a program with powerful facilities for simulating alternative flow scenarios.
At most of the transitions, the matrix formulation is an accounting framework incorporating availability estimates. These "accounting transitions" simply apply available data. However, at stages 6, 8, and 9 the model is more than an accounting device. At these stages, the model affords a comparison and potential reconciliation of alternate availability estimates. Thus, at stage 6, it estimates the inconsistency in cocaine availability estimates by comparing potential production with event-based estimates of cocaine departing South America. At stage 8, it compares predicted outputs derived from potential production, event-based data, and the Border Allocation Model. At stage 9, it judges the difference in availability estimates by incorporating domestic consumption estimates 6.
STAR Stage and Transition Details
The matrix formulation allows for differing assumptions – or scenarios – to be introduced at any transition and then carried forward (or backward) for comparisons with other scenarios. Although the following discussion presents stages in sequential order, the model does not have to be applied sequentially. Section 4 details how the model was used to yield estimates for 1996-1999, while the intent of this section is to provide specific details about data, stages, and transitions. The section also discusses both the Border Allocation Model and Domestic Allocation Model.
Potential Production Versus Actual Production
This paper makes a distinction between potential and actual production. Potential cocaine production is calculated, by year, beginning with hectares under coca cultivation and then multiplying by the leaf yield, alkaloid content, and base processing efficiency figures. These figures measure availability for world consumption, assuming all coca hectares are converted to cocaine product. Actual cocaine production is calculated by using the same conversion rates, but subtracts losses that occur during the process, such as leaf spoilage, licit consumption, and base and HCl seizures.
Table 3 summarizes the stage-by-stage summary of potential production estimates for each year (see Appendix A for details). Over the period 1996-1999, potential production has decreased 50-75 metric tons per year. These figures are worst-case estimates of cocaine availability in the Andean countries because they do not account for known losses such as consumption, or leaf, base, and HCl seizures. The STAR model expands on these estimates in order to calculate the actual availability of cocaine for export from South America.
For Colombia, Peru and Bolivia, estimates of the quantity of coca under cultivation are developed by CNC, using survey methods similar to those used by agricultural organizations estimating the size of licit crops. A survey is designed using statistically-based sampling techniques, ensuring that an adequate number of samples are collected over randomly selected areas, as well as sampling of known growing regions. Selected areas are then imaged, using satellites and aerial photography. Using these images, region-specific coca crop estimates are developed.
Throughout the 1990's, Colombia was assumed to be cultivating the poorer yielding variety of cocaine, E. coca var ipadu and was using processing techniques as efficient as Bolivia and Peru. However, recent research makes it clear that Colombia is not only a major cocaine producer, but also a leading coca cultivator. This new data has been used to revise historical Colombian production estimates back to 1995. The STAR model incorporates the revised data as of March 2000.
Figure 2 depicts changes in the distribution of Andean potential production. Note that the figure includes two lines for Colombia, the lower one representing earlier Colombian estimates and the higher one representing data as of March 2000. Revision of the Colombian conversion figures caused the total potential production figures to increase by nearly 200 metric tons per year, but a downward trend still remains.
These adjustments highlight the difficulty in maintaining consistent trends during periods of dynamic changes, such as the rapid increases in Colombian cultivation. The statistical nature of the imaging process allows standard errors to be calculated, which measures a portion of the uncertainty in the cultivation estimates. However, additional uncertainty is introduced by extrapolating the cultivation figures into potential production estimates. Uncertainties include the detection of new growing areas and eradication estimation (maturity of the eradicated crop, strength of the herbicide, and timing of the harvest). The Breakthrough 7 estimates provide the crop yield data and processing efficiency data to calculate the potential production from the crop cultivation estimates. These Breakthrough estimates are refined, as updated data becomes available. All of these estimates are snapshots in time, and must therefore be periodically updated. One example of a changing trend is that there have been reports that Peru’s coca industry may be recovering 8 .
Figure 2 also shows that while production in Bolivia and Peru has dropped, Colombian production has soared. Accounting for only 25% of total coca cultivation in 1995, Colombia’s contribution grew to 68% by 1999. Applying time-series techniques to the raw data could reduce what appears to be considerable random variation from year to year.Figure 2: Potential Cocaine Production, 1990-1999 (mt)
The STAR model expands on potential production in order to calculate the actual availability of cocaine for export from South America. This is done by subtracting losses such as spoilage, licit consumption, and seizures.
Stages 1 through 4 occur in each of the eighteen Andean growing regions (Guaviare, West Caqueta, East Caqueta, Norte de Santander, San Lucus, Arauca, Putamayo, and Macarena in Colombia; Upper Huallaga Valley, Central Huallaga Valley, Lower Huallaga Valley, Aguaytia, Pachitea, Apurimac, Cusco, and other Peruvian growing areas in Peru; Chapare, and Yungas/Apolo/Other in Bolivia). From these regions, coca base moves to cocaine HCl production labs (Stage 5), through base movement corridors.
Stage 1: Net Cultivation From Previous Growing Year
The STAR model starts with the estimates of hectares under cultivation. Stage 1 simply represents the previous year’s net cultivation estimates
Stage 2: Net Cultivation in Current Year
Stage 2 represents the current year’s net cultivation in each of the eighteen growing areas. Transition 1/2 is the computational link between the previous year’s net cultivation and the current year’s net cultivation. The computation considers new growth, field abandonment, and eradication.Stage 3 is the amount of net leaf yielded from coca plants, by growing region. The transition between Stages 2 and 3 applies leaf yield factors (shown in table A1, Appendix A) to transform the amount of net cultivation into potential leaf amounts, measured in metric tons. Colombian leaf yields represent amounts of wet leaf, whereas Peruvian and Bolivian leaf yields are for dry leaf. Transition 2/3 includes reductions for licit leaf consumption (obtained from the International Narcotics Control Strategy Report (INCSR)), leaf seizures, and for leaf not harvested – which is assumed to be one percent of mature hectares.
Stage 4 is the amount of base created from net leaf amounts, by growing region. Transition 3/4 applies leaf-to-cocaine conversion factors (detailed in table A2, Appendix A) for each growing region.
Stage 5: Cocaine Availability at Labs
Stage 5 measures the amount of cocaine produced at labs. Transition 4/5 follows coca base from growing regions to labs through base corridors of movement as defined in the IACM publications (beginning in 1997). In 1997, three base corridors of movement were identified (northeast, south, and northwest) and, in 1998, a fourth was added (west). The STAR model apportions base from growing regions to labs by the percentages of observed movement in the IACM. Reductions in the transition include cocaine base seizures.
Transitions 4/5 and 5/6 must be considered tentative for several reasons. First, data on movements of base and cocaine within the source countries are incomplete. Second, data on losses due to base spoilage and source country consumption are fragmentary, imprecise, or nonexistent. Finally, Transition 4/5 assumes that base movement corridors are independent of growing areas, and Transition 5/6 assumes that HCl movement corridors are independent of lab locations. Neither assumption is realistic. Nonetheless, it is useful to begin to model these two transitions, as base and HCl movement may become more detectable in the future.
Stage 6: Cocaine Departing South America
Transition 5/6 is the link between cocaine labs and South American departure points, through HCl corridors of movement as defined in the IACM publications. The model apportions the flow of HCl from labs to departure points by the percentages of observed South American cocaine movement described in the IACM.
Reductions taken in this transition include source country seizures and spoilage (assumed to be one percent). Source country consumption is not accounted for because estimates are tentative and exist for 1999 only.
Estimates of Cocaine Departing South America Using Event-Based Data on Cocaine Movements in the Transit ZoneThe IACM uses an event-based, interagency consensus methodology to quantify cocaine movement through the transit zone. Event-based data in the Consolidated Cocaine Database (CCDB) combines two efforts: the Interagency Counterdrug Performance Assessment Workgroup (ICPAWG) and the IACM. The ICPAWG -- established in 1992 to measure the performance of international drug interdiction -- maintains a database of known drug movements in the transit zone, with a destination of either the U.S. or Canada. Known events are designated by expert participants of an interagency working group on the basis of the following information: (1) seizure or observation of drugs; (2) observation of activity that could not be reasonably attributed to anything other than drug smuggling; (3) reliable intelligence.
In 1996, the interagency group developed a cocaine flow assessment methodology to determine the amount of cocaine that departs South America along major trafficking routes 9 . Three types of uncertainty exist in the data: uncertainty in the amount of cocaine transported, uncertainty in the existence of the event, and uncertainty about how much cocaine remains undetected. For example, if the quantity of cocaine recorded in the database for movements from South America to Florida come exclusively from seizures, then one can assume with a high degree of certainty that more cocaine was moved but not detected. This type of uncertainty is important because it can be used to show that cocaine movement via commercial means is underestimated.
Table 4 includes event-based estimates of cocaine departing South America for 1996 through 1998 10 . Part of the variability from year to year in these numbers is attributable to evolving methodology. There is considerable uncertainty in the magnitude of the IACM estimates, but the stable trend in the estimate of cocaine departing South America correlates well with other supply indicators. The stable, event-based estimates indicated a disconnect with the older Colombian cultivation estimates.
Stage 7A: Non-U.S./South American Markets
Figure 3 shows the split of the flow between that moving toward U.S. markets and that moving toward non-U.S./LTAM markets (primarily Europe). Stage 7A is the amount of cocaine that departs South America and successfully arrives at non-U.S./LTAM markets. Seizures in non-U. S. bound corridors are included in the transition.
This stage is the amount of cocaine that departs South America towards the United States. Transition 6/7B apportions cocaine from South American departure points through corridors of movement, via specific conveyances (noncommercial and commercial air, noncommercial and commercial maritime). Two assumptions are made: cocaine leaving from Colombia transits all three corridors; cocaine leaving from departure points in Peru, Ecuador or Bolivia transits through Mexico/Central America (MX/CA) only. Flows among corridors and conveyances are apportioned in the same proportion as flows in the event-based data.
Figure 3: Availability at U.S.-Bound Transshipment Corridors and Non-U.S./LTAM Markets (Stage 7)
During Transition 6/7B, event-based data is incorporated, which describes cocaine departing South America by corridor and conveyance combinations. Reductions taken in the transition include transit seizures and transit country consumption, which is assumed to be three percent of the flow.
Ideally, Transition 6/7B would include conveyance combinations. In the Mexican/Central American corridor, the most prevalent combination is to use noncommercial maritime to get part of the way through the transit zone and then to use land conveyance to travel the rest of the way. There are some secondary movement events listed in the CCDB, but they were not included in STAR.
Stage 8: Cocaine Availability at U.S. Border Entry Regions
Stage 8 is the amount of cocaine that successfully passes into the U.S., by border entry regions. Figure 4 illustrates the U.S. border entry regions used in the model. Transition 7B/8 converts the amount of cocaine passing through the transit zone – by movement corridor and conveyance type – into amounts entering U.S. borders by geographic region and by conveyance type. It is assumed that shipments passing through the Mexican/Central American corridor terminate at the southwest border and that shipments in the Caribbean and Direct to U.S. corridors are distributed in proportion to border seizures and conveyance combinations.
Reductions taken during this transition account for seizures at the border using an Enhanced Seizure Database created for the STAR model. At stage 8 the Border Movement Model provides estimates of cocaine arriving to U.S. regions, by conveyance type.
Enhanced Seizure Database
To determine reproducible domestic and border seizure amounts, an Enhanced Seizure Database was created, based on a variety of seizure databases. DEA’s Federal Drug Seizure System (FDSS) for calendar years 1991-1998 provided the bulk of the data for this effort. FDSS data contain no duplicate records – each seizure in the FDSS is uniquely identified by a Federal Drug Identifying Number (FDIN), eliminating the risk of double counting. The FDSS includes federal and federally-supported cocaine seizures of 500 grams or more. The Enhanced Seizure database only includes those FDSS seizures that were above the threshold set by the FDSS system.
FDSS contains limited details about each seizure, so the FDSS data was augmented with agency-specific seizure data. Customs seizure data includes country of origin and more detailed information about conveyance. Other supplementary data came from the Coast Guard, the El Paso Intelligence Center (EPIC) Border/Land Interdiction Seizure System (BLISS), and the CCDB. The EPIC data covers seizure events occurring at the United States/Mexican border and up to 150 miles inside the United States. Appendix D details specific variables from each of these data sources. The FDSS data was used as the "master" when conflicting data appeared across databases. The exception to this is that EPIC data are employed for southwest border seizures.
Border SeizuresSeizures at the border (arriving from foreign countries) were classified by conveyance types (noncommercial and commercial air, noncommercial and commercial maritime, noncommercial and commercial vehicle, rail and pedestrians) and geographic region (Florida, Gulf Coast, Northeast, Southwest Border, Puerto Rico/Virgin Islands, and Rest of U.S. – including Ports of Entry (POE) along the Canadian border)11.
EPIC has traditionally accounted for border seizures. There is a definitional difference in seizures at the southwest border and at all other border areas. EPIC’s definition of a southwest border extends 150 miles into the U.S., since the drugs likely came from Mexico. In Florida, by contrast, the border does not extend inland, although it would seem just as plausible that the drugs came across the Florida border. This issue points to the need for a consistent definition of a border seizure.
To identify a border seizure, and to classify it by conveyance type and by geographic region:
1. Seizures on the high seas were excluded from FDSS data because they are included in transit zone seizures.
2. To identify seizures along the southwest border, information from EPIC was used. Any car, four-wheel drive, motorcycle, pickup truck, recreational vehicle, towed vehicle, or van was classified as a noncommercial vehicle. Additionally, if the "type" variable indicated "intrusion by vehicle at border (not POE)" or "vehicle at POE" the conveyance was classified as a noncommercial vehicle. Conveyance was assigned as commercial vehicle for tanker truck, bus, tractor trailer, trailer, or wrecker. If the type variable indicated "on foot at border" or "pedestrian at POE", then the conveyance was assigned as pedestrian. And finally, if conveyance type was train, the seizure was assigned to the rail conveyance category.
3. To categorize maritime border seizures, Customs information was checked, specifically for whether the conveyance arrived from non-U.S. locations. If so, and if the conveyance was listed as a commercial vessel, then commercial maritime was assigned. If conveyance was listed as a fishing or private vessel, then noncommercial maritime was assigned. Coast Guard and CCDB information was used to identify maritime seizures that occurred outside of ports of entry.
4. To categorize border seizures from air conveyances, Customs information was checked to determine if the conveyance arrived from non-U.S. locations. If so, and if the conveyance was listed as commercial air, mail, or express consignment, then commercial air was designated as the conveyance type. If conveyance was listed as private aircraft, then noncommercial air was designated. CCDB data were consulted for air conveyance seizures.
5. Finally, 113 border seizures that were classified by Customs as "other" or "no transport involved" were examined individually, to determine if they were border seizures.
Figure 5 presents a plot of total border seizures for the years 1991-1999. The figure shows that overall, border seizures have decreased 31.5%, from 61.9 metric tons in 1991 to 42.4 metric tons in 1999. The chart also includes a two-year moving average line to smooth year-to-year variations. Table5 details border seizures by conveyance types.
In Figure 6, smoothed seizure (three-year moving average for southwest border and two-year moving average for all other areas 12 ) figures are plotted by region, for the period 1992 through 1999. Seizures on the southwest border (the solid line at the top of the figure) have remained relatively constant over the period. Seizures in Florida (the dotted line at the top of the chart) have declined over the period, while seizures in Puerto Rico/Virgin Islands have steadily increased.
(export quality metric tons)
(export quality metric tons)
The Border Allocation Model was developed to allocate the cocaine entering the U.S.(Stage 8) among the border entry regions. In particular, the model predicts the percentage of cocaine arriving at specific regions, by specific conveyance types. Cocaine amounts are then obtained by multiplying the percentages by the estimated total. The proportions can be employed in the allocation of amounts based on any estimate of the amount of cocaine arriving to the U.S. For example, using percentages generated by the Border Allocation Model, cocaine amounts estimated via event-based data can be allocated to specific U.S. border regions and conveyances (after subtracting transit zone seizures and consumption). Any amount that the STAR model incorporates (including potential production estimates) can be distributed into conveyance/border region combinations. Appendix E provides details about the methodology.
The Border Allocation Model uses data on U.S. border seizures and on the costs smugglers pay to transport cocaine from Colombia to the U.S. Data on U.S. border seizures were obtained from the Enhanced Seizure Database, and data pertaining to smuggler transportation costs were obtained from Customs Reports of Investigation.
Tables 6 and Table 7 show the average number of metric tons seized, and the percentage of the total amount seized, for each conveyance and border region combination. Note that seizures from land conveyances in Florida, Gulf Coast, Northeast, and Puerto Rico/Virgin Islands (PR/VI) are impossible and these region-conveyance combinations therefore contain structural zeros. This contrasts with observed zeros (such as that obtained for Gulf Coast, commercial air) where the region-conveyance combination is feasible, but no occurrences were observed.
Averaging over the eight year period, 45% of total seizures occurred at the southwest border (SWB) and 34% at the Florida border. In terms of conveyances, 31% of the seizures occurred upon commercial marine ships, while noncommercial vehicle, noncommercial marine, commercial vehicle, commercial air, and noncommercial air accounted for 28%, 16%, 12%, 10%, and 4%, respectively.
Table 8 shows how the Border Allocation Model allocates the total cocaine quantity arriving at U.S. borders to specific border regions and conveyance types. The model predicts that – averaged over the years 1991-1998 – 48% of cocaine destined for the U.S. arrives at Florida via commercial marine conveyances and 37% arrives at the southwest border via commercial and noncommercial vehicles. Note that the distribution of cocaine amounts (Table 8) differs considerably from the distribution of cocaine seizures (Table 7). This is because estimates of cocaine amounts are not simply proportional to seizures. For example, even though cocaine seizures for Florida via commercial marine are only 21% of total seizures, the proportion of the total amount transported through this region-conveyance combination is 48%. This occurs because transportation costs were relatively high in this case ($3,568 compared to the mean of $3,111), which, assuming constant total transportation costs, implies that the probability of seizure, and therefore seizure costs, were relatively low. Thus the amount seized was a relatively low percentage of the amount shipped to Florida via commercial marine.
Figure 7 plots the amount of cocaine arriving at each border region for the period 1991-1998. The model indicates that most cocaine entering the U.S. does so via Florida and the southwest border. Taking the eight-year period as a whole, quantities arriving at the southwest border have increased at the expense of quantities arriving at Florida. All other regions have remained fairly constant, with the exception of PR/VI, for which the model predicted a jump from 11 metric tons in 1996 to 42 metric tons in 1997.
Figure 8 plots model estimates by conveyance type. Conveyance types of choice appear to be commercial vehicle and commercial marine. Although it is likely that noncommercial air actually plays a large role in transporting cocaine, the model does not capture this because the typical flight stops just short of the U.S.- Mexican border. Figure 8 shows that, over the eight year period, conveyance by commercial vehicle has increased at the expense of conveyance by commercial marine: commercial vehicle increased by 78% (from 91 to 162 metric tons) and commercial marine decreased by 29% (from 286 to 203 metric tons). These estimates are consistent with Colombian drug lords allowing Mexico-based trafficking organizations to play an increasing role in shipping cocaine to the U.S. Indeed, taking Figures 7 and 8 together, it would appear that there has been a shift in smuggling from Florida via commercial marine to the southwest border, via commercial vehicle. Appendix C (Table C6) presents detailed estimates for each year.
Results of the Border Allocation Model indicate a higher proportion of cocaine flow to the Florida destination than current intelligence assessments. The results of the Border Allocation Model should be seen as developmental and not a conclusive result. But the model does provide an interesting perspective. The current intelligence assessment consistently underestimates smuggling via commercial conveyances, which would probably be the primary means of smuggling into the Florida corridor. Further research is needed in the critical border region to determine the more correct estimate.
Cocaine at this stage represents the amount of cocaine arriving to U.S. consumption regions from U.S. border entry regions. Figure 9 depicts domestic retail markets, which have been broken down into ten main regions. Transition 8/9 incorporates domestic (non-border) cocaine seizures. The arrows in the figure depict routes taken from border entry regions, based on the results of the Domestic Allocation Model.
The Enhanced Seizure Database was also used to quantify domestic seizures within the United States. Table 9 shows the annual domestic seizures allocated by census regions.
To allocate cocaine entering the U.S. to consumption regions, the Domestic Allocation Model was created. The premise of the model is consistent with the classic operations research transportation problem: given the quantities of cocaine entering the domestic market at the six border regions, and given the quantities demanded in each of the ten U.S. census divisions, it is assumed traffickers determine the allocation that satisfies demand in all divisions while minimizing total transportation costs. Standard linear programming techniques were used to solve this problem. Appendix E provides details of the model.
Table 10 shows, for each border entry region, the percentage of cocaine moved to each consumption region in 1998 (values for other years are shown in Appendix C, Table C6). Taking these estimates at face value, one could conclude that cocaine smuggled in at the Gulf Coast, Northeast, and Rest of U.S. stays in that general area, while shipments through Florida, Puerto Rico and the southwest border go to other regions. In particular, 90% of the southwest border’s imported cocaine is distributed to areas beyond the southwest border, reflecting the increased role of Mexico-based traffickers 13.
Notes
2 Event-based data are derived from a database of drug movements in the transit zone. These data are described in detail later in this section.
3 An overview of the model appears later in this section, and technical details are provided in Appendix D.4 An overview of the model appears later in this section, and technical details are provided in Appendix F.
5 At each of the eight stages, there is a transition matrix that transforms the input into the predicted output. At stage 1, v1 = v0 *M1, where "*" denotes matrix multiplication. At stage 2, v2 = v1 *M2. At stage 3, v3 = v2 *M3, and so on. The complete model can be written
v8 = v0 *M = v0 *M1 *M2 *M3 *M4 *M5 *M6 *M7 *M8 ,
where v0 denotes gross hectares by growing area and v8 denotes cocaine consumed by U.S. geographical subarea.
6 W. Rhodes, M. Layne, P. Johnston, L. Hozik, What America’s Users Spend on Illegal Drugs, 1988-1998, June 2000.
7 Drug Enforcement Administration, 1994, Operation Breakthrough: Coca Cultivation and Cocaine Production in Bolivia. Drug Enforcement Administration, 1997, Operation Breakthrough: Coca Cultivation and Cocaine Production in Peru.
8 Defense Intelligence Agency, 1999. Interagency Assessment of Cocaine Movement: August 1999 Eighteenth Edition, Mid-Year Review, p. 2.
9 The results are included in the transit zone section of the IACM publications.
10 Movement events from the CCDB were used for the calculations, and they differ slightly from figures published in the IACM. See Cala, 1999.
11 Our border seizures figures differ from those reported by EPIC in the IACM. A description of their methodology was unavailable.
12 A two-year moving average for the southwest border still yielded considerable variation from year to year.
13 Drug Enforcement Administration, August 1997, Changing Dynamics of the U.S. Cocaine Trade.
