Citizenship Status and Arrest Patterns for Violentand Narcot
Spatial analysis of Mexican Homicides and Proximity to the US Border
1. Kezia
E
Dinelt
UC
SAN
DIEGO
|
IRGN
490:
GIS
&
SPATIAL
DATA
ANALYSIS
PROFESSOR
GORDON
MCCORD
DECEMBER
17,
2014
Spatial
Analysis
of
Mexican
Homicides
and
Proximity
to
the
U.S.
Border
DOES
PROXIMITY
TO
THE
U.S.
BORDER
RESULT
IN
MORE
HOMICIDES
DUE
TO
HIGHER
LEVELS
OF
DRUG
TRAFFICKING
FROM
MEXICO
TO
THE
UNITED
STATES?
2. Introduction
There
has
been
substantial
research
done
in
recent
years
to
understand
the
worsening
situation
in
Mexico
in
regards
to
homicides
presumably
due
to
drug
trafficking
organization
(DTO)
violence.
In
2006,
Mexico's
president,
Felipe
Calderón
initiated
a
crackdown
against
DTOs,
causing
intense
backlash
throughout
the
country,
resulting
in
increased
violence
instead
of
a
desired
decrease
in
overall
violence.
It
was
reported
in
July
of
2010
that
upwards
of
28,000
people
had
died
as
a
result
of
drug-‐‑trafficking
violence
since
2006.1
The
United
States
has
tried
to
support
Mexico's
efforts
through
the
Mérida
Initiative,
through
$1.4
billion
of
assistance
in
the
form
of
equipment,
training,
and
other
means
from
2008
to
2012.2
This
situation
that
persists
today
is
interesting
because
Mexico
88𝑡ℎ
in
the
world
for
GDP
per
capita3
but
could
potentially
achieve
much
more
through
increased
tourism
and
FDI,
however
the
violence
seen
in
Mexico
has
been
known
to
cause
decreases
in
both.
The
GDP
ranking
is
a
reflection
of
the
problem
that
has
been
unfolding
in
Mexico
over
the
past
eight
years.
Hypothesis
Because
the
U.S.
is
a
large
market
for
drugs
coming
from
Mexico,
and
it
is
a
very
lucrative
business,
there
is
money
to
be
made
by
pushing
drugs
across
the
border.
Although
border
patrol
and
police
create
the
possibility
of
getting
caught,
this
does
not
deter
the
DTO
members;
one
can
assume
those
who
earn
a
living
from
smuggling
or
transporting
drugs
weigh
such
risks
less
heavily
than
the
benefits
received
from
completing
the
transactions.
Therefore,
patches
of
drug-‐‑related
murders
can
be
expected
to
be
prevalent
closer
to
the
border
than
further
from
these
points
of
entry.
Furthermore,
one
can
expect
there
to
be
substantial
U.S.
pressure
at
border
crossings
in
an
effort
to
thwart
trafficking
operations,
increasing
the
number
of
homicides
that
occur
at
or
very
near
to
the
border
points
of
entry
from
increased
security,
especially
since
9/11.
To
investigate
the
relationship
between
DTO
homicides
and
distance
from
border
crossings,
population
density
is
important
to
control
for
since
there
is
a
disparity
between
the
sizes
and
populations
within
each
municipality.
Inequality
also
plays
a
role,
which
is
controlled
for
by
a
municipal-‐‑level
Gini
Coefficient.
GDP
broken
down
at
this
level
would
be
of
great
interest
since
one
would
expect
lower
levels
of
crime
as
GDP
increases,
but
it
is
unavailable
at
this
time,
so
the
Gini
provides
a
similar
analysis.
Homicides
are
reported
as
DTO
murders
per
100,000
people
to
normalize
for
the
population
densities
throughout
the
country.
1
Congressional
Research
Services
2
Astorga
and
Shirk
(2010)
3
CIA
Factbook
3. Data
and
Methodology
To
look
at
the
relationship
between
DTO
homicides
and
distance
from
U.S.
border
entries
for
2010,
several
shapefiles
were
uploaded.
Those
include
shapefiles
for
national
boundaries,
Mexican
states
and
municipalities,
drug-‐‑related
murders
available
from
2010,
an
ocean
layer
as
well
as
world
cities.
These
layers
were
all
projected
to
North
American
Equidistant
Conic
since
the
region
of
interest
is
in
North
America,
and
because
distance
is
measured
from
border
points,
having
the
equidistant
conic
projection
best
suits
this
calculation.
To
look
only
at
points
of
entry
at
the
border,
a
layer
was
created
from
a
selection
of
cities
from
the
shapefile.
To
get
a
better
understanding
of
the
spatial
distribution
of
the
DTO-‐‑related
homicides,
10
classes
were
created
with
a
natural
breaks
(Jenks)
classification.
With
data
for
total
population
and
a
Gini
Coefficient
by
municipality
for
2010
from
CONAPO,
this
data
was
joined
to
the
"municipalities"
shapefile.
From
the
area
contained
in
the
same
shapefile,
area
was
converted
into
kilometers
squared
for
all
of
Mexico.
The
municipal
level
population
was
then
divided
by
km&
to
get
population
density
per
km&
,
which
would
serve
as
a
control
variable
in
the
regression.
In
a
similar
manner,
total
populatin
by
municipality
was
joined
to
the
DTO-‐‑related
homicides
dataset,
which
allowed
the
calculation
of
homicides
per
100,000
people
(the
reported
standard).
This
calculation
normalized
the
deaths
associated
with
DTOs
since
some
municipalities
are
quite
small,
in
which
case
one
murder
would
cause
a
high
murder
rate.
This
distribution
is
displayed
in
Figure
1.
To
create
the
main
independent
variable,
the
Euclidean
Distance
tool
was
used
to
create
a
raster
to
measure
the
distance
in
kilometers
from
the
U.S.
border
cities
to
areas
along
the
southern
borders
of
Mexico
(Figure
2).
Zonal
Statistics
was
used
to
calculate
the
average
distance
from
these
border
cities,
and
this
distance
data
was
joined
to
the
homicide
data.
This
information
was
joined
to
the
municipal
data
and
the
complete
table
was
exported
so
it
could
be
used
to
run
OLS,
both
in
ArcGIS
and
in
Stata.
This
dataset
contained
the
four
variables
of
interest:
DTO-‐‑related
homicides
(the
dependent
variable),
mean
distance
from
border
entries
at
the
U.S.-‐‑Mexican
border,
population
density
and
the
Gini
Coefficient
(independent
variable
and
control
variables,
respectively),
at
the
municipal
level,
the
unit
of
measurement.
Although
GDP
per
capita
would
have
been
an
ideal
control,
this
data
was
not
available,
so
the
Gini
Coefficient
at
least
controlled
for
the
distribution
of
inequality
throughout
the
country.
The
OLS
tool
was
used
in
ArcGIS
to
run
an
OLS
regression,
which
was
also
run
in
Stata
after
exporting
the
table
into
a
usable
format.
The
regression
results
are
shown
and
discussed
below
(Figure
3).
To
test
for
clustering
among
the
data,
the
Spatial
Moran's
I
tool
was
used,
adding
the
feature
class
of
municipalities
and
using
inverse
distance
as
the
conceptualization
of
spatial
relationships.
This
test
showed
that
autocorrelation
is
present,
with
a
Moran's
Index
of
0.1046
and
a
p-‐‑value
of
0.000,
which
means
that
there
is
internal
structuring
of
the
observations
and
that
these
4. observations
are
dependent
on
this
internal
structure.
Furthermore,
OLS
is
unbiased,
but
the
standard
errors
and
t
statistics
are
incorrect,
leading
to
incorrect
inferences.
To
account
for
this,
a
geographically
weighted
regression
(GWR)
would
be
used
to
account
and
adjust
for
this
issue.
As
can
be
seen
from
Figure
4,
there
was
an
error
that
caused
large
areas
of
"no
data."
The
source
of
this
error
is
unknown,
but
further
research
should
be
done
to
correct
for
this.
Adjusting
for
autocorrelation
by
use
of
the
Fotheringham
adjustment
and
running
GWR
in
ArcGIS,
a
more
robust
output
would
be
produced.
A
robustness
check
on
the
errors
can
also
be
run
in
Stata
by
clustering
the
regression
by
municipalities.
This
would
allow
for
arbitrary
spatial
autocorrelation
for
the
2,456
municipal
observations.
Due
to
the
errors
discussed,
GWR
was
not
utilized
in
this
analysis.
More
time
would
be
needed
to
find
the
error,
which
could
lie
in
the
data,
or
other
components
of
the
process.
Figure
1
5. Figure
2
Findings
From
the
OLS
regression
of
running
DTO
homicides
on
mean
distance
from
border
cities,
Gini
Coefficient,
and
population
density,
the
results
are
as
expected
in
that
all
variables
are
statistically
significant.
As
can
be
seen
from
Figure
3,
the
output
table
can
be
interpreted
in
the
following
way.
Holding
all
else
constant,
a
1
km&
increase
in
mean
distance
is
correlated
with
a
0.00002
decrease
in
homicides
per
100,000
people
on
average.
Although
this
number
is
extremely
small,
it
is
significant
at
the
95%
level.
Additionally,
holding
all
else
constant,
a
1-‐‑unit
increase
in
population
density
is
associated
with
a
0.002
decrease
in
homicides
on
average.
Again,
not
a
large
number,
but
this
is
also
significant
at
the
95%
level.
More
interesting,
on
average,
a
1-‐‑point
increase
in
the
Gini
Coefficient
(more
inequality)
is
associated
with
an
increase
of
125.15
homicides
per
100,000
people,
which
is
significant
at
the
95%
level
as
well.
This
seems
very
high
and
requires
further
investigation,
especially
since
one
would
expect
increases
in
violence
due
to
lower
levels
of
GDP
per
capita,
or
accessible
here,
more
inequality,
however
not
to
this
6. extent.
A
deeper
look
into
the
outcomes
of
inequality
in
Mexico
is
of
interest,
and
could
be
a
topic
of
further
study.
A
Probit
or
Logit
model
may
be
a
better
tool
for
analysis
here.
Figure
3
Limitations
There
are
several
limitations
to
be
addressed
in
this
analysis.
To
begin
with,
finding
complete
or
adequate
data
was
very
difficult.
Other
very
interesting
analyses
could
have
been
researched
if
more
data
were
available.
For
example,
having
access
to
transport
costs
involved
in
drug
trafficking
could
help
produce
a
cost-‐‑distance
calculation,
as
well
as
understanding
specific
geographic
components
of
roads
(tolls
versus
free
roads,
how
well
the
roads
are
maintained)
and
topographic
challenges,
such
as
how
mountainous
certain
regions
are,
making
the
costs
and
risks
easier
to
understand.
Having
this
knowledge
could
have
interesting
policy
implications,
not
only
for
the
U.S.-‐‑Mexican
border,
but
perhaps
for
borders
in
other
parts
of
the
world
as
well.
Other
controls
that
would
have
been
interesting
to
look
at
include
GDP
and
education
levels
by
municipality,
since
one
would
expect
as
these
two
variables
increase,
violence
and
homicides
would
decrease.
Electoral
outcomes
play
an
important
role
as
well,
especially
with
respect
to
the
crackdown
by
the
Calderón
administration
on
the
War
on
Drugs.
Controlling
for
the
party
in
power
by
district
could
provide
insight
on
how
Mexican
political
dynamics
affect
violence,
especially
because
corruption
and
lack
of
transparency
plays
a
role
in
this
situation.
A
time
series
or
panel
data
comparison
between
two
years,
say
2006
when
Calderón
began
to
crack
down
on
DTOs
and
related
offenders,
and
2010,
four
years
into
the
crackdown,
would
be
interesting
because
one
could
look
at
state
governance
and
make
inferences
about
levels
of
responsiveness
and
corruption
based
on
results.
Policing
data
would
bring
in
the
element
of
corruption
analysis,
however
this
data
is
extremely
insufficient.
There
is
a
corruption
index
but
not
at
the
municipal
level,
and
as
of
now,
there
is
no
data
on
the
distribution
of
police
bribery
or
extortion.
_cons -19.13542 8.206902 -2.33 0.020 -35.22981 -3.041042
popdens -.0021192 .0008281 -2.56 0.011 -.003743 -.0004953
gini_10 125.145 20.15789 6.21 0.000 85.61378 164.6762
mean -.0000208 3.99e-06 -5.20 0.000 -.0000286 -.0000129
hom_pc Coef. Std. Err. t P>|t| [95% Conf. Interval]
Total 4649498.69 2132 2180.81552 Root MSE = 45.99
Adj R-squared = 0.0301
Residual 4503022.15 2129 2115.08791 R-squared = 0.0315
Model 146476.539 3 48825.513 Prob > F = 0.0000
F( 3, 2129) = 23.08
Source SS df MS Number of obs = 2133
. reg hom_pc mean gini_10 popdens
7. Figure
4
Conclusion
In
closing,
Mexico's
drug
trafficking
organizations
are
violent
groups
that
have
contributed
to
the
deaths
of
tens
of
thousands
of
people
in
Mexico,
and
the
efforts
on
behalf
of
the
government
to
combat
them
have
worsened
the
situation,
as
discussed
above.
The
findings
from
this
analysis
show
that
being
nearer
to
border
crossings
is
associated
with
more
DTO-‐‑related
murders.
This
is
not
to
say
that
areas
further
from
the
border
are
not
violent,
as
there
are
hotspots
of
violence
that
fall
within
the
drug
trafficking
corridors.
Further
research
should
be
done,
specifically
by
comparing
instances
of
violence
between
years,
to
identify
specific
actions,
policies,
party-‐‑in-‐‑power
strongholds,
and
regions
where
DTO
homicides
have
occurred.
There
are
policy
implications,
which
include
but
are
not
limited
to
establishing
stronger
institutions
to
reduce
corruption
and
enforce
lawfulness
to
a
higher
extent.
Additionally,
analyzing
this
by
looking
at
drug
running
along
the
coasts
from
boats
and
ports
could
enhance
the
analysis
and
also
have
repercussions
on
policymaking.
By
improving
transparency
in
Mexico,
one
would
expect
to
see
less
8. corruption,
and
overall
violence
including
kidnappings,
bribery,
and
extortion,
and
could
contribute
to
an
increase
in
tourism
and
FDI;
Mexico
needs
both
of
these
things
to
experience
more
economic
growth.
9. Bibliography
Astorga,
Luis,
and
David
A.
Shirk.
"Drug
Trafficking
Organizations
and
Counter-‐‑Drug
Strategies
in
the
U.S.-‐‑Mexican
Context."
EScholarship.
University
of
California,
01
Jan.
2010.
Web.
12
Dec.
2014.
Beittel,
June
S.
"Mexico’s
Drug
Trafficking
Organizations:
Source
and
Scope
of
the
Rising
Violence."
Congressional
Research
Services.
N.p.,
11
Jan.
2011.
Web.
13
Dec.
2014.
"North
America:
Mexico."
Central
Intelligence
Agency.
Central
Intelligence
Agency,
n.d.
Web.
14
Dec.
2014.
Data
Resources
Population:
http://www.conapo.gob.mx/es/CONAPO/Proyecciones_Datos
GIS
DTO
related
murders:
http://esoc.princeton.edu/
GIS
Mexican
Municipalities
and
States:
http://esoc.princeton.edu/country/mexico
All
other
data
accessed
from
Gordon
McCord
(professor)