JEconGeogr 2014 McDougal Jeg Lbu021 PDF

See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/265127678
The Way of the Gun: Estimating Firearms Trafficking across the US–Mexico
Border
Article in Journal of Economic Geography · June 2014

DOI: 10.1093/jeg/lbu021
CITATIONS READS
20 985
4 authors, including:
Topher Mcdougal Robert Muggah

University of San Diego Igarapé Institute
46 PUBLICATIONS 55 CITATIONS 102 PUBLICATIONS 1,397 CITATIONS
SEE PROFILE SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Cost of Conflict in Nigeria View project
Political Economy of Rural-Urban Conflict View project
All content following this page was uploaded by Topher Mcdougal on 04 November 2014.
The user has requested enhancement of the downloaded file.

Journal of Economic Geography Advance Access published June 13, 2014
Journal of Economic Geography (2014) pp. 1–31 doi:10.1093/jeg/lbu021
The Way of the Gun: Estimating Firearms Trafficking

across the US–Mexico Border
y
Topher L. McDougal*,**, , David A. Shirk*, Robert Muggah*,**,*** and
John H. Patterson*
*Joan B. Kroc School of Peace Studies, 5998 Alcalá Park, San Diego, CA 92110, USA
**Centre on Conflict, Development, & Peacebuilding, Graduate Institute of International and Development
Studies, PO Box 136, 1211 Geneva 21, Switzerland
***Instituto de Relações Internacionais, Rua Marquês de São Vicente, 225 - Vila dos Diretórios, Casa 20,
Gávea - Rio de Janeiro - RJ, Brazil
y
Corresponding author: Topher L. McDougal, Joan B. Kroc School of Peace Studies, 5998 Alcalá Park, San
Diego, CA 92110, USA. email 5tlm@sandiego.edu4
Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Abstract
The volume of firearms sold in USA and trafficked across the US–Mexico border is
notoriously difficult to estimate. We consider a unique approach using GIS-generated
county-level panel data (1993–1999 and 2010–2012) of Federal Firearms Licenses to
sell small arms (FFLs) to estimate the realized demand for firearms based on the
distance by road from the nearest point on the US–Mexico border. We use a time-
series negative binomial model paired with a post-estimation population attributable
fraction (PAF) estimator. We do so to control determinants of domestic demand. We
are able to estimate a total demand for trafficking, both in terms of firearms and dollar
sales for the firearms industry. We find that nearly 2.2% (between 0.9% and 3.7%)
of US domestic arms sales are attributable to the US–Mexico traffic in the period
2010–2012, representing 212,887 firearms (between 89,816 and 359,205)
purchased annually to be trafficked.
Keywords: Firearms, trafficking, demand, North America, US–Mexico

JEL classifications: D74, F14, F52, K14, K42
Date submitted: 26 June 2013 Date accepted: 7 May 2014
1. Introduction
The total volume of firearms traffic from USA to Mexico has both intrigued and
frustrated those seeking to quantify and lower it. Previous efforts to address the issue
have usually focused only on the quantity of arms confiscated at the border, begging the
question of what fraction of the total they represent. We tackle the problem differently:
we create a unique GIS-generated county-level panel dataset (1993–1999 and 2010–
2012) of Federal Firearms Licenses to sell small arms (FFLs). We then model total
realized demand for firearms, in which the number of licensed gun retailers in each
county represents quantity, and the distance by road from each US county to the nearest
point on the US–Mexico border represents a price signal (in terms of fuel and time) to
potential traffickers. We use a time-series negative binomial model and control for
determinants of domestic demand (e.g. income, political leaning, population density and
spatial autocorrelation). Employing a post-estimation population attributable fraction
ß The Author (2014). Published by Oxford University Press. All rights reserved. For Permissions, please email: journals.permissions@oup.com
2 of 31 . McDougal et al.
(PAF) estimator, we are then able to estimate a total demand for trafficked weapons,
both in terms of firearms and dollar sales for the firearms industry.
The remainder of the article is structured as follows: Section 2 provides a background
on the issues surrounding firearms trafficking, including US legislation, the rise of
violence in Mexico, and existing studies of the issue; Section 3 lays out our empirical
strategy and basic methodology; Section 4 describes our results, including total volume
estimates; Section 5 concludes with a discussion of methodological considerations.
2. Background
As US legislators ponder the problem of gun violence associated with horrific shootings
that have occurred throughout the country in recent years, such violence has taken an
even greater toll in Mexico. A recent study found that there were approximately 120,000
homicides in Mexico from 2007 through 2012, and an estimated 60,000 of these were

organized-crime-style homicides that frequently involved the use of high-powered
firearms imported illegally from USA (Molzahn et al., 2013, 13).
Although there is some debate over the proportion of US guns smuggled into
Mexico, there is no doubt that USA is the world’s largest manufacturer and supplier of
firearms, and has a highly permissive regulatory environment for the production, sale,
purchase and ownership of firearms. According to the 2004 national firearms survey
conducted by Hepburn et al. (2007), there are an estimated 218 million privately owned
firearms in USA. One in four US citizens (26%) and two in five households (38%)
owned a firearm, nearly half of all individual gun owners (48%) possessed four or more
firearms, and 20% of all individual gun owners held 65% of all firearms.
The Federal Assault Weapons Ban (AWB), or Public Safety and Recreational
Firearms use Protection Act, was adopted as part of the Violent Crime Control and
Law Enforcement Act of 1994. The 10-year ban went into effect on 13 September 1994,
and was not renewed upon its expiration in 2004. As Figure 1 illustrates, the period
during which the AWB was in effect was characterized by low levels of firearms’
manufacture and import, relative to preceding and subsequent years.1 This period also
saw the dramatic decline in FFLs, though FFL numbers did not experience a rebound
in subsequent years, implying that since 2004, a growing number of firearms are in the
US market, sold by a dwindling number of retailers. Effects of the AWB were not
limited to USA. Chicoine (2011) has estimated that the expiration of the AWB is
responsible for a rise of at least 16.4% in the homicide rate in Mexico during 2004–2008
period.
Numerous reasons are cited for the dramatic decline in FFLs, most of them having to
do with new federal and state legislation and regulations. These include the Brady
Handgun Violence Prevention Act of 1993, the AWB, the Violent Crime Control and
Law Enforcement Act of 1994, and the launch of the FBI’s National Instant Criminal
Background Check System (NICS) in 1998 as mandated by the Brady Act and requiring
retailers to determine the eligibility of prospective customers to purchase a firearm
(GAO, 1996; Marks, 2006; Vernick et al., 2006). Each of these developments imposed
certain reporting and overhead costs on FFLs. It is likely that the passage of the Brady
1 It should be noted, however, that the anticipation of the ban taking effect caused a run on military-style
weapons, causing prices and production to rise in the short-run (Koper and Roth, 2002).
Firearms trafficking across the US–Mexico border . 3 of 31
14000000 300000
12000000
250000
Number of Firearms Retail Licenses

10000000
200000
Number of Firearms
8000000
150000
6000000
100000
4000000
50000
2000000

0 0
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
Year
Type 1 FFLs Manufactures Federal Assault
Imports Exports
Figure 1. Number of firearms manufactured, imported, exported and available for sale
(primary vertical axis) and federal retail licenses (secondary vertical axis), 1986–2012. Source:
ATF (2013); graph by the authors.
Act in 1993 heightened uncertainty among gun owners that they would be able to
purchase the types or quantities of firearms they wanted in the future, thereby
stimulating demand and causing the spike in purchases we observed in that year.
The US domestic firearms market is also large by dollar volume of sales. Figure 2
illustrates an imputed dollar volume of the market from 1983 to 2012, based on excise
taxes collected by the Internal Revenue Service (IRS) and the Bureau of Alcohol,
Tobacco, Firearms, and Explosives (ATF). Before the AWB started acting, the annual
volume spiked to over US$ 2.3 billion in 1993. We project, based on sales volume, that
revenues were over US$ 5.3 billion in 2012, the last year of the data reported in Figure 1.
The large number of guns available in USA is facilitated by a widespread availability
of retail firearm licenses, permissive import regulations, and largely unregulated, owner-
to-owner ‘kitchen-table’ sales. This makes many different types of deadly firearms,
ammunition and explosives legally available for purchase in USA. Moreover, the fact
that only 5% of the roughly 54,000 registered gun dealers2 in USA are inspected
annually (Marks, 2006), which suggests that there are few checks against illicit practices
such as sales to ‘straw purchasers’, who buy guns for the purpose of illegal resale or
trafficking (Violence Policy Center, 2009). Part of the difficulty in achieving more
2 Estimates for the total number of gun dealers in USA vary, but by all accounts have declined dramatically
over the last decade—from approximately 245,000 to 54,000 dealers—thanks to tougher enforcement by
ATF and tighter gun regulations (e.g. the Brady Handgun Violence Protection Act of 1993) (Marks, 2006;
Vernick et al., 2006).
$70,00,000
Total Sales (US$1,000)
$60,00,000
Total Excise (Current US$1,000)
$50,00,000
US$ (Thousands)
$40,00,000
$30,00,000
$20,00,000
$10,00,000

$0
2001*
2003*
2005*
2007*
2009*
2011*
1987
1989
1991
1995
1983
1985
1993
1997
1999
Year
Figure 2. Domestic revenue and tax excise volumes of the US firearms industry in thousands
of US dollars, 1983–2012. Source: ATF (2013) and the authors; graph by the authors. Excise
tax rates are 10% on pistols and revolvers, 11% on other firearms and ammunition.
Disaggregated volumes of excise volumes for different types of sales are based on 1996 relative
percentages, as ATF stopped collecting those statistics that year. Until FY 1990, the Internal
Revenue Service collected excise taxes. ATF assumed the collection in FY 1991. Dollar volume
is imputed for years 2001–2012 based on volume sold, past dollar volumes reported and
inflation rates.
effective firearm regulations is that powerful US gun lobbies have hamstrung efforts to
enforce existing laws or otherwise regulate access to deadly, high-powered weapons in
USA (Violence Policy Center, 2010, 2011).
Mexico, in contrast, maintains significant restrictions on firearms. While the Mexican
Constitution guarantees right of Mexicans to carry arms, Mexico’s Federal Firearms
and Explosives Law and other supporting legislation strictly regulate the possession and
sales of firearms, ammunition and explosives.3 Current restrictions on firearms in
Mexico date back to the period of civil unrest in the late-1960s, which led to a tightening
of regulations, which included the closure of retail firearm dealers and federal
requirements for the registration of all firearms. While law enforcement and military
personnel are permitted to use firearms by Article 160 of the country’s federal criminal
code, the Mexican army (SEDENA) oversees the sale of all firearms to private
individuals through the Federal Arms Registry. Person-to-person firearm sales are
3 Article 10 of the Mexican Constitution, amended in 1971, states that Mexicans have a ‘‘right to possess
arms in their homes for their security and legitimate defense with the exception of those prohibited by
federal law and those reserved for the exclusive use of the Army, Navy, Air Force, and National Guard.
Federal law shall determine the cases, conditions and place in which the inhabitants may be authorized to
bear arms.’’
35,000
Mexico 32,332
30,000
U.S. 29,824
25,000
Firearms Seizures
20,000
15,000
10,500
10,000
9,553
4,976
5,000
2,900
-
2005 2006 2007 2008 2009
Year

Figure 3. Firearms seizures by US and Mexican authorities, 2005–2009. Source: Goodman
and Marizco (2010), 170, 175–6; data on US seizures are approximate for 2007, and missing for
2006 and 2008. Data for Mexico seizures are missing for 2006, due to unreliable data.
prohibited by Article 164 of the criminal code. There are also strict penalties under
Article 162 of the federal criminal code for ordinary citizens who possess or carry
firearms without authorization.4 There are also restrictions on the caliber of firearms
that ordinary citizens may possess, which is limited to 0.380 or less (0.357 magnum and
9 mm parabellum are also prohibited), and in practice the lawful possession of firearms
above 0.22 caliber is limited by the difficulty of obtaining permits to do so. Hunters and
target shooters may obtain licenses for firearm possession, and gun collecting is allowed
with some exceptions. As a result of these strict controls, there were only about 4300
legally registered firearms in private possession in the country in 2007 (Grillo, 2007).
On the other hand, despite these strict regulations, there is a large volume of illegal
firearms in Mexico. In fact, in 2012, the Mexican military estimated that only one-third
of 1% of firearms in Mexico is legally registered, and 90% of all firearms are used for
criminal purposes.5 The Mexican federal government has seized a large number and
wide array of arms in recent years (see Figure 3), with dramatic growth in the
proportion of high-powered weapons (including 9-mm pistols, 0.38-caliber ‘super’
pistols also known as cop killers, 0.45-caliber pistols, and AR-15 and AK-47-type
assault rifles, grenades and bazookas).6 The jump in seizures from 2007 to 2008 is likely
4 In December 1998, the Mexican Congress enacted legislation relaxing unintentional violations involving a
single gun (by tourists): first-timers are fined US$1000, but not imprisoned. Previously, they would face
5–30 years in jail. However, these relaxed standards do not apply for military weapons or calibers.
5 Isaac M. Reyes Maza, ‘‘Con licencia de Sedena, 1% de las armas en el paı́s,’’ Quadratıń. Available online
at:http://www.quadratin.com.mx/Noticias/Con-licencia-de-Sedena-1-de-las-armas-en-el-pais
6 From January to June 2008, 7645 weapons were seized, compared to only 3801 during the same period for
the previous year. Of note was the increase in the proportion of ‘long guns’ (armas largas, or rifles and
shotguns), which went from 1877 or 49.3% of weapons seized in the first half of 2007 to 4465 or 58.4% of
weapons seized in the first half of 2008. President Felipe Calderón, Segundo Informe de Gobierno, 1
September 2008, p. 24. The rise of long guns is one of the aspects of the increasing militarization of the
US firearms market, as well as that portion of the stock trafficked into Mexico for use by cartels (Violence
Policy Center, 2009, 2011).
Imported to the United States
Recovered by US Border
Mexican Gov’t
Recovered by
Patrol/ Police
U.S.-Purchased
U S
to Traffic
to Mexico
Manufactured in the United States

Exported from the United States
Figure 4. Venn diagram depicting the relationship of total firearms purchased in USA with
various sub- and super-set of firearms. For illustrative purposes only; box sizes not
proportional to actual volumes.
due to increased scrutiny of firearms trafficking by the Calderón Administration, as

well as more frequent raids on Drug Trafficking Organizations (DTOs) that may be in
possession of illegal firearms (Goodman and Marizco, 2010, 175).
US and Mexican authorities recognize that a large quantity of firearms, ammunition
and explosives sold legally in USA are trafficked illegally into Mexico, primarily
through overland smuggling routes across the 2000-mile US–Mexico border (Goodman
and Marizco, 2010; Violence Policy Center, 2009). ATF efforts to trace firearms
provided in Mexico have consistently found that an overwhelming proportion of
firearms—as high as 90%—came to Mexico from USA (Serrano, 2008).7 A 2007 ATF
trace of firearms confiscated in Mexico found that 1805 (73.5%) of 2455 firearms came
from three of the four US border states: Arizona, California and Texas (Marks, 2006).
The accessibility of firearms in the border region is facilitated by the existence of an
estimated 6700 FFLs in the border region, which represent more than 12.5% of all
registered gun dealers in the country.
The total volume of firearms traffic from USA to Mexico has both intrigued and
frustrated those seeking to quantify and lower it. Two types of figures are routinely
examined with hopes of throwing light on the trade’s size: seizures by US authorities of
firearms bought or transported with the intention of being trafficked across the border;
and weapons seized by Mexican police and subsequetly traced back to a purchase
occurring in USA (Goodman and Marizco, 2010). Figure 4 is a conceptual depiction of
the relationship between various subsets of firearms. As data presented in Figure 1, the
total volume of firearms for sale in the domestic market is taken to be total
7 In May 2010, the Mexican Government claimed that around 80% of the firearms it had seized in the
preceding 3 years had come from USA (Goodman and Marizco, 2010, 170).
manufactured plus total imported minus total exported (M þ I X). Among these,
some quantity is purchased with the intention to traffic. Of that number, some subset of
arms is recovered by US law enforcement. Another, mutually exclusive, subset of
trafficked arms is recovered by Mexican authorities, along with firearms that were
acquired outside of USA.
Based on US prosecutions alone, 4976 firearms were bought with the intention to
traffic them to Mexico in FY 2009. This number corresponds to the proportion of US-
purchased firearms (represented by gray box in Figure 4) covered by horizontal stripes,
and is much lower than the Mexican government’s reported number of seizures for the
same year of 32,332 (Goodman and Marizco, 2010, 170, 176), which corresponds to the
entirety of the diagonally striped box. The Mexican Government claims that of
the approximately 75,000 firearms it seized from 2007 to 2010, roughly 80% were from
USA (Goodman and Marizco, 2010, 170). Those 20,000 or so firearms per year
correspond to the portion of overlap between the red and green boxes in Figure 4.

Few explicit attempts exist to estimate the total volume of arms purchased for
trafficking. Goodman and Marizco (2010) have perhaps conducted the most compre-
hensive study on the US–Mexico firearms traffic, but even they limit their research to
the types of firearms most usually trafficked, how DTOs are using those firearms, the
principal methods for buying and transporting the weapons, etc. Any attempt to
estimate the total traffic from firearms seizures will suffer from the obvious
disadvantage of relying upon educated guesses of the percentage of total traffic that
the confiscations represent. Therefore, we have chosen to use FFLs to retail firearms as
a proxy indicator for demand, and attempt to isolate the effect of purchasers intending
to traffic firearms across the US–Mexico border on that demand.
3. Empirical Specification
3.1. Data
We begin by noting the possibility that FFLs may tend to cluster at the US–Mexico
border in numbers disproportionate to the local US population they serve. Estimates
for the number of FFLs along the border vary, probably not only due to fluctuations in
FFL numbers, but also due to different definitions of the ‘border region’. In September
2007, Grayson estimated the number to be 1200. In January 2008, Mexican
Ambassador Arturo Sarukhán criticized the availability of weapons along the border:
‘Between Texas and Arizona alone, you’ve got 12,000 gun shops along that border with
Mexico’. The figure 6700—or three dealers for every mile of border—has been used in
reports about the number of gun shops along the border (see e.g. Corchado and
Connelly, 2008; Serrano, 2008).
We take the presence or absence of a firearms retail outlet as a proxy for the degree of
success of a local business, and therefore a proxy for realized demand. Such a method is
supported by Della Vigna and La Ferrara (2010), who similarly propose to employ a
measure of business success as a proxy for illegal arms trades (though they take as their
indicator stock prices of arms companies, and not the presence or absence of retail
licenses). Ideally, the dependent variable would be the total dollar volume of firearms
sales in each county-year. That information is unfortunately neither collected nor
reported by the ATF. Our next best option would be sales in terms of numbers of
firearms by county-year—also neither collected nor reported. The third best option
might be a count of instant background checks sent to the FBI’s National Instant
Criminal Background Check System (NICS) by FFLs to comply with US federal law
(the Gun Control Act of 1968 in particular, which bars certain categories of people—
convicted criminals most obviously—of purchasing a gun). There are at least three
drawbacks to this measure: (i) one background check is performed per person
purchasing, rather than per weapon sold; (ii) background checks are not required for
person-to-person sales or at gun shows; (iii) we have been unable to obtain those
records, and as to the best of the authors’ knowledge, use of those data is restricted to
agencies authorized by the FBI. Our best available option then was to use FFLs
operating per county-year. There are precedents for using these data: Wiebe et al. (2009)
used the 1993–1999 portion of this panel as a proxy for supply-side ‘access to guns’,
which they use as an independent variable to predict homicides in USA.
All US–Mexico traffickers incur the risk of jail time and punitive fines for attempting
to cross the border, and so we disregard those costs entirely. Distance from the border

varies drastically by US county, however, and presents two principal costs to traffickers
in the form of fuel and time. Therefore, distance from the border may be regarded as
giving price signals to potential firearms purchasers intending to traffic the goods across
the border. The negative relationship between those price signals and the quantity of
FFLs in any given US county is used to estimate the total realized demand for US-
retailed firearms sold to buyers with the intent to traffic south of the US–Mexico border.8
In our analysis, we employ decile distances rather than absolute measures of distance
for three reasons: (i) decile distance, as an ordinal categorical variable, can be interacted
with Period easily, whereas a continuous variable would produce an impossibly large,
mostly empty matrix; (ii) as the decile distance, unlike a continuous distance measure,
enters the regression equation as a categorical variable (in effect, a series of 10 dummy
variables), there is no guarantee that the adjusted predictions will produce the expected
downward slope at all; (iii) employing decile distance makes the PAF estimation much
simpler; if absolute distance measures were used, the adjusted predictions curve would
in essence have to be weighted by the number of counties at each distance value. Using
decile distance, however, the number of counties in each category is by definition equal,
and so weighting is rendered unnecessary.
We have data covering two separate, contiguous time periods: 1993–1999 and 2010–
2012. We include in the outcome variable, FFLs, both ‘Type 1’ (firearms dealer) and
‘Type 2’ (pawnbroker) licenses issued by ATF. Following Wiebe et al. (2009), we
exclude all other licenses. This is primarily because other types of FFLs do not allow for
retail of firearms, rather their manufacturing, import, collection, etc.
Figure 5 shows the number of FFLs by type over a 35-year period. Striking is the
precipitous decline in dealer (Type 1) licenses, and the concomitant rise in collector
(Type 3) licenses. Most of others are relatively small in number.
Of course, in order to isolate the effect of distance from the US–Mexico border on
level of economic demand, we must first control the local (and other) demand for
firearms. We do this with a series of control variables listed in Table 1. The controls
8 The likelihood of being caught by the US Customs and Border Protection, Immigration and Customs
Enforcement, and state border patrol forces probably decreases with distance from the border. Those
agencies do not enforce firearms purchasing laws or audit FFLs. As we are assuming that all traffickers
are passing overland, heightened security and road blocks in this zone may effectively increase the cost of
travel near the border further incentivizing FFL clustering.
300000
250000
Type 1: Dealer
200000 Type 2: Pawnbroker
Type 3: Collector
150000
Type 4: Ammunion
Manufacturer
100000
Type 5: Firearms

Manufacturer
Type 6: Importer
50000
0
1977
1985
1991
1997
2005
2011
1975
1979
1981
1983
1987
1989
1993
1995
1999
2001
2003
2007
2009
Figure 5. Federal Firearms Licenses (FFLs) by type, 1975–2010. Source: ATF (2013); graph
by the authors.
include the most fundamental determinants of domestic demand: total population,

population density, mean income and political voting record of each county i in year t.9
DCB represents the distance from the Canadian border, and might be used to estimate
the overland economic demand for firearms from Canada.
We also split our data into four periods, manifest in the variable Period. Period is
defined as a 1 for pre-AWB 1993, and the deals with the remaining 12 years by even
splits: 2 ¼ 1994–1996, 3 ¼ 1997–1999 and 4 ¼ 2010–2012. The reason behind introducing
the Period variable is to allow us to interact time with the main predictor, decile distance
from the US–Mexico border, thereby revealing how the realized demand from south of
the border has changed over time. We decided to split the years up into 3-year periods
because our 2010–2012 data already lent itself to that length of time, and because it
allowed us to single out 1993, which was unusual in many ways due to the spike in
demand discussed above, and the fact that it preceded the tighter regulations that
characterized 1994–1999.
Other control variables, of particular interest, attempt to control for the so-called
‘gun culture’ in USA. The first such control is simply longitude, as the farther west the
county is, the more deeply the Wild West attitude may be entrenched. The second
9 We did consider creating another dependent variable that would be FFL count per population, but there
are some serious drawbacks to such an approach. First, the result would no longer be a strict count
outcome, and therefore would not lend itself defensibly to analysis using a negative binomial estimator.
Second, the output of a model based on a normalized dependent variable would be less amenable to our
post-estimation Population Attributable Fraction (PAF) analysis (see Subsection 3.3).
Table 1. Control variables
Variable Description Source
Period Time period (1 ¼ 1993, 2 ¼ 1994–1996, 3 ¼ 1997–1999, and

4 ¼ 2010–2012
Statehood Year of state entering the union
Lon Longitude GIS generation
Pop Population in county i in year t US Census Bureau
FFLiDist Control for Spatial Autocorrelation
DCB Distance to the Canadian border by road from county i (km) GIS generation
GasDef Inflation-adjusted gasoline prices in that US state in year t EIA (2013)
Areakm2 Area of county i (km2) GIS generation
PopDens Population density of county i in year t US Census Bureau/
GIS generation
PopIDist The inverse distance-weighted measure of all other popula-

tions in counties other than i
DemPer Percentage of general election vote in county i going to the http://uselectionatlas.
Democratic presidential candidate org/
RepPer Percentage of general election vote in county i going to the http://uselectionatlas.
Republican presidential candidate org/
DemXRep Interaction: Democratic and Republican Voting Percentages
PersInc Median personal income in county i in year t US Bureau
DistBIA Distance by road from county i to the nearest Bureau of GIS generation
Indian Affairs lands
DistBLM Distance by road from county i to the nearest Bureau of Land GIS generation
Management lands
DistBOR Distance by road from county i to the nearest Bureau of GIS generation
Reclamation lands
DistDOD Distance by road from county i to the nearest Department of GIS generation
Defense lands
DistFS Distance by road from county i to the nearest U.S. Forest GIS generation
Service lands
DistFWS Distance by road from county i to the nearest Fish and GIS generation
Wildlife Service lands
DistNPS Distance by road from county i to the nearest National Park GIS generation
Service lands
DistMexX Distance from Border Patrol stations along the US–Mexico GIS generation
border
NAFTAYR Years since NAFTA went into effect
CrimdIndx Total Crime Index National Archive of
Criminal Justice
Data
control is the year that the state was granted statehood by the federal government, as
such statehood status came along with some minimum level of law enforcement from
the federal government.
FFLiDist represents a control for spatial autocorrelation. Spatial autocorrelation
statistics measure the degree of geospatial dependency among observations (in this case,
counties). The basic intuition here is that FFLs, like many other forms of economic
activity, may tend to cluster in space. Geographic clustering offers at least three types of
pecuniary benefits to firms: (i) reduced costs of transport of intermediate goods between
related businesses; (ii) ‘thick’ labor markets that allow faster, smoother hiring; and
(iii) reputational benefits due to the fact that customers come to know of the geographic
concentration (see e.g. Fujita et al., 1999). In order to control for spatial autocorrel-
ation, we construct an inverse distance squared metric of the outcome, FFLs, in all
counties other than the county of interest. We choose the inverse squared distance
metric over its cousin, the inverse distance, because theoretical and empirical evidence
suggests that spillovers from the economies mentioned above lead to increasing returns
to scale in highly localized settings and decline dramatically over space (e.g. Krugman,
1991, 1998; Gertler, 1995; Ratanawaraha and Polenske, 2003; Florida, 2005).
We define Y as the vector of FFLs for all counties, D as the matrix of travel times Di
between each possible pair of counties, and D0 1=D as the matrix of inverse travel
times whose elements are 1=Di , unless Di ¼ 0, in which case D0i ¼ 0. We then multiply in
0 0
scalar terms Y D D ¼ Z to obtain the squared distance-weighted influence of FFLs
in all counties on all counties. We then column-summed Z to obtain vector S, which we
then transpose to obtain ST , denoting the total spatial influenceP of all other counties’

T 10
FFLs on the county i in question, such that ST 3STi ¼ Sj ¼ i Zij . A similar
method is used to control for spillover effects from neighboring populations of potential
customers, creating the PopiDist variable. The latter, however, does not serve to control
for spatial autocorrelation, so much as account for nearby market potential.
We also tested various alternative ways of accounting for population in the model,
because population coefficient estimates are crucial to control for domestic demand.
For instance, we reran the model substituting FFLs per capita as an outcome, but that
model backed up or did not converge in almost every instance. We also tried accounting
for population as a relative difference from the mean term [i.e. ðPopi Pop= Pop], but
the term seemed not to afford any greater predictive power than population alone. We
also tried including a simple square term for population, which was not statistically
significant. See Appendix and Section 5 for further reflection on the difficulties of
capturing the effect of population on realized demand for firearms in these models.
3.2. Regression analysis

As the variance of the outcome variable greatly exceeds the mean (see Table 2), we
employed a negative binomial generalized linear regression.
The relationship in a negative binomial model between EðYit Þ, the expected number
of FFLs in county i at time t, and DCBi and TCBi —respectively, the distance to the
nearest stretch of border from county i and the time expected to drive that distance—
can be expressed as
EðYit Þ ¼ it ¼ expð0 þ Di þ 1 Xit1 þ 2 Xit2 þ . . . þ q Xitq þ "it Þ, ð1Þ
where and 1 ,2 , . . . ,q are regression coefficients, Di is the distance from the border
to county i, and Xit1 ,Xit2 , . . . ,Xitq are control variables.
The appropriateness of a negative binomial model is confirmed by the estimates of
random effects dispersion parameters lnðrÞ and lnðsÞ, both of which are highly
significant. Moreover, the incidence rate ratios for the model hover at around 1—not
nearly high enough to suggest convergence to a Poisson distribution.
10 For more on the inverse distance squared method, see Lloyd (2007), Ch. 2.1: ‘Approaches to local
adaptation’.
Table 2. Descriptive statistics for the outcome and underlying predictor variables, 1999
Statistic Total FFLs Distance to border crossing (km) Distance to Mexican border (km)
Minimum 1 12.02298 1.78513

Median 15 1792.648 1782.25
Maximum 787 3929.954 3905.144
Std. Dev. 37.70969 712.8791 710.5672
Mean 25.64258 1726.72 1711.911
Variance 1422.02 508196.7 504905.7
Observations 3086 3109 3109
3.3. Post-estimation analysis

In order to estimate the total volume of firearms traffic, we employ the epidemiological
concept of population attributable fraction (PAF), also called the etiologic fraction.
PAF is usually used to determine the proportion of disease incidence that is attributable
to exposure to the risk factor (Greenland and Drescher, 1993; Last, 2001, 137).11 Here,
we want to attribute the proportion of FFL ‘incidence’ that is attributable to exposure
to Mexico. The PAF for adjusted predictions is often defined as:

PrðEjDÞ RRadjusted 1
PAF ¼ , ð2Þ
RRadjusted
where PrðEjDÞ represents the prevalence of exposure given the disease (in our case, the
decile distance from Mexico given the number of FFLs), and RRadjusted represents the
adjusted relative risk of disease (here, the adjusted prediction of FFLs per population
with the inclusion of the Mexico variable, normalized by that without it) (see e.g.
Doidge et al., 2012). Take the simple example of smoking (exposure) and lung cancer
(disease). In this scenario, PrðE\vertDÞ represents the probability of being a smoker,
given that one has lung cancer. In effect, this term identifies the total population, in
proportional terms, that could be impacted if smoking were eliminated. RRadjusted , in this
scenario, represents the ratio of the probability of having cancer given smoking, to the
probability of having cancer given no smoking, controlling for age, genetic predispos-
ition, etc. In other words, it
is the percentage
increase in the risk of cancer associated
with smoking. Conversely, RRadjusted 1 =RRadjusted is then the percentage reduction in
adjusted risk of lung cancer given the elimination of smoking. Putting the two terms
together yields the proportion of lung cancers among the smoking population that
would be prevented with the elimination of smoking. It assumes that no lung cancers
among the non-smoking population could be prevented (e.g. through the removal of
second-hand smoke), using the unexposed population as a baseline.
11 There is some confusion in the epidemiologic literature regarding the difference between PAF and
Population Attributable Risk (PAR). While some authors suggest that the two terms are synonymous
(Bruzzi et al., 1985; Last, 2001), it is more common to see a differentiation made, such that the PAR
represents the decreased risk of a disease given the removal of a single factor, while PAF is the
proportion of the total number of cases that would be eliminated with that removal (Deubner et al.,
1975; Rockhill et al., 1988; Steenland and Armstrong, 2006; Rückinger et al., 2009).
Likewise in our application of the PAF, we assume that the decile of counties farthest
from Mexico experiences no demand from Mexico at all. It serves as our ‘unexposed’
baseline, and that any rise above that level is then attributable to demand from the US–
Mexico traffic. Unlike in most of the epidemiologic applications, our exposure variable,
road distance from the US–Mexico border, is categorical and not binary, though this
does not present any problems. In addition, the equation above tends to overestimate
the PAF, as it assumes perfect additivity of effects of variables when included in a single
model—i.e. EA þ EB ¼ EAþB (Walter, 1976, 1983). To mitigate that risk, we have
implemented Stata’s PUNAF module (Newson, 2012), which relies upon the Greenland
and Drescher (1993) formulation allowing for EA þ EB > EAþB . Returning briefly to
the smoking example, we might find that the removal smoking reduces the total number
of lung cancers by, say, 60%, but that elimination of smog would be predicted to reduce
lung cancers by 30%. The PUNAF module does not assume that all lung cancers
prevented by the elimination of smoking are mutually exclusive of those prevented by

the elimination of smog—many are exposed to both, and the elimination of one would
suffice to prevent their would-be lung cancer. Thus, the total percentage reduction of
lung cancers may, by the elimination of both smog and smoking, be likely less than
ð1 ð1 0:6Þ ð1 0:3ÞÞ 100 ¼ 72%.
In this case, the PAF function estimates the total FFLs in the country by summing
the adjusted predictions of FFLs across all decile distance groups of counties. This is
somewhat akin to taking the integral of the (presumably) downward-sloping adjusted
predictions curve of FFLs as a function of decile distance. The PAF function then
estimates the proportion of total FFLs in the countries that are ‘unattributable’ to the
predictor variable. This implies that the tenth decile distance group of counties
(containing, e.g., counties in the state of Maine) is assumed to experience 0 firearms
demand from south of the US–Mexico border, even though the adjusted prediction for
that group is greater than 0. Therefore, the adjusted prediction for the tenth decile
distance group of counties is assumed, ceteris paribus, to be the baseline level of FFLs
that would exist in a hypothetical alternative scenario in which all counties in USA were
removed to the distance from Mexico of the farthest 10% of US counties. Any rise
above that baseline level of FFLs is then considered to be ‘attributable’ to the presence
of the US–Mexico border. In effect, the PAF function ascribes that adjusted baseline
level of domestic demand to each of the county decile groups, and subtracts out the sum
of those attributed values. The remainder is then treated as a numerator and normalized
by the sum of adjusted predictions across decile distance groups to obtain a proportion.
4. Results
Table 3 shows only a list of variables, not regression results the number of FFLs as a
function of distance from the nearest border crossing (distance is broken into deciles),
time period, and an interaction between distance and period. The coefficients for the
categorical distance variable are uniformly highly significant, and robust and stable in
both sign and magnitude under a number of different tested control models. The results
of Model 3 are illustrated in Figure 6, with the ‘price’ (in decile distance) on the
horizontal axis and the quantity (in FFLs) on the vertical axis. The downward trend in
numbers of FFLs overall as well as the demand effect of US–Mexico border proximity
are clearly visible. However, the fact that US FFLs are becoming increasingly
Table 3. Random-effects negative binomial regression of total FFLs on distance from the US–Mexico
border crossings
(1) (2) (3)

Variables FFL12 FFL12 FFL12
Deciles of distance from Mexico by road

1 1.339*** 1.126***
(0.179) (0.180)
2 1.044*** 0.941***
(0.148) (0.149)
3 0.879*** 0.678***
(0.125) (0.126)
4 0.708*** 0.488***
(0.108) (0.109)
5 0.622*** 0.409***

(0.0964) (0.0984)
6 0.606*** 0.340***
(0.0899) (0.0920)
7 0.628*** 0.352***
(0.0801) (0.0823)
8 0.371*** 0.122
(0.0728) (0.0753)
9 0.427*** 0.217***
(0.0652) (0.0688)
10 0 0
(0) (0)
Time period (reference category ¼ 1993)
1994–1996 –0.0777*** –0.102***
(0.00535) (0.00970)
1997–1999 –0.119*** –0.218***
(0.0109) (0.0143)
2010–2012 1.393*** 1.213***
(0.0392) (0.0408)
Interactions between decile distance from Mexico and time period
Distance decile 1 in 1994–1996 0.0367**
(0.0153)
Distance decile 1 in 1997–1999 0.107***
(0.0176)
(0.0204)
Distance decile 2 in 1994–1996 –0.00616
(0.0153)
Distance decile 2 in 1997–1999 0.0329*
(0.0176)
(0.0206)
(0.0156)
(0.0177)
(0.0204)
(0.0157)
(continued)
Table 3. Continued
(1) (2) (3)


(0.0177)
(0.0205)
Distance decile 5 in 1994–1996 0.00309
(0.0157)
(0.0178)
(0.0205)

(0.0160)
(0.0181)
(0.0207)
(0.0147)
(0.0167)
(0.0189)
(0.0141)
(0.0161)
(0.0184)
(0.0135)
(0.0154)
(0.0179)
Domestic Demand Controls Yes Yes Yes
Constant 2.280 2.568* –0.795

(1.619) (1.422) (1.580)
Ln(r) constant 0.996*** 1.256*** 1.301***
(0.0279) (0.0287) (0.0290)
Ln(s) constant 0.363*** 0.469*** 0.473***
(0.0275) (0.0276) (0.0279)
Observations 30,809 30,809 30,809
Number of FIPS 3,107 3,107 3,107
Log likelihood –101901 –100133 –99814
Pseudo R2 0.242 0.252 0.251
Standard errors in parentheses.

***p50.01, **p50.05, *p50.1.
Effect sizes of 0(0) suppressed.
Bold coefficients are statistically significant at the p50.1 level.
350
300
250
Esmated Number of FFLs
1993
95% CI
200
1994-1996
95% CI
150
1997-1999
95% CI

100 2010-2012
95% CI
50
0
1 2 3 4 5 6 7 8 9 10
County Distance from U.S.-Mexico Border by Decile
Figure 6. Adjusted negative binomial predictions of number of FFLs per county by decile
distance from the US–Mexico border. Confidence interval between dashed lines; based on
Model 7-3.
dependent upon demand originating south of the border is less visible, as discussed
below.
Figure 6 indicates one encouraging indicator that the PAF is appropriately applied to
the whole of USA, rather than a subset of counties more proximate to the US–Mexico
border. As observed in this, adjusted predictions of firearms sales as a function of
decile distance from the Mexican border decline monotonically (with the exception of
minor upward jumps in deciles 7 and 9), and the declines are generally decreasing in size
with decile distance. As decile distance, unlike a continuous distance measure, enters the
regression equation as a categorical variable, there was no guarantee that the adjusted
predictions would produce the hypothesized slope at all. The fact that it nevertheless
implies that, as the ATF trace data suggest, most of the firearms purchasing in question
occur in border states. Counties in the fourth decile are predicted to realize only about
20% of the sales as counties in the first decile—much smaller, but not negligible,
numbers.12
12 We have also run the analyses using a continuous distance predictor. That specification does not allow us
to parse out time periods, but does exhibit the same downward sloping curve, steep at first and easing off
later. In fact, the resulting curve is much smoother than those we show, as the predictor enters the
equation as a single continuous variable, rather than a categorical variable that is treated as a series of
dummies.
Table 4. Population attributable fractions of FFLs on exposure to Mexican demand
Period Estimate (%) 95% LB (%) 95% UB (%)
1993 37.4 28.2 45.0

1994–1996 36.7 28.1 44.0
1997–1999 41.8 33.8 48.3
2010–2012 46.7 39.4 52.7
Given the regression results above, the PAF of FFLs attributable to the US–Mexican
arms trade is given in Table 4. It is important to note that the PAF does not tell us the
volume of total US demand for firearms attributable to US–Mexico traffic. Rather, it
estimates a percentage of FFLs that would go out of business in the absence of US–

Mexico traffic demand (for more on this, see Rowe et al., 2004). Therefore, the
hypothetical scenario in which demand arising from south of the border was removed
would imply a decrease in total demand at least equal to profit margins for the
corresponding number of businesses.
It is a well-known principle in microeconomics that the viability of a firm is deemed
to be largely, if not solely, a function of its profitability, and firms are assumed at all
times to behave so as to maximize profits (Stiglitz and Walsh, 2006, Ch. 7: ‘The
competitive firm’). In this instance, if the PAF predicts a certain fraction of FFLs that
would go out of business in the absence of demand from Mexico, we can therefore
assume that they would do so only if their profit margins have been eroded to 0. (In
fact, it would probably take negative profit margins to drive an FFL out of business,
but this assumption is a conservative one, as it implies a smaller percentage of the
domestic US firearms market attributable to the presence of Mexico.) In other words, it
is not the case that all the revenues of those FFLs predicted to go out of business are
impacted by the hypothetical removal of Mexican demand. Rather, only their profit
margins need be affected. Therefore, we want to multiply the average profit margin of
an FFL by the percentage of FFLs that are predicted to go out of business in the
absence of Mexican demand, thereby yielding a percentage of total FFL revenues that
would disappear in such a scenario.
To calculate the total volume of the US–Mexico trade, we must make some basic
assumptions about the profit margins in the firearms retail business. Ideally, we would
use survey data to derive these profit margins, but such data do not exist, according to
the authors’ knowledge.13 Informal communications with gun manufacturers and
retailers suggest that it is reasonable to assume a roughly 15% markup on firearms at
retail outlets on the primary market, with much larger markups for accessories. They
estimate that perhaps around half of the markup then go to pay overhead costs,
bringing the profit margin down to around 7%. However, the size of the retailer
obviously matters a great deal, and so we have included low-end, mid-range and high-
end estimates of FFL profit margins.14 We approximated low-end margins by averaging
13 The National Shooting Sports Federation conducts annual business surveys of FFLs, but the data are
neither publicly available nor numerically specific as to profit margins.
14 On profit margins, see Butler Consultants (2012).
Table 5. Assumed profit margins and resulting percentage estimates of total US domestic arms sales
represented by US–Mexico firearms trafficking demand
Profit Margin 0.0468 0.0234 0.07

Period Mid-Range Low-End High-End
1993 1.75% 0.87% 2.61%

(1.32%, 2.11%) (0.66%, 1.05%) (1.97%, 3.15%)
1994–1996 1.72% 0.86% 2.57%
(1.32%, 2.06%) (0.66%, 1.03%) (1.97%, 3.08%)
1997–1999 1.95% 0.98% 2.92%
(1.58%, 2.26%) (0.79%, 1.13%) (2.37%, 3.38%)
2010–2012 2.18% 1.09% 3.27%
(1.84%, 2.46%) (0.92%, 1.23%) (2.76%, 3.69%)

Table 6. Estimated annual total realized firearms demand for US–Mexico trade by period (units sold)
Period Mid-Range Low-End High-End
1993 134,045 67,023 200,495

(101,086; 101,086) (50,543; 80,730) (151,197; 241,501)
1994–1996 92,001 46,000 137,608
(70,440; 110,041) (35,220; 55,020) (105,359; 164,591)
1997–1999 87,890 43,945 131,460
(71,194; 101,718) (35,597; 50,859) (106,487; 152,142)
2010–2012 212,887 106,444 318,421
(179,632; 240,154) (89,816; 120,077) (268,681; 359,205)
Source: Calculations by the authors, based on ATF (2013).
those of gas stations (around 1.68%) and food and beverage stores (at around 3%). We
derived a mid-range estimate by averaging the profit margins of hardware stores
(4.45%), electronics stores (4.68%) and general stores (4.91%). We took the 7% figure
above as representing a high-end estimate. Using these estimates, we derived low-end,
mid-range and high-end estimates of US domestic arms sales attributable to US–
Mexico traffic (see Table 5).
Multiplying the above percentages by the total number of firearms sold in the US
domestic market, we can obtain the total number trafficked in each period. Results are
shown in Table 6.
We next monetized the value of these estimated volumes. We began by back-
calculating the total domestic revenues of the firearms industry from the reported
volume of excise taxes collected on domestic firearms and ammunition sales and their
relative excise tax rates (Figure 2) (ATF, 2002). We then applied same percentages
derived above to obtain the total monetary value of the US–Mexico firearms traffic. We
arrived at an annual dollar volume of around US$ 102 million in 2010–2012.
Robustness check of the results presented in this section using OLS models on the
untransformed and logged FFL outcome yields similar results (see Appendix).
However, when we normalized FFLs by population, the results did not hold. The
Appendix discusses this anomaly and reports on some statistical tests, suggesting that
the size of FFLs in terms of sales numbers and annual revenue may vary in accordance
with the size of the nearby market.
5. Discussion
Our estimates of numbers of firearms trafficked far exceed the total volume seized by
both the Mexican or American governments combined in recent years—that is, around
5000 and 32,300 by the US and Mexican authorities, respectively, in 2009 (Goodman
and Marizco, 2010, 170, 176). The mid-range estimate implies that the combined
seizures of roughly 37,000 firearms by Mexican and US authorities represent roughly
18.2% (between 10.8 and 43.1%) of the weapons bought for trafficking in recent years.
There are at least three possible reasons for the high numbers reported here: the first

two of which constitute major caveats for these estimates. First, proximity to the
Mexican border may be a proxy for other (domestic) factors than the US–Mexico
firearms trade. One possible example is the so-called ‘Minutemen’ militia, a group of
self-appointed citizen border patrollers who carry firearms. Law enforcement officers
associated with border security—for example, ICE and CBP officers—may themselves
contribute to demand. Another example of this type of unaccounted for domestic
demand, legal or undocumented immigrants may tend to settle close to the border, and
may also exhibit a greater per capita demand for firearms than the American populace
at large. Alternatively, border communities may also purchase more firearms in
response to immigrant influxes (though large immigrant communities are hardly unique
to the US–Mexico border region). More generally, the ‘Wild West’ gun culture may
really be a ‘Wild Southwest’ gun culture, so that the longitude variable may only
imperfectly control for gun culture. This is an important weakness of the study, but it is
difficult to measure ‘gun culture’ on the county-level. One convincing proxy might be
membership in the National Rifle Association, but those data are not publically
available. Perhaps, an even better proxy would be survey-derived gun ownership
statistics, such as those used in Miller et al. (2002). However, that study does not make
use of a county-level panel, but rather a state-level panel. Adding such data to a
county-level dataset could not qualify as a ‘fixed effect’, but rather a commission of the
ecological fallacy. Future research might usefully leverage similar survey on gun
ownership to better control for gun culture. Finally, as noted above, the likelihood of
traffickers being caught at Border Patrol road checkpoints probably diminishes with
distance from the border, effectively increasing traffickers’ ‘cost’ of traveling on roads
close to the border. However, as distance from the border is modeled using a categorical
decile variable, rather than a continuous variable, each decile is allowed to vary
independently of the others. Consequently, any additional clustering that might result
from this phenomenon would likely show up as a ‘balloon effect’, with lower values for
decile 2 made up for by higher ones in decile 1.
The second possible reason for the high numbers is that the FFL size may itself vary
in response to proximity of the border—or some other factor that correlates with
distance from the border. For instance, the demand for firearms may make it preferable
for FFLs to become smaller, but more numerous, and therefore harder for the ATF to
audit. A parallel from the (legal) tire trade may support this possibility. USA exports
large numbers of used and waste tires to Mexico. However, Mexican states issue quotas
to restrict the total number of tires that may cross the border. Consequently, some large
tire shipments may to be transported by semi to depots near the border. There,
informal, unpermitted drivers in pickup trucks load them up and cross in non-
commercial lanes at official ports of entry (Institute for Regional Studies of the
Californias, 2009, 15). A similar splintering of the market to hamper the enforcement of
regulations may be at work in the US firearms market. This suggestion would prima
facie seem to run counter to the decline in numbers (and consequent rise in presumed
sales volume), as stable or even rising demand would imply more, now fewer, FFLs if
their size is shrinking. However, controlling for the overall decline in FFLs, one might
imagine a relative increase in FFL numbers (and consequent relative decrease in size) in
areas where the ATF regularly conducts audits. As we have no records for volume of
firearms sold at specific FFLs or even within specific counties, we cannot discount this
possibility.
The Appendix discusses a similar problem associated with the estimated effect of

population and population proximity on FFLs. In our models, these terms’ coefficients
were universally negative, wrongly implying that the fewer people there are, the higher
the demand will be. The Appendix considers the possibility that, in metropolitan areas,
where smaller counties may be clustered tightly together, a single large gun store may
serve a large (but proximate) population. In contrast, FFLs in rural areas may be more
numerous but smaller, serving a small (but far-flung) population. Indeed, there is some
evidence that this phenomenon may be at work. These findings should lead us to interpret
with caution the results presented in Table 7 and subsequent calculations based thereon.
The large, sparsely populated counties near the US–Mexico border may naturally host a
number of FFLs that are disproportionate to the total volume of sales. These results may
require more nuanced modeling of the role played by population proximity in
determining FFL size. The major obstacle, of course, is the lack of data on sales
volume by county. Without such data, our best attempt at controlling for population
proximity (PopiDist) may miss the point, driving down the predicted FFLs in high-
density areas, but saying nothing about potentially larger volumes sold.
The third possible reason for such high numbers is that traffic that exists may include
volumes en route to destinations other than Mexico, be it in Central and South
America, or even elsewhere around the globe.
Table 7. Estimated annual total realized firearms demand for US–Mexico trade by period (US$ 1000)
(figures for 2001–2012 have been imputed)
Period Mid-Range Low End High End
1993 33,425 16,713 49,995

(25,207; 25,207) (12,603; 20,131) (37,702; 60,220)
1994–1996 33,724 16,862 50,443
(25,821; 40,337) (12,911; 20,169) (38,621; 60,334)
1997–1999 32,013 16,006 47,882
(25,931; 37,049) (12,966; 18,525) (38,786; 55,415)
2010–2012 107,160 53,580 160,282
(90,941; 120,885) (45,210; 60,643) (135,245; 180,812)
Source: Calculations by the authors, based on (ATF, 2002, 2013).

On the other hand, there are three reasons to believe that our estimates are credible.
First, we assumed in this article that the level of demand exhibited by the US–Mexico
firearms trade in the 10% of counties situated farthest from the US–Mexico border is 0.
In fact, this may not be so. We see evidence of heightened demand in the 9th decile
relative to the 10th (Figure 6), and there is nothing to say that if USA stretched farther
north, demand would not be felt there, as well. Mexican DTOs have documented
operations stretching into the northernmost reaches of USA. The National Drug
Intelligence Center (NDIC) (2009) reported the presence of DTOs over the period 2006–
2008 in such northern US cities as Bellingham, WA; Bismark, ND; Buffalo, NY; and
even Portland, ME (Finklea et al., 2011, Figure 2, 8). As the assumption of zero
demand in the 10th decile of US counties need not be the case, our estimates may not be
counting a latent level of demand underlying all deciles.
Second, our calculation of the percentage of US domestic sales represented by the
US–Mexico traffic rested on the very minimum amount of demand decrease that could

possibly cause an FFL to close. To the extent that an FFL might run budget deficits for
a number of years before finally closing, the decrease in demand required to put an FFL
out of business may be larger than that assumed here. That said, there is some anecdotal
evidence from online gun enthusiast forums that profit margins of the majority of FFLs
are much closer to those of gas stations than hardware stores. If so, then our low-end
estimates may be more realistic.
Third, our estimates are derived solely from data on regulated, above-board sales
occurring in ATF-licensed stores. To the extent that such venues may be under closer
scrutiny by the ATF than gun shows and private sales, the population of FFLs is likely
to exhibit attenuated effect sizes for US–Mexico traffic effects, vis-à-vis the wider group
of firearms sales outlets.
If our estimates are roughly correct, it may be that Mexico serves as a crossroad in
the international firearms trade. Shirk (2011, 7) reports that roughly 450,000 people
depend upon the drug trade in Mexico for their livelihood. If those are the only buyers
of arms, on average, each of them would have to purchase 0.33 firearms per year (or
one firearm every 3 years)15 in order generate the volume of demand we are estimating
here. However, DTOs may not represent the entirety of the Mexican demand. Illegal
traffic may be going to buyers in the general Mexican population (currently estimated
to be 115 million people) who are not affiliated with DTOs. Demand for firearms for
personal protection might also logically rise in an increasingly dangerous country.
Moreover, legal traffic may also be taking place, if Mexican police forces and others
purchase from FFLs in US border states. Nevertheless, the possibility of a transnational
trade remains.
Our study demonstrates that it is possible to leverage the locational decisions of
retailers to estimate the volume of an illicit trade. Despite important caveats that must
qualify our estimates, this technique represents one possible new tool to add to the
arsenal of economists and policymakers interested in quantifying illicit trades in the
usual context of scarce data. Future research might usefully improve and expand upon
the present study. We have discussed above various weaknesses using not only certain
variables—particularly the outcome (FFLs) as an adequate proxy for realized demand,
15 This figure is based on our mid-range estimate of 185,409 firearms purchased, minus approximately
37,000 seized by US and Mexican authorities, and divided by 450,000.
but also certain control variables that serve as proxies for ‘gun culture’. Future research
might make use of better variables. As discussed in the Appendix, it might also attempt
to model population’s effect on FFL market entry in the manner of Bresnahan and
Reiss (1991). Finally, future research might also make a more direct spatial link between
the use of firearms in Mexico (e.g. municipality-level homicide statistics) and the sale of
firearms in USA. Establishing such a link would require careful thinking as to how to
account for the dynamics of stocks and flows of firearms, including the rate at which
they become unusable. Moreover, the relationship between demand for firearms south
of the border and their supply north of the border would likely exhibit endogeneity.
Acknowledgments
The authors gratefully acknowledge Wiebe et al. (2009) for generously sharing their data on

Federal Firearms Licensees (FFLs) for the period 1993-1999. We also express our thanks to
Tilman Brück, Paul Holtom, Neil T.N. Ferguson, Chris Fariss, Barbara Buckinx, Kristian
Behrens, and two anonymous reviewers for their comments and suggestions on various drafts of
this paper.
References
ATF, (2002) Firearms Commerce in the United States 2001/2002. Washington, DC: Bureau of
Alcohol, Tobacco, Firearms, and Explosives.
ATF, (2013) Firearms Commerce in the United States, 2013. Washington, DC: Bureau of Alcohol,
Tobacco, Firearms and Explosives.
Bresnahan, T. F., Reiss, P. C. (1991) Entry and competition in concentrated markets. The Journal
of Political Economy, 99: 977–1009.
Bruzzi, P., Green, S. B., Byar, D. P., Brinton, L. A., Schairer, C. (1985) Estimating the population
attributable risk for multiple risk factors using case-control data. American Journal of
Epidemiology, 122: 904–915.
Butler Consultants, (2012) Free industry statistics - sorted by highest gross margin Available
online at: http://research.financial-projections.com/IndustryStats-GrossMargin [accessed 5
March 2013].
Chicoine, L. (2011) Exporting the Second Amendment: U.S. Assault Weapons and the Homicide
Rate in Mexico. South Bend, IN: Department of Economics, University of Notre Dame.
Corchado, A., Connelly, T. (2008) Analyst: U.S. anti-drug aid proposal could heighten violence
in Mexico. Dallas Morning News, 03 January 2008.
Della Vigna, S., La Ferrara, E. (2010) Detecting illegal arms trade. American Economic Journal:
Economic Policy, 2: 26–57.
Deubner, D. C., Tyoler, H. A., Cassel, J. C., Hames, C. G., Becker, C. (1975) Attributable risk,
population attributable risk, and population attributable fraction of death associated with
hypertension in a Biracial population. Circulation, 52: 901–908.
Doidge, J. C., Segal, L., Gospodarevskaya, E. (2012) Attributable risk analysis reveals potential
healthcare savings from increased consumption of dairy products. The Journal of Nutrition:
Methodology and Mathematical Modeling, 142: 1772–1780.
EIA, (2013) Gasoline and Diesel Fuel Update. Washingon, DC: U.S. Energy and Information
Administration.
Finklea, K. M., Krouse, W. J., Rosenblum, M. R. (2011) Southwest Border Violence: Issues in
Identifying and Measuring Spillover Violence. Washington, DC: Congressional Research
Service.
Florida, R. (2005) The world is spiky. The Atlantic Monthly, October, pp. 48–51.
Fujita, M., Krugman, P., Venables, A. J. (1999) The Spatial Economy: Cities, Regions, and
International Trade. Cambridge, MA: MIT Press.
GAO, (1996) Federal Firearms Licensees: Various Factors Have Contributed to the Decline in the
Number of Dealers. Washington, DC: GAOAvailable online at: http://www.gpo.gov/fdsys/pkg/
GAOREPORTS-GGD-96-78/html/GAOREPORTS-GGD-96-78.htm [accessed 25 June 2013].
Gertler, M. S. (1995) ‘‘Being There’’: proximity, organization, and culture in the development and
adoption of advanced manufacturing technologies. Economic Geography, 71: 1–26.
Goodman, C., Marizco, M. (2010) U.S. Firearms Trafficking to Mexico: New Data and Insights
Illuminate Key Trends and Challenges, pp. 167–203. Washington, DC: Woodrow Wilson
International Center for Scholars.
Grayson, G. (2007) Drug war allies. San Diego Union-Tribune, 02 September 2007.
Greenland, S., Drescher, K. (1993) Maximum likelihood estimation of the attributable fraction
from logistic models. Biometrics, 49: 865–872.
Grillo, I. (2007) ‘Cop-killer’ guns from U.S. seen crossing into Mexico. Associated Press, 18
August 2007.
Hepburn, L., Miller, M., Azreal, D., Hemenway, D. (2007) The U.S. gun stock: results from the
2004 National Firearms Survey. Injury Prevention, 13: 15–19.
Institute for Regional Studies of the Californias, (2009) The Flow of Used and Waste Tires in the
California-Mexico Border Region. Sacramento, CA: California Integrated Waste Management

Board.
Koper, C. S., Roth, J. A. (2002) The impact of the 1994 Federal Assault Weapons Ban on gun
markets: an assessment of short-term primary and secondary market effects. Journal of
Quantitatiûe Criminology, 18: 239–266.
Krugman, P. (1991) Increasing returns and economic geography. Journal of Political Economy,
99: 483–499.
Krugman, P. (1998) Space: the final frontier. The Journal of Economic Perspectives, 12: 161–174.
Last, J. M. (ed.) (2001) A Dictionary of Epidemiology, 4th edn. New York: Oxford University
Press.
Lloyd, C. D. (2007) Local Models for Spatial Analysis. Boca Raton: Taylor & Francis.
Marks, A. (2006) Why gun dealers have dwindled. Christian Science Monitor, 14 March 2006.
Miller, M., Azreal, D., Hemenway, D. (2002) Rates of household firearm ownership and
homicide across US regions and states, 1988–1997. American Journal of Public Health, 92:
1988–1993.
Molzahn, C., Rodriguez Feirrera, O., Shirk, D. A. (2013) Drug violence in Mexico: data and
analysis through 2013. San Diego, CA: Trans-Border Institute, USD.
National Drug Intelligence Center (NDIC), (2009) National Drug Threat Assessment, 2009.
Washington, DC: U.S. Department of Justice.
Newson, R. (2012) PUNAF: stata module to compute population attributable fractions for
cohort studies (Version Stata 12). Available online at: http://ideas.repec.org/c/boc/bocode/
s457193.html#cites [accessed 15 Feburary 2013].
Ratanawaraha, A., Polenske, K. R. (2003) Measuring geography of innovation: a literature
review. In K. R. Polenske (ed.) The Economic Geography of Innovation, pp. 30–59. New York:
Cambridge University Press.
Rockhill, B., Newman, B., Weinberg, C. (1988) Use and misuse of population attributable
fractions. American Journal of Public Health, 88: 15–20.
Rowe, A. K., Powell, K. E., Flanders, W. D. (2004) Why population attributable fractions can
sum to more than one. American Journal of Preventive Medicine, 26: 243–249.
Rückinger, S., von Kries, R., Toschke, A. M. (2009) An illustration of and programs estimating
attributable fractions in large scale surveys considering multiple risk factors. BMC Medical
Research Methodology, 9: 1–6.
Serrano, R. A. (2008) U.S. guns arm Mexican drug cartels. Los Angeles Times, 10 August 2008.
Shirk, D. A. (2011) The Drug War in Mexico: Confronting a Shared Threat. Washington, DC:
Council on Foreign Relations.
Steenland, K., Armstrong, B. (2006) An overview of methods for calculating the burden of
disease due to specific risk factors. Epidemiology, 17: 512–519.
Stiglitz, J., Walsh, C. E. (2006) Economics, 4th edn. New York: WW Norton.
Vernick, J. S., Webster, D. W., Bulzacchelli, M. T., Mair, J. S. (2006) Regulation of firearm
dealers in the United States: an analysis of state law and opportunities for improvement. The
Journal of Law, Medicine, and Ethics, 34: 765–775.
Violence Policy Center, (2009) Iron River: Gun Violence & Illegal Firearms Trafficking on the
U.S.-Mexico Border. Washington, DC: Violence Policy Center.
Violence Policy Center, (2010) Lessons Learned: The Gun Lobby and the Siren Song of Anti-
Government Rhetoric. Washington, DC: Violence Policy Center.
Violence Policy Center, (2011) The Militarization of the U.S. Civilian Firearms Market.
Washington, DC: Violence Policy Center.
Walter, S. D. (1976) The estimation and interpretation of attributable risk in health research.
Biometrics, 9: 531–541.
Walter, S. D. (1983) Effection of interaction, confounding, and observational error on
attributable risk estimation. American Journal of Epidemiology, 117: 598–604.
Wiebe, D. J., Krafty, R. T., Koper, C. S., Nance, M. L., Elliott, M. R., Branas, C. C.
(2009) Homicide and geographic access to gun dealers in the United States. BMC Public
Health, 9: 199.
Appendix

As robustness checks, we repeat the analyses described in Sections 3 and 4 using OLS
estimators, first for the untransformed outcome variable (FFLs), then for the log-
transformed outcome [ln(FFLs)]. The results confirm the general results from our
negative binomial models. We then discuss rather more problematic robustness check
outcomes when we use an OLS estimator to predict FFLs normalized by population,
which fail to conform to our expectations.
A. OLS with untransformed outcome

The OLS estimator assumes a normal distribution in the outcome variable that clearly
does not obtain in this case of an untransformed FFL outcome. Nevertheless, the OLS
model yields fairly similar results to those of the binomial estimator, in terms of
coefficient signs and significances (Table 8).
When we plot the adjusted predictions from Model 8-3 above, it yields the graph as in
Figure 7. Note that periods 1 (1993) and 2 (1994–1996) are characterized by heightened
levels of FFLs at the farther deciles, which runs counter to our hypotheses, but that the
type of monotonic decay curve illustrated in Figure 6 is more the case here in periods 3
(1997–1999) and 4 (2010–2012).
Note, too, that the ninth decile is, in the first three periods, abnormally high. In order
to see why this might be the case, we created a time-series of GIS-generated maps of US
counties color-coded by FFLs per capita, FFLs per km2, and FFLs per capita per km2.
Overlaying the shapes of federal lands that may serve as hunting grounds (and a 100-km
buffer zone) onto this map blocked out the areas we might expect to have elevated per
capita numbers of gun shops. Of those states with counties in the ninth decile distance
from the US–Mexico border (Florida, Maryland, Michigan, Minnesota, North
Carolina, North Dakota, Ohio, Pennsylvania, South Carolina, Virginia, Wisconsin
and West Virginia), the uncovered areas of the map revealed swathes of Minnesota and
Pennsylvania that demonstrated higher-than-average FFL concentrations. We included
dummy variables for the eight states that had over 10 counties in the ninth decile
(Michigan, Minnesota, North Carolina, Ohio, Pennsylvania, Virginia, Wisconsin, and
West Virginia) and found that, indeed, Minnesota and Pennsylvania have FFL
concentrations that are higher than expected. Ohio’s dummy variable also had a
positive coefficient, but it narrowly missed the cutoff for statistical significance.
Table 8. Random-effects OLS regression of total FFLs on distance from the US–Mexico border crossings
(1) (2) (3)


1 65.63*** 63.80***
(8.156) (8.653)
2 39.37*** 16.30**
(6.628) (7.263)
3 28.30*** –2.513
(5.573) (6.336)
4 23.97*** –7.223
(4.877) (5.753)
5 19.07*** –11.84**
(4.375) (5.351)

6 14.09*** –23.24***
(4.053) (5.091)
7 11.83*** –13.54***
(3.695) (4.805)
8 11.88*** –11.54**
(3.419) (4.599)
9 17.60*** 5.758
(3.103) (4.383)
1994–1996 –16.02*** –35.16***
(1.083) (2.760)
1997–1999 –19.72*** –53.54***
(1.936) (3.181)
2010–2012 70.92*** 50.81***
(5.837) (6.383)
(3.792)
(3.807)
Distance decile 1 in 2010–2012 –8.374**
(3.879)
(3.785)
(3.793)
(3.861)
(3.786)
(3.794)
(3.862)
(3.787)
(3.796)
(continued)
Table 8. Continued
(1) (2) (3)


(3.861)
(3.791)
(3.798)
(3.853)
(3.785)

(3.793)
(3.849)
(3.784)
(3.787)
(3.813)
(3.783)
(3.786)
(3.810)
(3.786)
(3.789)
(3.812)
Constant 261.9*** 133.0** 22.39
(69.25) (64.51) (68.87)
Observations 30,809 30,809 30,809

Number of FIPS 3,107 3,107 3,107
Overall R2 0.520 0.538 0.549
Between R2 0.594 0.611 0.622
Within R2 0.392 0.412 0.424
Chi Square 23,210 25,004 26,225

***p50.01, **p50.05, *p50.1.
180
160
140
120
1993
100 95% CI
1994-1996
80
95% CI
60 1997-1999
95% CI
40

2010-2012
20
0
1 2 3 4 5 6 7 8 9 10
-20
-40
Figure 7. Adjusted OLS predictions of number of FFLs per county by decile distance from
the US–Mexico border. Confidence interval between dashed lines; based on Model 8-3.
(Surprisingly, dummy variables for Virginia, West Virginia and Wisconsin were all
significantly negative.) While this cursory analysis suffers from the ecological fallacy, it
provides some evidence that Minnesota, Pennsylvania and possibly Ohio counties are
responsible for higher-than-expected FFLs in the ninth decile.
The PAF analysis cannot be performed on this model because predicted outcomes for
the 10th decile are slightly negative in periods 3 and 4. No percentage change in a
negative number can create a positive number (as the estimates for the first decile
invariably are).
B. OLS with a log-transformed outcome

We, then, log-transform the outcome variable in order to normalize its distribution, and
thereby better conform to the requirements of the OLS. Again, this OLS model yields
fairly similar results to those of the binomial estimator, in terms of coefficient signs and
significances (Table 9).
We plot the adjusted predictions from Model 9-3 above to obtain the graph in
Figure 8. In this model, all periods are characterized by heightened levels of FFLs at the
farther deciles, especially in deciles 8 and 9. Again, though, this phenomenon is less
pronounced in periods 3 and 4 and the decay curve we expect is generally evident.
The PAF analysis in this case roughly conforms to the results of PAF based on the
negative binomial model (Table 10). In both the cases, the percentage of FFLs
Table 9. Random-effects OLS regression of logged total FFLs on distance from the US–Mexico border
crossings
(1) (2) (3)

Variables lnFFL12 lnFFL12 lnFFL12

1 1.293*** 1.203***
(0.177) (0.174)
2 0.968*** 0.928***
(0.144) (0.142)
3 0.501*** 0.454***
(0.120) (0.120)
4 0.389*** 0.293***
(0.105) (0.105)
5 0.292*** 0.225**

(0.0941) (0.0945)
6 0.0724 –0.0274
(0.0869) (0.0877)
7 0.204*** 0.0978
(0.0789) (0.0800)
8 0.329*** 0.224***
(0.0727) (0.0742)
9 0.325*** 0.255***
(0.0670) (0.0689)
1994–1996 –0.191*** –0.238***
(0.00741) (0.0190)
1997–1999 –0.311*** –0.444***
(0.0132) (0.0219)
2010–2012 0.744*** 0.527***
(0.0407) (0.0449)
(0.0260)
(0.0262)
(0.0269)
(0.0260)
(0.0260)
(0.0267)
(0.0260)
(0.0261)
(0.0267)
(0.0260)
(0.0261)
(continued)
Table 9. Continued
(1) (2) (3)

Variables lnFFL12 lnFFL12 lnFFL12

(0.0267)
(0.0260)
(0.0261)
(0.0266)
(0.0260)

(0.0260)
(0.0266)
(0.0260)
(0.0260)
(0.0262)
(0.0259)
(0.0260)
(0.0262)
(0.0260)
(0.0262)
(0.0262)
Constant 3.531** 1.276 –0.300

(1.490) (1.389) (1.460)
Observations 30,809 30809 30,809
Observations 30,809 30,809 30,809
Number of FIPS 3,107 3,107 3,107
Overall R2 0.366 0.417 0.435
Between R2 0.255 0.311 0.334
Within R2 0.737 0.764 0.768
Chi Square . . .

***p50.01, **p50.05, *p50.1.
140
120
100
1993
95% CI
80
1994-1996
95% CI
60
1997-1999
95% CI

40 2010-2012
95% CI
20
0
1 2 3 4 5 6 7 8 9 10
Figure 8. Adjusted OLS predictions of logged FFLs per county, given decile distance from the
US–Mexico border. Confidence interval between dashed lines; based on Model 9-3.
Table 10. Population attributable fractions of FFLs on exposure to Mexican demand (OLS model on
logged-FFLs)
Period Estimate (%) 95% LB (%) 95% UB (%)
1993 30.6 19.8 39.5

1994–1996 33.6 23.3 42.0
1997–1999 39.1 29.6 46.7
2010–2012 43.1 34.1 50.1
attributable to demand arising south of the US–Mexico border is rising over time, and
all estimates hover between 25% and 45%. The OLS model yields slightly lower
estimates for periods 1–3 than the negative binomial model, but higher ones in period 4.
C. OLS with FFLs per population

As a final robustness check, we used an OLS model to estimate a newly-generated
outcome variable: FFLs per population. The results were counter-intuitive, and
coefficients for decile distance were all negative. That is, the models predict that at
greater distances from the US–Mexico border, FFLs per capita will be fewer. In fact,
the explanation is fairly straightforward, and implies that FFLs per capita is not an
appropriate outcome measure.
The crux of the matter is that FFLs may differ in the volume they sell based on
whether they are located in densely or sparsely populated parts of the country. This
possibility is supported by the fact that, while unreported in previous regression tables,
the coefficients for population are negative in the negative binomial and OLS models,
and that for the measure of nearby populations is negative in all three specifications.
The population coefficient was 5.58e-07 (s.d.: 3.56e-08) in model 3-3, 0.000113 (s.d.:
3.09e-06) in model 8-3, and 6.36e-07 (s.d.: 3.99e-08) in model 9-3. The coefficient for the
inverse distance-weighted population was 1.53e-09 (s.d.: 2.77e-10) in model 3-3,
9.76e-08 (s.d.: 1.70e-08) in model 8-3, and 9.11e-10 (s.d.: 3.15e-10) in model 9-3. In
effect, the models are implying that the fewer the people (and therefore smaller the
potential market), the greater the total market demand, which of course runs counter to
microeconomic logic.16 But for instance, in metropolitan areas where smaller counties

may be clustered tightly together, a single large gun store may serve a large (but
proximate) population. In contrast, FFLs in rural areas may be more numerous but
smaller, serving a small (but far-flung) population.
To test whether this explanation was indeed the case, we stratified our sample by
quintiles of nearby populations (PopiDist), and reran the model 7-3 stratified by those
subsamples. The population coefficient became positive (though not significant in the
two lowest market potential quintiles) in each subsample, implying that FFLs do differ
in sales volume based on proximity to nearby populations.
16 The coefficient for income is also always negative, implying that firearms are an inferior good.
View publication stats

JEconGeogr 2014 McDougal Jeg Lbu021 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

JEconGeogr 2014 McDougal Jeg Lbu021 PDF

Uploaded by

Copyright:

Available Formats

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

Article in Journal of Economic Geography · June 2014

Topher Mcdougal Robert Muggah

SEE PROFILE SEE PROFILE

Cost of Conflict in Nigeria View project

Political Economy of Rural-Urban Conflict View project

The user has requested enhancement of the downloaded file.

Journal of Economic Geography (2014) pp. 1–31 doi:10.1093/jeg/lbu021

The Way of the Gun: Estimating Firearms Trafficking

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Keywords: Firearms, trafficking, demand, North America, US–Mexico

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Number of Firearms Retail Licenses

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Imported to the United States

Manufactured in the United States

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

due to increased scrutiny of firearms trafficking by the Calderón Administration, as

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

200000 Type 2: Pawnbroker

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

include the most fundamental determinants of domestic demand: total population,

Table 1. Control variables

Variable Description Source

Period Time period (1 ¼ 1993, 2 ¼ 1994–1996, 3 ¼ 1997–1999, and

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

3.2. Regression analysis

Minimum 1 12.02298 1.78513

3.3. Post-estimation analysis

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

(1) (2) (3)

Deciles of distance from Mexico by road

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

(1) (2) (3)

Distance decile 4 in 1997–1999 0.183***

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Constant 2.280 2.568* –0.795

Standard errors in parentheses.

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Table 4. Population attributable fractions of FFLs on exposure to Mexican demand

Period Estimate (%) 95% LB (%) 95% UB (%)

1993 37.4 28.2 45.0

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Profit Margin 0.0468 0.0234 0.07

1993 1.75% 0.87% 2.61%

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Period Mid-Range Low-End High-End

1993 134,045 67,023 200,495

Source: Calculations by the authors, based on ATF (2013).

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Period Mid-Range Low End High End

1993 33,425 16,713 49,995

Source: Calculations by the authors, based on (ATF, 2002, 2013).

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014

Downloaded from http://joeg.oxfordjournals.org/ by guest on June 14, 2014