ArticlePDF Available

The Emergent Network Structure of the Multilateral Environmental Agreement System

Authors:
The
emergent
network
structure
of
the
multilateral
environmental
agreement
system
Rakhyun
E.
Kim
a,b,
*
a
Fenner
School
of
Environment
and
Society,
The
Australian
National
University,
Australia
b
Institute
of
Advanced
Studies,
United
Nations
University,
Japan
1.
Introduction
It
is
generally
accepted
that
a
de
facto
‘system’
of
international
environmental
law
and
governance
has
emerged
(Freestone,
1994;
Boyle
and
Freestone,
1999;
Najam
et
al.,
2004;
Bodansky,
2006).
This
acknowledgement
stems
from
the
observation
that
interna-
tional
norms
and
institutions
do
not
exist
in
isolation
but
as
embedded
in
a
maze-like
structure
(Young,
1996,
2002).
However,
we
know
little
about
the
macroscopic
structure
and
evolutionary
dynamics
of
this
system
(Biermann
and
Pattberg,
2008;
Young,
2010a).
Our
understanding
has
not
advanced
much
beyond
‘congestion’
and
‘fragmentation’
rhetoric
based
on
anecdotal
evidence
(Ivanova
and
Roy,
2007).
There
is
a
clear
need
to
study
the
system
empirically
and
in
toto,
and
unravel
this
alleged
institutional
maze.
Such
an
understanding
of
the
emerging
complexity
would
prove
useful
in
improving
the
alignment
between
the
governance
system
and
the
multifaceted
challenges
of
governing
the
interactions
of
different
Earth
system
processes
(Rockstro
¨m
et
al.,
2009;
Walker
et
al.,
2009;
Galaz
et
al.,
2012;
Nilsson
and
Persson,
2012).
This
study
fills
the
knowledge
gap
by
revealing
and
analysing
dynamic
patterns
in
the
structural
organization
of
international
environmental
law
and
governance.
I
take
a
network-based
approach,
which
uncovers
the
underlying
system
architecture
by
reducing
the
system
to
an
abstract
structure
capturing
only
the
basics
of
connection
patterns
between
its
components
(Newman,
2010).
The
core
analytic
unit
is
neither
the
whole
system
nor
individual
components,
but
rather
the
relation
between
compo-
nents
that
gives
rise
to
large-scale
connection
patterns.
The
emergent
patterns
are
then
treated
as
mathematical
objects
or
graphs,
and
analyzed
with
network
metrics
such
as
modularity,
clustering
coefficient,
and
average
path
length.
These
topological
properties
reflect
differences
in
the
governing
system
structure
that
may
lead
to
significant
differences
in
governance
processes
and
outcomes
(Bodin
and
Crona,
2009;
Orsini
et
al.,
2013).
For
constructing
a
network
representation
of
the
institutional
structure
of
international
environmental
governance,
I
chose
multilateral
environmental
agreements
as
nodes
and
their
cross-references
as
links
that
define
the
relation
between
the
agreements.
Multilateral
environmental
agreements
are
treaties,
Global
Environmental
Change
23
(2013)
980–991
A
R
T
I
C
L
E
I
N
F
O
Article
history:
Received
3
August
2012
Received
in
revised
form
10
July
2013
Accepted
11
July
2013
Keywords:
International
environmental
law
Global
environmental
governance
Fragmentation
Defragmentation
Polycentricity
Network
analysis
A
B
S
T
R
A
C
T
The
conventional
piecemeal
approach
to
environmental
treaty-making
has
resulted
in
a
‘maze’
of
international
agreements.
However,
little
is
known
empirically
about
its
overall
structure
and
evolutionary
dynamics.
This
study
reveals
and
characterizes
the
evolving
structure
of
the
web
of
international
environmental
treaty
law.
The
structure
was
approximated
using
1001
cross-references
among
747
multilateral
environmental
agreements
concluded
from
1857
to
2012.
Known
network
analysis
measures
were
used
to
answer
the
following
questions:
has
a
complex
system
of
international
environmental
treaty
law
emerged?
If
so
when,
and
what
does
it
look
like?
What
are
its
topological
properties?
To
what
extent
is
the
institutional
complex
fragmented?
The
network
analysis
suggested
that
multilateral
environmental
agreements
have
self-organized
into
an
interlocking
system
with
a
complex
network
structure.
Furthermore,
the
system
has
defragmented
as
it
coevolved
with
the
increasing
complexity
and
interconnectivity
of
global
environmental
challenges.
This
study
demonstrates
the
need
to
approach
multilateral
environmental
agreements
in
the
context
of
a
complex
networked
system,
and
recommends
against
assuming
the
overall
institutional
structure
is
fragmented.
Proposals
for
global
environmental
governance
reform
should
pay
attention
to
this
network’s
emergent
polycentric
order
and
complexity
and
to
the
implications
of
these
features
for
the
functioning
of
the
multilateral
environmental
agreement
system.
!
2013
Elsevier
Ltd.
All
rights
reserved.
*
Correspondence
address:
United
Nations
University-Institute
of
Advanced
Studies,
International
Organizations
Center,
1-1-1
Minato
Mirai,
Nishi-ku,
Yokohama
220-8502,
Japan.
Tel.:
+81
45
221
2300;
fax:
+81
45
221
2302.
E-mail
address:
rakhyunkim@gmail.com
Contents
lists
available
at
ScienceDirect
Global
Environmental
Change
jo
ur
n
al
h
o
mep
ag
e:
www
.elsevier
.co
m
/loc
ate/g
lo
envc
h
a
0959-3780/$
see
front
matter
!
2013
Elsevier
Ltd.
All
rights
reserved.
http://dx.doi.org/10.1016/j.gloenvcha.2013.07.006
conventions,
charters,
statutes,
or
protocols
between
three
or
more
governments
relating
to
the
environment
(Mitchell,
2003;
Carruthers
et
al.,
2007).
They
typically
include
cross-references
to
a
number
of
other
such
agreements
that
their
parties
consider
relevant.
According
to
Kiss
and
Shelton
(2007),
these
cross-
references
can
be
viewed
as
extending
the
legal
effect
of
cited
texts
to
the
texts
that
cite
them.
I
selected
a
list
of
747
multilateral
environmental
agreements
concluded
between
1857
and
2012,
and
identified
1001
cross-
references
to
other
agreements
in
the
list.
Using
this
dataset,
I
produced
a
series
of
agreement-level
connectivity
maps
of
international
environmental
treaty
law.
I
investigated
the
structural
dynamics
of
the
network
by
focusing
on
the
following
questions:
has
a
complex
polycentric
system
emerged
among
multilateral
envi-
ronmental
agreements
through
self-organization?
If
so,
when,
and
what
does
it
look
like?
What
are
its
topological
properties?
To
what
extent
is
the
institutional
complex
fragmented?
The
questions
relating
to
the
dynamics
on
the
network,
that
is,
how
the
functioning
of
the
system
depends
on
its
topological
properties,
are
beyond
the
scope
of
this
paper.
Such
an
enquiry
would
require
representing
each
multilateral
environmental
agreement
as
a
dynamic
system
in
itself
(Churchill
and
Ulfstein,
2000;
Brunne
´e,
2002,
2012;
Gehring,
2007;
Wiersema,
2009;
Young,
2010a,
2010b)
and
further
specifying
the
causal
mecha-
nisms
of
institutional
interaction
(Young,
2002;
Gehring
and
Oberthu
¨r,
2009).
As
the
institutional
citation
network
is
an
abstract
representation
of
symbolic
relationships,
it
is
yet
unclear
how
its
network
measures
such
as
modularity
should
be
interpreted
with
respect
to
their
consequences
for
some
process
on
the
network.
Nonetheless,
where
possible,
explanations
were
offered
by
juxtaposing
the
observed
structural
changes
with
what
had
actually
happened
in
the
real
world.
The
paper
starts
with
a
brief
review
of
relevant
literature
to
which
the
present
network
analysis
contributes.
The
methods
section
then
follows,
explaining
what
cross-references
mean
in
the
context
of
multilateral
environmental
agreements
and
how
the
data
were
collected.
Key
empirical
findings
are
presented
in
two
sections
focusing
respectively
on
the
evolution
of
network
topology
from
1857
to
2012,
and
static
topological
properties
of
the
network
in
2012.
I
conclude
by
identifying
implications
of
the
analysis
of
this
structure
for
governance
outcomes.
2.
Fragmentation,
polycentricity,
and
networks
Institutional
fragmentation
has
received
significant
scholarly
attention
as
a
macroscopic
feature
of
international
environmental
law
and
governance
(e.g.,
Doelle,
2004;
Stephens,
2007;
Carlarne,
2008;
van
Asselt
et
al.,
2008;
Biermann
et
al.,
2009;
Boyd,
2010;
Scott,
2011;
van
Asselt,
2012;
Zelli
and
van
Asselt,
2013).
Although
there
is
no
consensus
on
its
meaning
and
implications
(Biermann
et
al.,
2009;
Zelli
and
van
Asselt,
2013),
the
underlying
idea
can
be
traced
to
the
notion
of
treaty
congestion
(Brown
Weiss,
1993;
see
also
Hicks,
1999;
Anton,
2012),
that
institutional
proliferation
has
led
to
chaos
and
anarchy.
From
a
polycentric
perspective,
however,
‘‘fragmentation
at
the
international
level
does
not
imply
anarchy’’
(Galaz
et
al.,
2012,
p.
22).
Numerous
independent
centres
of
decision-making
may
self-
organize
and
make
mutual
adjustments
that
order
their
relation-
ships
with
one
another
(Ostrom,
1999b,
2010).
This
process
may
give
rise
to
different
forms
and
degrees
of
polycentric
order,
where
stronger
forms
can
be
denoted
as
polycentric
systems
(Galaz
et
al.,
2012).
These
systems
are
comparable
in
their
structure
and
function
to
complex
adaptive
systems
(Ostrom,
1999a),
which
have
the
capacity
to
adapt
to
external
conditions
by
changing
their
rules
as
experience
accumulates
(Holland,
1995;
Levin,
1998;
Arthur,
1999;
Miller
and
Page,
2007;
Mitchell,
2009).
Because
of
the
complexity-handling
capacity
of
these
systems,
polycentrism
has
been
considered
as
one
appropriate
model
for
international
environmental
law
and
governance
(e.g.,
Folke
et
al.,
2005;
Ostrom,
2010).
However,
empirical
research
on
fragmentation
and
polycen-
tricity
at
the
international
level
has
been
hampered
by
inadequate
methods
and
a
lack
of
large
datasets.
For
example,
whereas
these
concepts
are
about
macro-level
architecture
in
a
time-dependent
sense,
most
previous
studies
have
examined
isolated
cases
of
dyadic
institutional
interaction
over
a
limited
period
of
time
(Zelli
and
van
Asselt,
2013).
We
need
to
go
beyond
such
reductionist
methodologies
and
study
the
architecture,
that
is,
the
system
of
institutions
at
the
macro-level
(Biermann,
2007).
Many
important
questions
remain
unexplored
from
a
dynamic
systems
perspective.
Network
theory
has
recently
emerged
as
a
widely
applied
tool
kit
for
studying
complex
systems
(Amaral
and
Ottino,
2004;
Newman,
2011).
The
most
important
breakthrough
in
network
science
has
been
the
discovery
of
striking
regularities
in
the
macro-
structures
of
many
complex
systems
that
exist
in
the
real
world
(Baraba
´si
and
Albert,
1999;
Watts
and
Strogatz,
1998;
Ravasz
et
al.,
2002).
These
common
design
principles
provide
a
powerful
justification
for
a
network
approach.
By
providing
a
common
language
and
empirical
methods,
network
theory
has
the
potential
to
bring
together
fragmentation,
polycentricity,
and
complexity
studies,
and
provide
some
novel
insights
into
the
structure
and
dynamics
of
international
environmental
law
and
governance
(e.g.,
Orsini
et
al.,
2013).
3.
A
citation
network
perspective
on
international
environmental
treaty
law
This
study
used
cross-references
as
proxies
for
the
evolving
structure
of
international
environmental
treaty
law,
a
strategy
justified
and
explained
below.
3.1.
Cross-references
as
proxies
for
relationships
among
multilateral
environmental
agreements
To
construct
the
complete
network
of
multilateral
environmen-
tal
agreements,
I
needed
to
define
objective
criteria
to
connect
them.
In
this
study,
I
used
‘‘interrelated
or
cross-referenced
provisions
from
one
instrument
to
another’’
(Kiss
and
Shelton,
2007,
p.
74)
or
simply
citations
or
cross-references
(these
terms
are
used
inter-
changeably
in
this
paper)
as
proxies
for
an
approximation
of
the
relationships
among
multilateral
environmental
agreements.
Most
agreements
contain
references
to
a
small
number
of
pre-existing
agreements
by
including
their
titles
in
the
treaty
texts,
often
in
preambles,
that
the
negotiating
states
consider
as
being
highly
relevant.
This
cross-referencing
has
been
noted
as
a
unique
common
characteristic
of
modern
environmental
treaties
(Kiss
and
Shelton,
2007).
Kiss
and
Shelton
(2007,
p.
87)
observed
that:
recent
environmental
agreements
increasingly
cross-reference
other
international
instruments.
Marine
environmental
trea-
ties,
for
example,
often
cite
to
[the
International
Convention
for
the
Prevention
of
Pollution
from
Ships,
1973,
as
modified
by
the
Protocol
of
1978]
or
[the
United
Nations
Convention
on
the
Law
of
the
Sea],
including
their
rules
by
reference.
The
result
could
be
to
extend
the
legal
effect
of
these
instruments
to
states
that
have
not
ratified
them
but
which
ratify
the
texts
that
cite
them,
especially
when
the
citation
affirms
the
norms
as
customary
international
law.
States
drafting
and
negotiating
a
multilateral
environmental
agreement
would
cross-reference
other
agreements
for
various
reasons.
The
most
frequently
observed
instances
are
when
states
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
981
acknowledge
the
positive
relevance
of
the
cited
agreement
on
the
issue
and
build
upon
it.
This
type
of
cross-reference
usually
appears
in
the
preamble
where
the
parties
to
the
agreed
agreement
are,
for
example,
‘noting’,
‘recalling’,
‘reaffirming’,
‘recognizing’,
‘bearing
in
mind’,
or
‘taking
into
account’
relevant
agreements.
A
typical
example
can
be
found
in
the
preamble
to
the
1992
United
Nations
Framework
Convention
on
Climate
Change,
where
its
parties
recalled
the
1985
Vienna
Convention
for
the
Protection
of
the
Ozone
Layer
and
the
1987
Montreal
Protocol
on
Substances
that
Deplete
the
Ozone
Layer.
In
some
cases,
such
as
in
the
United
Nations
Convention
to
Combat
Desertification,
a
multilateral
environmental
agreement
includes
a
cross-reference
to
recognize
the
contribution
that
it
can
make
to
the
cited
agreement.
Furthermore,
regional
agreements
often
cite
relevant
global
agreements,
such
as
the
United
Nations
Convention
on
the
Law
of
the
Sea
(UNCLOS),
to
include
the
basic
norms
previously
articulated
in
those
instruments
(UNEP,
2001a;
Kiss
and
Shelton,
2007).
Such
cross-references
are
also
used
when
sharing
defini-
tions
of
key
terms,
such
as
‘‘pollution
of
the
marine
environment’’
(UNCLOS
Article
1.1(4))
and
‘‘dumping’’
(UNCLOS
Article
1.1(5)),
creating
consistency
across
international
regimes.
Another
key
reason
for
citing
a
multilateral
environmental
agreement
is
to
define
the
relationship
between
the
citing
and
the
cited
agreements,
typically
in
conflict
clauses
(Wolfrum
and
Matz,
2003)
or
choice-of-law
provisions
(Kiss
and
Shelton,
2007).
For
example,
the
1992
North
American
Free
Trade
Agreement
gives
priority
to
the
obligations
set
out
in
named
environmental
agreements
in
the
event
of
any
inconsistency
(Article
104).
Moreover,
a
protocol
to
a
framework
convention
often
includes
a
specific
provision
that
defines
its
relationship
to
the
convention.
Less
frequently,
a
multilateral
environmental
agreement
cross-
references
when
replacing
an
existing
agreement
to
define
the
relationship
between
the
old
and
new
agreements
until
the
former
terminates.
It
should
be
carefully
noted
at
the
outset
how
citation
networks
differ
from
other
networks
(Leicht
et
al.,
2007;
Radicchi
et
al.,
2012).
First,
citation
networks
are
directed.
Citations
go
from
one
document
to
another,
involving
an
inherently
asymmetric
relationship
between
the
agreements
involved.
Second,
citation
networks
are
acyclic,
meaning
there
are
no
closed
loops
of
citations
of
the
form
‘A
cites
B
cites
C
cites
A’,
or
longer.
In
other
words,
when
a
new
agreement
is
added
to
the
network,
it
can
cite
any
of
the
previously
existing
agreements,
but
it
cannot
cite
agreements
that
have
not
yet
been
created.
This
gives
the
network
an
‘arrow
of
time’,
with
all
links
pointing
backwards
in
time.
Third,
the
time
evolution
of
citation
networks
takes
a
special
form,
in
that
nodes
and
links
are
added
to
the
network
at
a
specific
time
and
cannot
be
removed
later
(see
Appendix
A.1).
This
permanence
of
nodes
and
links
means
that
the
structure
of
the
network
is
mostly
static:
it
changes
only
at
the
leading
edge
of
the
network,
as
new
agreements
are
added.
In
principle,
citations
suggest
links
that
do
not
require
any
preceding
or
anticipated
institutional
interplay.
They
simply
capture
the
interests
of
the
parties
at
the
time
of
treaty
negotiations.
Therefore,
the
multilateral
environmental
agreement
citation
network
should
be
considered
as
a
‘symbolic’
network,
a
network
representation
of
abstract
relations
between
discrete
entities,
as
opposed
to
an
‘interactive’
network,
whose
links
describe
tangible
interactions
that
are
capable
of
transmitting
information,
influence,
or
material
(Watts,
2004).
In
other
words,
one
should
be
cautious
in
assuming
that
the
legal
and
governance
processes
as
practiced
are
reflected
in
the
citation
network
structure.
For
the
purpose
of
estimating
the
basic
system
architecture
of
international
environmental
treaty
law,
however,
the
citation
data
should
suffice.
The
validity
of
such
a
citation
network
approach
to
unravelling
legal
and
institutional
complexity
has
been
proven
in
previous
studies.
For
example,
several
scholars
have
used
legal
cross-references
when
studying
the
aggregate
structures
of
the
United
States
case
law
(Post
and
Eisen,
2000;
Fowler
et
al.,
2007;
Smith,
2007;
Fowler
and
Jeon,
2008),
the
United
States
Code
(Katz
and
Stafford,
2010;
Bommarito
and
Katz,
2009,
2010),
and
the
French
legal
system
(Boulet
et
al.,
2010,
2011).
In
particular,
Smith
(2007,
pp.
310–311)
considered
cross-references
as
linking
‘‘cases,
statutes
and
other
legal
authorities’’
together,
hence
allowing
a
study
of
law’s
overall
shape,
that
is,
‘‘how
law
is
organized
and
evolves’’.
Furthermore,
given
the
technical
difficulties
associated
with
collecting
other
types
of
connection
data
(see
Appendix
A.2),
cross-references
provide
practical
and
reliable
proxies
for
the
purpose
of
this
research.
3.2.
Dataset
compilation
Agreeing
on
what
is
and
what
is
not
a
multilateral
environ-
mental
agreement
is
not
a
straightforward
task
(Mitchell,
2003;
Scott,
2003;
Kiss
and
Shelton,
2007).
To
be
as
objective
and
comprehensive
as
possible
in
building
my
dataset,
I
combined
the
lists
of
multilateral
environmental
agreements
contained
in
the
two
most
comprehensive
international
environmental
agreement
databases:
the
IEA
Database
(Mitchell,
2013)
and
the
ECOLEX
(IUCN
et
al.,
2013).
I
also
added
a
small
number
of
agreements
that
were
missing
from
both
of
these
databases,
and
ended
up
with
747
in
my
dataset
(see
Table
A.1
in
the
appendix).
Amendments
were
excluded,
as
they
are
not
separate
agreements
but
form
an
integral
part
of
a
convention
or
a
protocol
(Carruthers
et
al.,
2007).
Examination
of
the
texts
(title,
preambular
paragraph,
opera-
tional
provisions,
and
annexes)
of
747
multilateral
environmental
agreements
identified
1001
cross-references
(see
Appendix
A.3
for
citation
data
collection
rules).
A
computer
programmed
and
automated
search-and-find
operation
was
not
considered
feasible,
as
formal
titles
of
these
agreements
were
not
used
consistently
across
the
agreements.
After
compiling
the
dataset,
I
constructed
and
visualized
the
institutional
network.
I
conducted
various
analyses
on
it
with
tools
developed
by
network
scientists
(e.g.,
Albert
and
Baraba
´si,
2002;
Newman,
2003).
Network
analysis
computer
programmes,
Pajek
and
Netminer,
were
used
to
provide
graphical
and
statistical
representations
of
the
system.
4.
Evolution
of
the
institutional
network
structure
from
1857
to
2012
Analysis
of
topological
changes
of
the
multilateral
environ-
mental
agreement
citation
network
between
1857
and
2012
provide
insight
into
the
evolution
of
the
institutional
system.
4.1.
Network
connectivity
The
network
representation
of
the
multilateral
environmental
agreement
system
I
constructed
evolved
in
156
steps,
from
a
single
node
in
1857
to
747
nodes
with
1001
directed
links
(or
986
undirected
links
with
multiple
lines
removed)
in
2012.
Fig.
1
shows
eight
graphical
snapshots
of
the
network
taken
at
ten-year
interval
from
1941
to
2011
(and
2012).
Increases
in
the
cumulative
number
of
agreements
adopted
and
cross-references
made
since
1857
are
shown
in
Fig.
2(a).
Multilateral
environmental
agreements
concluded
before
the
mid-1940s
often
contained
no
cross-
references.
The
average
number
of
cross-references
made
(i.e.,
outward
citations)
per
agreement
grew
rapidly
after
1992,
when
the
number
of
outward
citations
made
each
year
clearly
surpassed
the
number
of
agreements
adopted
each
year
(Fig.
2(a)).
The
total
number
of
outward
citations
surpassed
the
total
number
of
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
982
Fig.
1.
Graphical
representations
of
the
multilateral
environmental
agreement
citation
network
as
at
1941,
1951,
1961,
1971,
1981,
1991,
2001,
and
2011
(and
2012)
drawn
using
the
layout
algorithm
of
Fruchterman
and
Reingold
(1991).
The
nodes
of
the
largest
components
appear
in
blue.
(For
interpretation
of
the
references
to
colour
in
this
figure
legend,
the
reader
is
referred
to
the
web
version
of
the
article.)
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
983
agreements
in
1996
when
each
agreement
adopted
thus
far
had,
on
average,
one
outward
citation.
By
2012,
the
average
multilateral
environmental
agreement
made
and
received
1.3
citations
to
and
from
other
agreements,
which
means
that
an
average
agreement
has
2.6
direct
neighbours.
The
number
of
outward
citations
varies
from
0
to
18
with
a
standard
deviation
of
1.9
and
a
median
of
1.
The
number
of
inward
citations
varies
from
0
to
66
with
a
standard
deviation
of
3.7
and
a
median
of
0.
Among
the
747
agreements,
595
(80
percent)
have
at
least
one
connection
(i.e.,
either
inward
or
outward
citation),
and
152
(20
percent)
stand
alone
as
isolated
components.
4.2.
(De)fragmentation
Before
the
United
Nations
was
established,
there
were
only
a
few
multilateral
environmental
agreements,
most
of
which
were
not
related
to
each
other.
Roughly
coinciding
with
the
conclusion
of
the
Charter
of
the
United
Nations
in
1945,
the
number
of
agreements
increased
incrementally
over
the
next
three
decades,
but
without
fundamentally
changing
their
macro-structure.
Small
discrete
components
grew
bigger
in
size,
but
at
the
same
time
more
isolated
nodes
or
dyads
randomly
appeared
on
the
institutional
landscape.
This
network
representation
corresponds
to
Birnie
(1977)
who
observed
that
the
development
of
interna-
tional
environmental
law
at
the
time
was
not
systematic.
The
network
was
becoming
an
increasingly
disaggregated
set
of
discrete
international
institutions.
This
process
conforms
to
the
classic
definition
of
fragmentation
as
‘‘the
process
or
state
of
breaking
or
being
broken
into
small
or
separate
parts’’
(Oxford
English
Dictionary,
1989).
Such
structural
changes
could
be
quantified
by
a
simple
measure
of
the
fraction
of
the
largest
component,
which
I
plotted
in
Fig.
2(b).
The
fraction
of
the
largest
component
was
1
with
a
single
node
in
1857.
It
continued
to
decrease,
as
more
and
more
nodes
with
no
links
were
inserted
into
the
network,
until
the
fraction
reached
the
minimum
at
0.056
(or
5.6
percent)
in
1975.
The
network
then
consisted
of
252
multilateral
environmental
agreements
grouped
into
small
and
separate
128
components,
with
the
largest
component
consisting
of
only
14
agreements.
Since
1976,
however,
the
fraction
of
the
largest
component
has
increased
until
today,
and
it
stabilized
around
0.564
(56
percent).
If
we
accept
a
definition
of
fragmentation
based
solely
on
the
fraction
of
the
largest
component,
the
international
environmental
governance
architecture
was
most
structurally
fragmented
in
1975.
Furthermore,
the
institutional
network
has
since
increas-
ingly
defragmented.
I
acknowledge
that
such
a
structuralist
definition
might
be
overly
simplistic
by
neglecting
the
complex
nature
of
institutional
interaction,
which
may
be
functionally
cooperative
or
disruptive
(Gehring
and
Oberthu
¨r,
2006;
Biermann
et
al.,
2009).
The
definition
adopted
here,
however,
focuses
on
a
different
aspect
of
the
institutional
architecture.
Whereas
the
existing
scholarship
focuses
primarily
on
the
fragmented
implementation
of
multilat-
eral
environmental
agreements,
this
study
is
directed
towards
their
texts,
each
of
which
is
a
product
of
negotiation.
Therefore,
the
findings
on
defragmentation
should
not
be
seen
as
completely
contradicting
the
existing
literature
on
institutional
fragmenta-
tion,
but
as
providing
a
complementary
perspective.
There
are
also
other
ways
in
which
the
concept
of
defragmen-
tation
relates
to
and
differs
from
the
mainstream
understanding
of
Fig.
2.
(a)
Cumulative
number
of
multilateral
environmental
agreements
and
cross-references;
and
number
of
new
agreements
each
year,
and
different
distributions
of
inward
and
outward
citations
as
a
function
of
the
year
in
which
cited
and
citing
agreements
were
adopted,
respectively.
This
network
is
symmetric,
where
the
total
number
of
inward
citations
equals
the
total
number
of
outward
citations.
(b)
Number
of
components,
the
size
of
the
largest
component,
and
the
fraction
of
the
largest
component.
(c)
The
average
path
length
and
the
clustering
coefficient
of
the
multilateral
environmental
agreement
network.
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
984
the
fragmentation
of
global
environmental
governance
(e.g.,
Biermann
et
al.,
2009;
Zelli
and
van
Asselt,
2013).
Like
fragmentation,
defragmentation
is
a
value-free
concept
for
describing
the
overall
institutional
structure.
The
concept
of
defragmentation,
however,
is
intended
to
place
a
greater
emphasis
on
the
need
to
harness,
rather
than
manage,
institutional
complexity
(c.f.,
Axelrod
and
Cohen,
1999).
It
differs
from
integration,
a
term
commonly
used
as
an
antonym
of
fragmenta-
tion
(e.g.,
Keohane
and
Victor,
2011;
Young,
2011),
in
that
integration
generally
involves
interplay
management
towards
policy
coherence
(Nilsson
et
al.,
2012).
Defragmentation,
on
the
other
hand,
is
not
a
management
response
to
fragmentation,
but
a
self-organizing,
counteracting
process
which
may
occur
simulta-
neously
with
fragmentation.
The
beginning
of
structural
defragmentation
of
the
multilateral
environmental
agreement
network
roughly
coincided
with
the
emergence
of
modern
international
environmental
law,
which
was
marked
by
the
1972
United
Nations
Conference
on
the
Human
Environment
(also
known
as
the
Stockholm
Conference)
(Bod-
ansky
et
al.,
2007).
The
1970s
also
witnessed
the
births
of
the
earliest
forms
of
modern
multilateral
environmental
agreements,
such
as
the
1971
Convention
on
Wetlands
of
International
Importance,
especially
as
Waterfowl
Habitat,
the
1972
Convention
on
the
Prevention
of
Marine
Pollution
by
Dumping
of
Wastes
and
Other
Matter,
and
the
1973
Convention
on
International
Trade
in
Endangered
Species
of
Wild
Fauna
and
Flora.
Furthermore,
the
Regional
Seas
Programme
was
launched
with
the
creation
of
the
United
Nations
Environment
Programme,
which
has
led
to
the
conclusion
of
a
number
of
regional
seas
agreements.
The
emergence
of
these
modern
agreements
contributed
significantly
to
the
increasing
network
connectivity.
Does
structural
defragmentation
alone
indicate
a
‘systematiza-
tion
of
anarchy’
(c.f.,
Backer,
2012)?
Just
as
fragmentation
does
not
imply
anarchy
(Galaz
et
al.,
2012),
defragmentation
does
not
necessarily
imply
order.
Although
the
Stockholm
Conference
brought
about
a
change
in
the
old
laissez-faire
thinking,
it
is
still
questionable
whether
it
introduced
a
new
system
of
law
(Birnie,
1977).
4.3.
Systematization
of
anarchy
The
density
of
local
neighbourhoods,
as
measured
by
the
clustering
coefficient,
began
to
increase
in
the
1980s
(Fig.
2(c)).
The
clustering
coefficient
measures
the
fractions
of
potential
connec-
tions
among
network
neighbours
that
are
realized
(Watts
and
Strogatz,
1998).
In
other
words,
it
quantifies
how
close
the
local
neighbourhood
of
a
multilateral
environmental
agreement
is
to
being
part
of
a
‘clique’,
in
which
every
agreement
is
connected
to
every
other
agreement.
Therefore,
the
increasing
average
cluster-
ing
coefficient
of
the
network
indicates
a
corresponding
increase
in
the
level
of
redundancy
and
cohesiveness.
The
1990s
was
a
particularly
critical
period
in
the
evolution
of
the
multilateral
environmental
agreement
network.
The
network
reached
a
critical
level
of
connectivity
at
which
its
structure
changed
from
a
loose
collection
of
small
clusters
to
a
system
dominated
by
a
single
‘giant
component’
(Janson
et
al.,
1993;
Dorogovtsev
et
al.,
2008;
Newman,
2010).
This
system
state
transition
happened
around
1992
when
new
agreements
brought
a
few
shortcuts
into
the
network.
These
shortcuts
shrunk
the
size
of
the
network
while
maintaining
the
level
of
local
clustering.
The
average
path
length,
which
had
consistently
increased
since
1857,
started
decreasing
after
reaching
the
peak
of
6.53
in
1991
(Fig.
2(c)).
The
average
path
length
is
the
average
number
of
links
that
must
be
traversed
in
the
shortest
path
between
any
two
reachable
pair
of
nodes,
and
it
can
be
understood
as
a
global
measure
of
separation
(Watts,
1999,
2004).
By
1992,
the
average
path
length
dropped
from
6.53
to
5.47
(Fig.
2(c)).
The
network
diameter,
which
is
the
maximum
internode
distance,
also
declined
from
16
to
13
between
1991
and
1992.
In
1992,
the
multilateral
environmental
agreement
network
started
to
become
a
small-
world,
and
it
has
become
smaller
ever
since.
It
can
be
argued
that,
during
the
1990s,
the
‘‘partial
and
uneven’’
body
of
international
environmental
law
(Schachter,
1991,
p.
457)
underwent
systematization.
The
analysis
of
institutional
cross-
references
questions
the
conventional
wisdom
that
‘‘since
1992,
there
had
been
a
fragmentation
of
environmental
governance
and
issues’’
(UNEP,
2001b,
p.
2).
Empiricism
rather
confirms
the
claim
that
a
system
of
international
environmental
law
emerged
on
the
landscape
in
1992
from
a
mere
collection
of
environmental
norms
(Freestone,
1994;
Boyle
and
Freestone,
1999;
see
also
Najam
et
al.,
2004).
This
emergence
coincided
with
the
Earth
Summit
in
1992,
when
states
adopted
the
landmark
Rio
Declaration
on
Environ-
ment
and
Development,
Agenda
21,
the
Convention
on
Biological
Diversity,
and
the
United
Nations
Framework
Convention
on
Climate
Change.
4.4.
Self-organized
growth
The
multilateral
environmental
agreement
system
matured
in
the
2000s,
when
only
a
few
agreements
were
concluded
(Fig.
2(a)).
This
recent
trend
can
be
attributed
to
what
some
called
‘‘negotiation
fatigue’’
(Najam,
2000,
p.
4048;
see
also
Mun
˜oz
et
al.,
2009).
Anton
(2012),
for
example,
observed
that,
since
2002
and
more
noticeably
2005,
the
negotiation
and
adoption
of
multilateral
environmental
agreements
have
slowed.
Struggling
to
meet
current
treaty
obligations,
states
may
have
become
less
interested
in
creating
new
agreements
and
more
concerned
about
making
the
law
work.
This
is
also
reflected
in
the
2002
Johannesburg
Plan
of
Implementation.
The
noticeable
shift
of
resources
towards
implementation
after
three
decades
of
interna-
tional
cooperation
can
be
considered
as
a
sign
of
system
maturity
and
self-regulation
of
its
own
growth.
Although
the
horizontal
expansion
of
the
multilateral
environ-
mental
agreement
network
has
almost
halted,
its
internal
complexity
has
increased.
This
has
occurred
primarily
through
decisions
and
amendments
adopted
by
treaty
bodies,
which
this
study
did
not
consider.
The
internal
changes
have
often
been
made
in
response
to
new
scientific
information
about
the
state
of
the
target
environmental
phenomenon
(Gehring,
2007;
Huitema
et
al.,
2008;
Wiersema,
2009;
Brunne
´e,
2012).
Such
‘‘coherence
under
change’’
(Holland,
1995,
p.
4)
exhibited
in
recent
years
implies
that
the
multilateral
environmental
agreement
system
may
have
self-
organized
at
a
critical
state
of
‘stable
disequilibrium’
(Bak,
1996).
That
is
to
say,
international
environmental
law
has
reached
maturity
as
a
complex
system
which
displays
a
degree
of
institutional
resilience
and
adaptability.
This
may
also
suggest
that
the
system
as
a
whole
is
now
at
a
stage
where
further
institutional
stresses
may
trigger
abrupt,
non-linear
changes,
through
which
a
radically
new
system
is
installed
(Young,
2010b;
see
also
Walker
et
al.,
2009;
Biermann
et
al.,
2012).
4.5.
A
periodization
of
the
network
evolution
From
a
structural
evolutionary
perspective,
the
development
of
the
multilateral
environmental
agreement
system
can
be
divided
into
six
stages:
(1)
from
the
1850s
to
the
mid-1940s
(the
‘beginning’);
(2)
from
the
mid-1940s
to
the
mid-1970s
(the
period
of
‘incoherency’);
(3)
from
the
mid-1970s
to
the
1980s
(the
period
of
‘clustering’);
(4)
the
1990s
(the
period
of
‘emergence’);
(5)
the
2000s
(the
period
of
‘consolidation’);
and
(6)
the
2010s
(the
period
of
‘criticality’).
It
is
interesting
to
compare
this
periodization
with
the
conventional
description
of
the
historical
evolution
of
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
985
international
environmental
law,
which
identifies
the
years
1945,
1972,
and
1992
as
critical
transition
points
(Brown
Weiss,
1993;
Steiner
et
al.,
2003;
Redgwell,
2006;
Sand,
2007;
Birnie
et
al.,
2009;
Sands
and
Peel,
2012).
The
network
analysis
also
supports
the
contention
that
these
years
indeed
were
critical
turning
points
in
the
course
of
development,
given
that
we
accept
a
lag
of
a
few
years
since
the
year
1972
until
an
increasing
number
of
modern
multilateral
environmental
agreements
started
to
appear
in
the
mid-1970s.
5.
Analysis
of
static
topological
properties
Topological
properties
of
the
multilateral
environmental
agreement
network
in
2012
are
characterized
below
with
key
network
measures
and
metrics.
5.1.
Small-world
The
network
has
a
single
giant
component
of
421
multilateral
environmental
agreements
and
870
citations,
constituting
56
and
87
percent
of
the
entire
network,
respectively
(Fig.
1).
The
average
path
length
is
4.70
(4.71
for
the
giant
component)
(Fig.
2(c)),
and
the
two
reachable
agreements
that
are
furthest
apart
are
12
steps
away
(Fig.
3).
The
clustering
coefficient
for
the
network
is
0.43
(0.41
for
the
giant
component)
(Fig.
2(c)),
which
is
orders
of
magnitude
higher
than
0.005
(
!0.002),
the
clustering
coefficient
of
a
corresponding
Erdo
˝s–Re
´nyi
random
network
which
has
the
same
number
of
nodes
and
links
(Erdo
˝s
and
Re
´nyi,
1960).
The
high
clustering
coefficient
and
short
characteristic
path
length
suggest
that
the
giant
component
is
a
small-world
network.
In
other
words,
most
agreements
in
the
component
can
be
reached
from
every
other
by
a
small
number
of
steps.
This
is
so
despite
the
fact
that
the
network
contains
a
large
number
of
agreements,
that
each
agreement
is
connected
to
relatively
few
other
agreements,
and
that
the
network
has
no
dominant
central
agreement
to
which
most
others
are
directly
connected.
5.2.
Scale-free
The
multilateral
environmental
agreement
network
has
an
approximately
scale-free
topology.
This
means
that
the
degree
distribution,
the
probability
that
a
node
selected
uniformly
at
random
has
a
certain
number
of
links,
is
far
from
random,
but
heterogeneous
with
a
highly
skewed
tail
that
follows
a
particular
mathematical
function
called
a
power
law
(Baraba
´si
and
Albert,
1999).
I
tested
whether
the
network
is
scale-free
using
the
method
developed
by
Clauset
et
al.
(2009).
This
method
combines
maximum-likelihood
fitting
methods
with
goodness-of-fit
tests
based
on
the
Kolmogorov–Smirnov
statistic
and
likelihood
ratios.
After
goodness-of-fit
tests
with
1000
iterations,
with
the
null
hypothesis
that
the
degree
distribution
follows
a
power
law,
the
result
was
P-values
of
0.39
and
0.75
for
the
indegree
and
outdegree
distributions,
respectively.
P-values
significantly
larger
than
0.1
support
the
conclusion
that
they
are
drawn
from
a
power-law
distribution
(Clauset
et
al.,
2009).
Furthermore,
the
degree
distributions
in
log-log
scale
(Fig.
4)
show
that
straight
lines
would
fit
reasonably
well
through
the
dots,
which
is
roughly
suggestive
of
power-law
scaling.
The
heavily
right-skewed
degree
distributions
point
to
the
presence
of
relatively
few
agreements
with
extraordinary
numbers
of
links,
hence
power
and
authority,
despite
the
few
links
that
an
average
agreement
has.
In
fact,
the
top
10
percent
of
the
747
multilateral
environmental
agreements
garnered
about
65
percent
of
the
total
cross-references.
The
presence
of
such
‘hubs’
has
originated
from
a
micro-process
called
‘preferential
attachment’,
whereby
new
agreements
are
more
likely
to
make
connections
to
those
that
already
have
many
links
(Baraba
´si
and
Albert,
1999).
From
a
network
theoretical
perspective,
such
degree
heterogeneity
fosters
system
resilience
to
random
failures
but
system
vulnerability
to
the
failure
of
hubs
(Albert
et
al.,
2000;
Tu,
2000;
see
also
Young,
2010b).
To
identify
the
hubs,
I
used
a
variety
of
node-level
algorithms
and
measures,
such
as
the
Hyperlink-Induced
Topic
Search
Fig.
3.
Distribution
of
shortest
path
lengths
between
all
reachable
pairs.
Fig.
4.
Inward
citation
and
outward
citation
distributions
in
log-log
scale.
The
data
have
been
binned
logarithmically
to
reduce
noise.
R.E.
Kim
/
Global
Environmental
Change
23
(2013)
980–991
986
(Kleinberg,
1999)
and
betweenness
that
measures
‘‘the
degree
to
which
a
point
falls
on
the
shortest
path
between
others’’
(Freeman,
1977,
p.
35;
see
also
Wasserman
and
Faust,
1994).
In
2012,
the
United
Nations
Convention
on
the
Law
of
the
Sea
had
66
citations
and
is
currently
the
most
structurally
central
and
authoritative
multilateral
environmental
agreement.
A
possible
explanation
for
its
central
position
in
the
network
is
the
sheer
number
of
agreements
relating
to
regional
fisheries
management,
most
of
which
cite
the
Law
of
the
Sea
Convention.
The
runner
up
is
the
Convention
on
Biological
Diversity
with
34
inward
citations
and
1
outward
citation.
5.3.
Modularity
Modules
are
locally
dense
subgroups
of
multilateral
environ-
mental
agreements
that
are
relatively
densely
connected
to
each
other
but
sparsely
connected
to
agreements
in
other
dense
groups
(Porter
et
al.,
2009;
Fortunato,
2010).
In
governance
terminology,
modules
are
‘agreement
clusters’
(von
Moltke,
2005)
or
‘regime
complexes’
for
different
issue
areas
such
as
plant
genetic
resources
(Raustiala
and
Victor,
2004),
climate
change
(Keohane
and
Victor,
2011),
or
the
Arctic
(Young,
2011).
The
notion
of
clustering
of
agreements
has
been
the
subject
of
increasing
interest
to
governance
scholars,
especially
for
those
concerned
about
the
challenges
of
institutional
fragmentation
and
coordination
(Oberthu
¨r,
2002;
Roch
and
Perrez,
2005;
von
Moltke,
2005).
However,
their
arguments
have
been
largely
normative
and
based
on
anecdotal
evidence
of,
for
example,
deliberate
efforts
in
‘clustering
experiment’.
Here
I
take
a
broader
view
of
the
multilateral
environmental
agreement
system
and
present
empir-
ical
evidence
for
the
presence
of
naturally
emergent,
topical
agreement
modules.
Modularity
does
not
always
mean
clear-cut
subgroups,
but
there
may
be
overlap
between
modules.
To
find
the
best
partition
of
the
network
into
modules,
I
applied
a
community
detection
algorithm
developed
by
Newman
(2006).
This
algorithm
frames
the
problem
of
detecting
modules
as
an
optimization
task
in
which
one
searches
for
the
maximal
value
of
‘modularity’
over
possible
divisions
of
a
network
(Newman,
2006).
Modularity
is
quantified
by
calculating
‘‘the
number
of
edges
falling
within
groups
minus
the
expected
number
in
an
equivalent
network
with
edges
placed
at
random’’
(Newman,
2006,
p.
8578).
The
results
showed
that
the
multilateral
environmental
agreement
network
exhibits
a
modular
structure
consisting
of
a
high
modularity
score
of
0.75
(with
a
maximum
of
1),
which
is
comparable
to
the
modularity
of
a
co-authorship
network
of
scientists
working
in
condensed
matter
physics
(0.72)
(New-
man,
2006).
Newman’s
algorithm
identified
20
modules
within
the
giant
component.
A
scan
of
agreements
in
each
module
revealed
that
they
share
similar
subject
matter
or
topic,
confirming
the
presence
of
homophily
(McPherson
et
al.,
2001).
Sizeable
and
clearly
distinguishable
modules
include
the
marine
environment,
biodiversity,
maritime
safety
and
liability,
watercourses,
atmosphere,
hazardous
wastes,
plant
protection,
and
nuclear-related.
The
modular
structure
con-
formed
to
the
conventional
organization
of
law
with
its
modules
correlating
highly
with
underlying
legal
semantics
(UNEP,
2001a;
von
Moltke,
2005;
Smith,
2007).
Furthermore,
the
high
modularity
score
suggests
the
presence
of
sparse
inter-module
connections
called
‘weak’
ties
(Granovet-
ter,
1973).
These
weak
ties
play
an
important
role
in
global
connectivity.
For
example,
the
network
would
still
retain
its
macro-structure
even
if
some
of
the
‘strong’
intra-module
ties
were
removed,
whereas
removal
of
the
same
number
of
‘weak’
inter-module
ties
may
lead
to
a
fragmentation
of
the
entire
network.
5.4.
Nested
hierarchy
Low-degree
agreements
tend
to
belong
to
highly
cohesive
neighbourhoods
whereas
higher-degree
agreements
tend
to
have
neighbours
that
are
less
connected
to
each
other
(Fig.
5).
Such
an
inverse
correlation
between
degree
and
clustering
coefficient,
taken
together
with
a
heterogeneous
degree
distribution
and
modularity,
suggest
a
hierarchically
nested
organization
(Ravasz
et
al.,
2002;
see
also
Dorogovtsev
and
Mendes,
2002).
This
hierarchical
organization
does
not,
however,
refer
to
dominance
and
subservience
but
to
the
nested
structure
of
separate
but
interrelated
layers
that
expand
exponentially
in
width.
In
other
words,
agreement
modules
are
generally
made
up
of
smaller
and
more
cohesive
modules,
which
themselves
are
made
up
of
smaller
and
more
cohesive
modules
(Ravasz
et
al.,
2002).
6.
Interpreting
the
emergent
complexity:
from
structure
to
function
What
can
we
make
of
the
measured
structural
features
in
terms
of
collective