Relational collaboration among spatial multipoint competitors

Abstract

The presence of network ties between multipoint competitors is frequently assumed but rarely examined directly. The outcomes of multipoint competition, therefore, are better understood than their underlying relational mechanisms. Using original fieldwork and data that we have collected on an interorganizational network of patient transfer relations within a regional community of hospitals, we report and interpret estimates of Exponential Random Graph Models (ERGM) that specify the probability of observing network ties between organizations as a function of the degree of their spatial multipoint contact. We find that hospitals competing more intensely for patients across multiple geographical segments of their market (spatial multipoint competitors) are significantly more likely to collaborate. This conclusion is robust to alternative explanations for the formation of network ties based on organizational size differences, resource complementarities, performance differentials, and capacity constraints. We show that interorganizational networks between spatial multipoint competitors are characterized by clear tendencies toward clustering and a global core-periphery structure arising as consequences of multiple mechanisms of triadic closure operating simultaneously. We conclude that the effects of competition on the structure of interorganizational fields depends on how markets as physical and social settings are connected by cross-cutting network ties between competitors.

Full text

This article appeared in a journal published by Elsevier. The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy
Social
Networks
34 (2012) 101–
111
Contents
lists
available
at
ScienceDirect
Social
Networks
journa
l
h
o
me
page:
www.elsevier.com/locate/socnet
Relational
collaboration
among
spatial
multipoint
competitors夽
Alessandro
Lomia,b, Francesca
Pallottia,∗
aCenter
for
Organizational
Research
(CORe),
Faculty
of
Economics,
University
of
Lugano,
Switzerland
bUniversity
of
Bologna,
Italy
a
r
t
i
c
l
e
i
n
f
o
Keywords:
Multipoint
competition
Mutual
forbearance
Hospitals
Interorganizational
networks
Exponential
random
graph
models
a
b
s
t
r
a
c
t
The
presence
of
network
ties
between
multipoint
competitors
is
frequently
assumed
but
rarely
examined
directly.
The
outcomes
of
multipoint
competition,
therefore,
are
better
understood
than
their
underlying
relational
mechanisms.
Using
original
fieldwork
and
data
that
we
have
collected
on
an
interorganiza-
tional
network
of
patient
transfer
relations
within
a
regional
community
of
hospitals,
we
report
and
interpret
estimates
of
Exponential
Random
Graph
Models
(ERGM)
that
specify
the
probability
of
observ-
ing
network
ties
between
organizations
as
a
function
of
the
degree
of
their
spatial
multipoint
contact.
We
find
that
hospitals
competing
more
intensely
for
patients
across
multiple
geographical
segments
of
their
market
(spatial
multipoint
competitors)
are
significantly
more
likely
to
collaborate.
This
conclu-
sion
is
robust
to
alternative
explanations
for
the
formation
of
network
ties
based
on
organizational
size
differences,
resource
complementarities,
performance
differentials,
and
capacity
constraints.
We
show
that
interorganizational
networks
between
spatial
multipoint
competitors
are
characterized
by
clear
tendencies
toward
clustering
and
a
global
core-periphery
structure
arising
as
consequences
of
multiple
mechanisms
of
triadic
closure
operating
simultaneously.
We
conclude
that
the
effects
of
competition
on
the
structure
of
interorganizational
fields
depends
on
how
markets
as
physical
and
social
settings
are
connected
by
cross-cutting
network
ties
between
competitors.
© 2010 Elsevier B.V. All rights reserved.
1.
Introduction
Multipoint
contact
between
two
organizations
is
observed
when
they
encounter
each
other
simultaneously
in
multiple
seg-
ments
of
their
market.
Market
segments
may
be
distinguished
along
a
variety
of
dimensions
such
as,
for
example,
products,
cus-
tomers,
and
price
levels
(Baum
and
Haveman,
1997).
When
market
segments
are
defined
in
term
of
the
geographical
location
of
rele-
vant
production
and/or
consumption
activities
–
i.e.,
when
markets
are
considered
as
located
institutions
–
organizations
encountering
each
other
in
multiple
locales
are
spatial
multipoint
competitors
(Haveman
and
Nonnemaker,
2000).
夽We
gratefully
acknowledge
the
financial
support
of
the
Swiss
National
Science
Foundation
(Fonds
National
Suisse
de
la
Recherche
Scientifique;
Research
Grant
Number
124537).
Our
gratitude
extends
to
Nuffield
College,
University
of
Oxford,
and
to
the
Department
of
Psychology,
University
of
Melbourne
for
their
hospitality
during
part
of
the
work
leading
to
the
current
paper.
We
thank
the
Department
of
Public
Health,
Catholic
University
of
Rome
for
making
part
of
the
data
available.
Finally,
we
are
pleased
to
acknowledge
our
debt
of
gratitude
with
Garry
Robins,
Philippa
Pattison,
Johan
Koskinen,
Dean
Lusher,
Tom
Snijders,
and
Christian
Steglich
for
their
constant
encouragement
and
contribution
to
the
development
of
our
ideas.
Mistakes
are
ours.
∗Corresponding
author
at:
University
of
Lugano,
Institute
of
Management,
Via
G.
Buffi
13,
6900
Lugano,
Switzerland.
Tel.:
+41
058
6664471.
E-mail
addresses:
alessandro.lomi@usi.ch
(A.
Lomi),
francesca.pallotti@usi.ch
(F.
Pallotti).
Because
production
and
exchange
activities
are
frequently
con-
centrated
in
space
(Krugman,
1991)
spatial
multipoint
contact
is
an
empirical
regularity
of
contemporary
organizational
landscapes.
While
studies
of
multipoint
competition
have
produced
a
number
of
replicable
empirical
results,
extant
research
suffers
from
two
related
limitations
which
we
want
to
redress
in
this
paper.
Rooted
in
the
older
law
and
economics
literature
on
collusion
and
mar-
ket
power
(Edwards,
1955),
the
first
limitation
has
to
do
with
the
fact
that
typical
models
estimated
in
empirical
studies
only
look
for
(and
find)
evidence
of
implicit
collaboration
or
tacit
collusion
between
multipoint
competitors.
With
few
notable
exceptions
the
actual
mechanisms
that
make
collaboration
(and
collusion)
effec-
tively
possible
are
left
unspecified
(Baker
and
Faulkner,
1993).
This
problem
is
summarized
by
the
so
called
mutual
forbearance
hypothesis
according
to
which
each
multipoint
competitor
may
gain
by
allowing
the
other
to
take
a
dominant
position
in
one
domain,
while
conceding
dominance
in
another.
However,
the
for-
mation
of
tangible
network
ties
allowing
competitors
to
share
information
and
coordinate
their
actions
across
markets
is
rarely
examined
directly.
Available
studies
focus
on
some
of
the
possible
consequences
that
tacit
coordination
may
induce
such
as,
for
exam-
ple,
reduced
competitive
pressures,
increased
stability
in
market
shares,
and
collusion
(Barnett,
1993;
Bernheim
and
Whinston,
1990;
Gimeno,
1999).
In
this
paper
we
want
to
focus
instead
on
the
effect
of
spatial
multipoint
contact
on
the
presence
of
tangible
network
ties
between
competing
organizations.
0378-8733/$
–
see
front
matter ©
2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.socnet.2010.10.005

Author's personal copy
102 A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111
Directly
related
to
the
first,
the
second
limitation
concerns
the
level
of
analysis
at
which
multipoint
contact
operates
and
its
con-
sequences
supposedly
unfold.
Interorganizational
dyads
represent
the
smallest
unit
at
which
competition
can
be
meaningfully
mea-
sured
(Sohn,
2001),
but
extant
empirical
research
typically
focuses
on
organization-level
consequences
of
multipoint
contact.
Even
studies
based
on
dyadic
observation
plans
maintain
their
analyt-
ical
focus
on
organizational
outcomes
such
as
market
entry,
sales
growth,
market
share,
and
survival.
Yet
the
effects
of
multipoint
competition
may
be
fully
understood
only
by
considering
the
sys-
tem
of
local
dependencies
implied
by
attempts
to
coordinate
action
across
multiple
markets
(Fuentelsaz
and
Gómez,
2006;
Mizruchi
and
Marquis,
2006).
We
confront
this
limitation
by
estimating
new
Exponential
Random
Graph
Models
(ERGM)
that
afford
consider-
able
flexibility
in
specifying
patterns
of
dyadic
and
extra-dyadic
dependence
among
network
ties,
while
providing
a
rigorous
infer-
ential
framework
for
examining
the
effects
of
non-relational
factors
(Robins
et
al.,
2009,
2007;
Snijders
et
al.,
2006).
The
need
to
account
for
dependencies
generated
by
coordination
between
organiza-
tions
across
markets
has
been
recently
recognized
in
the
context
of
market
entry
(Fuentelsaz
and
Gómez,
2006).
However,
to
the
best
of
our
knowledge
no
study
is
available
that
has
analyzed
directly
the
structural
logic
of
such
dependencies
and
their
implications
for
the
tendency
of
multipoint
competitors
to
collaborate.
To
address
these
issues
empirically
we
rely
on
original
fieldwork
and
data
that
we
have
collected
on
patient
transfer
relations
within
a
regional
community
of
hospital
organizations.
Hospitals
provide
an
almost
ideal
example
of
spatial
multipoint
competitors
because
the
market
for
health
care
services
may
be
defined
meaningfully
only
with
reference
to
the
geographical
location
of
its
core
demand
(patients)
and
supply
(hospitals)
components
(Sohn,
2002).
As
it
is
typical
for
service
organizations,
and
unlike
most
manufactur-
ing
organizations,
hospitals
render
services
that
must
be
consumed
where
they
are
produced.
As
a
consequence
markets
for
health
care
services
and
competition
between
hospitals
are
highly
localized
(Garnick
et
al.,
1987).
For
example,
Phibbs
and
Robinson
(1993)
estimate
that
in
California
the
average
hospital
attracts
75%
of
its
patients
from
within
a
radius
of
only
8.5
miles,
and
90%
from
within
a
radius
of
18
miles.
No
definition
of
“market”
for
a
hospital
is
com-
plete
that
does
not
take
into
account
the:
“[R]elational
aspects
of
geographic
markets
and
the
division
of
labor
in
product
markets.”
(Dranove
and
White,
1994:
191).
While
it
is
now
common
to
emphasize
the
economic
aspects
of
hospitals
as
organizations
competing
on
price
and
quality
for
profitable
patients
(Gaynor
and
Vogt,
2003),
hospitals
do
differ
from
most
conventional
business
organizations
because
hospitals
operate
in
environments
that
are
jointly
technical
and
institutional
(Scott
and
Meyer,
1991[1983]).
As
organizations
operating
in
a
technical
environment
hospitals
are
undoubtedly:
“[R]ewarded
for
effective
and
efficient
control
of
their
production
system”
(Scott
and
Meyer,
1991:
123).
As
organizations
operating
in
an
institu-
tional
environment
hospitals
are
also
subjected
to:
“[I]nstitutional
rules
and
requirements
to
which
(they)
have
to
conform
if
they
are
to
receive
support
and
legitimacy”
(Scott
and
Meyer,
1991:
123).
Competition
and
opportunities
for
collaboration
between
hospi-
tals
can
be
understood
only
through
the
association
of
each
with
the
other
precisely
because
the
institutional
and
technical
compo-
nents
in
hospitals’
environments
are
inextricably
interwoven
(Ruef
and
Scott,
1998)
–
a
conclusion
also
reached
by
contemporary
anal-
yses
of
the
welfare
implications
of
competition
between
hospitals
(Kessler
and
McClellan,
2000).
The
intensity
of
multipoint
competition
between
two
hospi-
tals
is
a
function
of
the
overlap
in
the
spatial
distribution
of
their
actual
patients
(Sohn,
2002).
Other
conditions
being
equal,
hospi-
tals
attracting
patients
from
exactly
the
same
geographical
areas
are
likely
to
compete
more
intensely.
Our
main
contention
in
this
paper
is
that
spatial
locations
in
which
competitors
meet
create
social
settings
that
facilitate,
rather
than
inhibit,
interorganiza-
tional
collaboration.
Our
analytical
focus
is
on
the
presence
of
actual
network
ties
between
competitors
rather
than
on
their
pos-
sible
consequences.
Clearly,
competition
may
not
be
the
sole,
or
perhaps
even
the
main,
source
of
network
ties
between
organiza-
tions
in
general,
and
between
hospitals
in
particular.
For
example,
Iwashyna
et
al.
(2009a)
discuss
the
potential
implications
of
trans-
ferring
critically
ill
patients
from
low
to
high
performing
hospitals
–
a
possibility
that
we
explicitly
entertain
in
the
empirical
part
of
the
paper.
Whatever
the
source
of
network
ties,
patient
trans-
fer
is
not
possible
without
a
significant
level
of
explicit
relational
coordination
between
sender
and
receiver
hospitals.
Patient
transfers
reflect
an
underlying
relationship
between
the
partner
hospitals
involved
(Iwashyna
et
al.,
2009b).
Building
on
our
fieldwork,
we
treat
the
presence
of
patient
transfer
relations
as
the
observable
counterpart
of
the
latent
propensity
of
hospitals
to
col-
laborate
via
the
creation
of
network
ties.
We
model
the
probability
of
observing
network
ties
between
two
hospitals
as
a
function
of
the
degree
of
their
spatial
multipoint
contact
while
controlling
for
competitive,
institutional
and
organizational
factors
that
that
may
affect
the
propensity
of
organizations
to
collaborate.
As
we
explain
below,
the
modeling
framework
that
we
adopt
allows
us
to
keep
important
local
dependencies
into
account
and
specify
how
they
might
affect
endogenously
the
presence
of
network
ties.
Therefore
this
framework
provides
more
reliable
evidence
on
key
questions
of
collaboration
between
organizations
competing
across
spatial
domains.
2.
Collaboration
between
multipoint
competitors:
from
mutual
forbearance
to
interorganizational
networks
Departing
from
a
more
conventional
understanding
of
markets
as
exclusively
competitive
arenas,
a
major
line
of
contemporary
organizational
research
reframes
markets
as
social
settings
that
encourage
coordination
and
communication
among
participants
(White
and
Eccles,
1987).
In
this
view,
competitive
interde-
pendence
increases
reciprocal
awareness
and
facilitates
mutual
understanding
(White,
1988).
Because
competition
derives
from
similarity
in
positions
occupied
in
the
resource
space,
two
organi-
zations
compete
to
the
extent
that
they
depend
on
similar
resources
or,
in
other
words,
to
the
extent
that
their
niches
overlap
(Podolny
et
al.,
1996;
Popielarz
and
Neal,
2007).
But
the
development
of
similar
structural
interests
inherent
in
the
shared
dependence
on
material
and
symbolic
resources,
also
makes
organizations
more
likely
to
collaborate
(Ingram
and
Yue,
2008;
Trapido,
2007).
This
core
organizational
insight
is
perhaps
best
articulated
in
current
research
on
mutual
forbearance
between
multipoint
competitors.
Recasting
at
the
organizational
level
Georg
Sim-
mel’s
(1950)
interpretation
of
conflict
as
a
socially
binding
force,
the
mutual
forbearance
hypothesis
posits
that
the
potential
for
cooperation
will
be
higher
between
rivals
interacting
in
multiple
domains,
than
between
rivals
interacting
in
a
single
domain.
Fear
of
retaliation
across
domains,
mutual
deterrence,
collusion,
learn-
ing
and
linkage
negotiation
strategies
are
all
candidate
behavioral
mechanisms
that
may
be
underlying
mutual
forbearance
between
multipoint
rivals.
In
order
to
operate
effectively,
however,
each
one
of
these
possible
mechanisms
requires
some
form
of
con-
tact,
information
sharing
and
coordination
between
competitors
or,
in
a
word,
the
presence
of
a
meaningful
social
setting
(Baker
and
Faulkner,
1993;
White,
1988).
Contact
between
organizations
competing
simultaneously
in
multiple
markets
provides
one
such
setting
which
produces
common
perceptions
and
understandings
by
encouraging
competitors
to:
“[S]hare
a
great
deal
of
information
about
the
style
of
behavior
that
each
firm
adopts
vis-à-vis
the
oth-

Author's personal copy
A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111 103
ers,
that
is
the
social
context
in
which
they
operate”
(White,
1988:
228).
Organizations
typically
establish
their
presence
in
multiple
seg-
ments
of
the
market,
as
well
as
in
multiple
markets.
When
market
segments
are
defined
with
reference
to
spatial
considerations,
organizations
encountering
each
other
simultaneously
in
multiple
segments
are
spatial
multipoint
competitors.
An
extensive
liter-
ature
links
multipoint
competition
to
important
policy,
business,
and
organizational
implications.
The
probability
of
entering
into
new
market
segments
(Baum
and
Korn,
1996),
the
probability
of
exiting
from
established
markets
(Barnett,
1993),
organizational
growth
rates
(Haveman
and
Nonnemaker,
2000),
the
quality
of
ser-
vices
rendered
(Prince
and
Simon,
2009),
the
stability
of
market
shares
(Gimeno,
1999),
and
the
level
of
price
charged
across
market
segments
(Gimeno
and
Woo,
1999)
are
only
some
of
the
substan-
tive
concerns
that
have
been
related
to
the
presence
of
multipoint
contact
between
competing
organizations.
Contemporary
studies
of
multipoint
competition
have
inherited
their
almost
exclusive
focus
on
organizational
outcomes
from
the
older
tradition
of
antitrust
law
and
economics
(Edwards,
1955),
and
from
earlier
studies
on
collusive
behavior
of
business
firms
(Bernheim
and
Whinston,
1990).
But
the
presence
of
actual
net-
work
ties
which
would
make
such
outcomes
effectively
possible
is
rarely
examined
directly.
As
a
consequence,
the
outcomes
of
spatial
multipoint
contact
are
far
better
understood
than
the
net-
work
mechanisms
that
generate
and
sustain
them.
To
put
this
problems
in
a
sharper
focus
consider
carefully
Boeker
et
al.
(1997)
study
of
multimarket
competition
among
283
hospitals
–
the
same
kind
of
organization
that
we
analyze
in
the
empirical
part
of
this
paper.
These
Authors
find
that
hospitals
competing
across
multi-
ple
market
segments
are
less
likely
to
exit
from
segments
in
which
they
meet
their
multipoint
rivals.
Rather
reasonably,
Boeker
and
colleagues
interpret
this
outcome
as
a
consequence
of
interorgani-
zational
coordination
efforts.
The
important
result
that
they
report
is
that
interorganizational
coordination
really
does
affect
market
exit
decisions.
As
they
clearly
explain:
“A
critical
requirement
for
organizational
coordination
is
that
firms
maintain
relationships
with
each
other
and
monitor
each
other’s
activities
through
participation
in
similar
sets
of
markets.
Firms
that
maintain
a
presence
in
several
markets
in
which
their
competitors
also
participate
can
draw
on
cross-market
information
flows
to
enhance
their
knowledge
about
the
type
and
likelihood
of
competitive
behaviors
by
their
rivals”
(Boeker
et
al.,
1997:
128.
Emphasis
ours).
This
argument
typifies
the
reasoning
behind
the
hypothe-
sis
of
mutual
forbearance
and
involves
four
logical
steps.
First,
multipoint
contact
generates
interorganizational
relations
that
rivals
build
to
manage
their
joint
resource
dependencies.
Second,
the
presence
of
relations
facilitates
communication,
information
sharing,
and
coordination
between
organizations
across
mar-
kets.
Third,
coordination
alleviates
competitive
pressures.
Fourth,
reduced
competitive
pressures
produce
desirable
consequences
for
multipoint
competitors
(such
as
increased
growth
rates,
prof-
its,
and
opportunities
to
enter
new
markets).
But
the
presence
of
actual
interorganizational
relations
between
multipoint
competi-
tors
(which
sets
the
whole
process
in
motion
in
step
1)
is
assumed
rather
than
analyzed.
Only
possible
outcomes
of
relations
(described
in
step
4)
are
examined.
The
majority
of
available
studies
of
mutual
forbearance
among
spatial
multipoint
competitors
follow
a
similar
line
of
argument.
When
organizations
meet
each
other
in
multi-
ple
market
locations,
they
form
(dyadic)
relations.
These
dyadic
relations
“embed”
individual
organizations
into
broader
interor-
ganizational
networks
across
markets,
and
give
rise
to
complex
dependencies.
But
the
properties
of
these
dependencies
and
their
underlying
organizational
principles
remain
unspecified
because,
as
Fuentelsaz
and
Gómez
(2006)
observe,
coordination
assump-
tions
are
only
implicit
in
the
hypothesis
of
mutual
forbearance.
The
analysis
of
interorganizational
relations
curiously
assum-
ing
the
absence
of
network
dependencies
has
encouraged
the
adoption
of
analytical
approaches
based
on
assumptions
of
dyadic
independence.
Assumptions
of
dyadic
independence
are
partic-
ularly
troubling
in
the
study
of
multipoint
competition
because
simultaneous
presence
in
multiple
markets
gives
rise
to
a
“Net-
work
of
multimarket
contacts
in
which
firms
meet
each
other
in
several
markets”
(Greve,
2006:
249.
Emphasis
ours).
Current
studies
typically
maintain
the
crucial
assumption
of
independence
between
interorganizational
dyads
–
the
elementary
micro-
structural
components
of
networks
(Laumann
and
Marsden,
1982).
As
a
consequence
relatively
little
is
known
about
the
mechanisms
linking
patterns
of
local
(dyadic)
interaction
between
actual
or
potential
competitors
to
the
global
structure
of
interorganizational
networks.
Learning
about
such
mechanisms
is
particularly
impor-
tant
in
the
study
of
health
care
fields
because
inter-hospital
transfer
of
patients
frequently
gives
rise
to
unexpected
network
structures.
For
example,
in
the
specific
case
of
critical
care
Iwashyna
et
al.
(2009a)
find
that
U.S.
hospitals
tend
to
transfer
patients
to
several
other
hospitals
in
contrast,
at
least
in
part,
with
the
hub-and-spoke
model
that
is
typically
expected
to
coordinate
secondary
and
ter-
tiary
care
activities.
3.
The
empirical
setting:
patient
transfer
relations
between
hospitals
We
conducted
extensive
field
research
to
help
specify
our
model
of
collaboration
between
spatial
multipoint
competitors.
Over
a
period
of
approximately
one
year
we
conducted
a
number
of
face-to-face
and
telephone
interviews
with
hospital
managers,
executives,
presidents,
and
with
medical
doctors
working
in
some
of
the
hospitals
in
our
sample.
Our
field
experience
reveals
that
hospital
managers
consider
competition
a
meaningful
concept,
and
they
are
aware
that
patients
are
the
main
source
of
competitive
interdependence.
In
the
words
of
a
hospital
executive
we
interviewed:
“attempting
to
attract
patients
from
the
same
areas
is
one
of
the
main
source
of
compe-
tition
for
any
hospital.
The
ability
to
attract
patients
from
a
broad
area
gives
a
hospital
a
fundamental
competitive
advantage.”
As
it
is
common
for
most
service
industries,
the
market
for
health
care
has
a
clear
spatial
structure
because
health
care
services
have
to
be
rendered
on
demand
at
locations
that
are
ultimately
chosen
by
patients.
Markets
for
health
care
services
tend
to
be
highly
geo-
graphically
heterogeneous
giving
rise
to
fine-grained
variations
in
competitive
conditions
across
locations
(Gaynor
and
Haas-Wilson,
1999).
To
examine
the
effects
of
spatial
multipoint
contact
on
collabo-
ration
between
competitors
we
analyze
the
presence
of
network
ties
established
between
hospitals
by
virtue
of
patient
transfer
relations.
Patients
with
therapeutic
needs
for
which
the
referent
hospital
has
no
physical
capacity
available,
patients
with
complex
pathologies
for
which
the
hospital
does
not
have
adequate
tech-
nologies
or
clinical
competences,
or
patients
who
could
be
treated
more
efficiently
and
effectively
elsewhere
may
be
transferred
between
hospitals.
In
any
case,
observed
patients
transfer
events
reflect
the
presence
of
an
underlying
relationship
between
hospi-
tals
(Iwashyna
et
al.,
2009b).
In
public
health
care
systems
such
as
the
one
we
analyze,
data
on
inter-hospital
patient
mobility
are
routinely
collected
for
policy
purposes
to
assess
the
geographical
structure
of
demand
for
health
care
services,
and
the
institutional
coordination
requirements
that
such
structure
imposes
on
hospi-
tals
(Ugolini,
2001).
The
decision
to
refer
a
patient
to
a
different
hospital
may
be
taken
by
medical
doctors
only
in
the
best
interest
of
the
patient.
Yet,
choosing
a
partner
involves
an
organizational
decision
to
involve
another
hospital
in
the
search
for
a
common
solution
to
specific

Author's personal copy
104 A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111
clinical
or
therapeutic
problems.
To
be
effectively
possible,
patient
transfer
requires
a
high
level
of
interorganizational
coordination,
communication,
and
information
sharing
or,
in
a
word,
collabora-
tion
between
“sender”
and
“receiver”
hospitals.
Lack
of
adequate
coordination
between
hospitals
may
have
adverse
consequences
for
patients
being
transferred.
Building
on
research
on
interorgani-
zational
fields
in
health
care
sectors,
and
on
our
own
fieldwork,
we
consider
patient
transfer
as
an
observable
signal
of
inter-hospital
collaboration
(Levine
&
White,
1961;
Van
Raak
et
al.,
2005).
Our
fieldwork
produced
at
least
two
main
observations
in
sup-
port
of
our
understanding
of
patient
transfer
as
a
reliable
signal
of
collaboration
and
relational
coordination
between
hospitals
(Gittell,
2002).
The
first
is
that
transferring
patients
is
typically
not
considered
by
hospitals
as
a
solution
to
contingent
one-off
prob-
lems.
Rather,
patient
transfer
happens
in
a
highly
structured
social
context
that
facilitates
and
actively
promotes
interpersonal
contact
between
doctors
working
in
different
hospitals.
As
the
Director
of
Health
Care
Services
in
a
large
university
polyclinic
in
our
sample
revealed:
“More
than
getting
doctors
to
read
the
intricate
clinical
and
administrative
protocols
for
managing
patient
transfer,
the
real
problem
is
to
help
doctors
working
in
different
hospitals
to
get
to
know
each
other
–to
help
them
recognize
faces
and
remembering
names
so
that
when
they
have
to
talk
they
know
each
other
on
a
personal
basis.
This
is
the
main
purpose
of
meetings,
conferences,
workshops,
dinners
and
training
events
that
we
organize
with
other
hospitals
in
the
region.”
Our
fieldwork
produced
a
second
observation
in
support
of
our
understanding
of
patient
transfer
as
a
meaningful
signal
of
rela-
tional
collaboration
between
hospitals.
As
the
General
Manager
of
a
nationally
prominent
university
hospital
simply
put
it
in
an
inter-
view:
“Patients
that
we
receive
from
other
hospitals
are
a
source
of
revenues
for
us.”
But
if
patients
are
resources
for
the
control
of
which
hospitals
compete,
transferring
them
to
other
hospitals
is
a
decision
that
involves
at
least
some
element
of
collabora-
tion.
What
hospital
is
more
likely
to
be
chosen
as
partner?
Our
attempt
to
address
this
question
concentrates
specifically
on
the
transfer
of
in-patients.
In-patients
are
individuals
who
have
already
acquired
the
status
of
“admitted
patient”
and,
therefore,
have
con-
sented
to
follow
the
clinical
and
therapeutic
paths
proposed
by
professional
medical
staff
who
are
clinically
responsible
and
legally
liable
for
their
conditions.
This
is
an
important
qualification
because
individual
network
ties
induced
by
in-patients
transfer
are
the
outcome
of
organizational
decisions
over
which
patients
have
sur-
rendered
control
at
admission.
Of
course,
in-patients
retain
the
right
to
refuse
transfer
in
the
same
way
as
they
retain
the
right
to
refuse
treatment.
However,
they
cannot
choose
where
they
will
be
transferred
–
a
decision
that
remains
a
joint
prerogative
of
the
hos-
pital
transferring
the
patient
and
the
hospital
accepting
her.
Hence,
the
network
structure
of
in-patient
(henceforth
simply
“patients”)
transfer
between
hospitals
in
our
sample
can
be
legitimately
seen
–
and
modeled
–
as
the
outcome
of
a
complex
system
of
interrelated
organizational
decisions.
4.
Data,
models
and
methods
4.1.
Sample
and
data
We
examine
patient
transfer
relations
among
members
of
a
community
of
hospital
organizations
located
in
Lazio,
a
large
geo-
graphical
region
in
central
Italy.
Extended
over
approximately
17,000
square
kilometers,
Lazio
has
a
resident
population
of
roughly
5,300,000
inhabitants,
more
than
60%
of
whom
live
in
Rome,
the
capital
city.
The
health
system
in
Lazio
is
part
of
the
Ital-
ian
National
Health
Service
(NHS)
that
provides
universal
coverage
through
a
single
payer.
The
majority
of
hospitals
in
Italy
are
publicly
owned,
but
a
significant
number
of
investor-owned,
and
not-for-
profit
hospitals
also
receive
public
funding
through
a
contracting
system
with
the
NHS.
The
administrative
structure
of
the
Italian
health
care
system
is
inspired
by
federalist
principles:
regions
enjoy
high
levels
of
autonomy
in
the
organization
and
provision
of
health
care
services.
The
regional
health
system
in
Lazio
is
partitioned
into
12
Local
Health
Units
(LHUs).
LHUs
are
well-defined
territorial
and
admin-
istrative
units
responsible
for
the
financing,
organization,
and
(in
some
regional
contexts)
the
provision
of
health
care
services
within
their
jurisdiction
defined
in
terms
of
zip
codes.
LHUs
can
provide
care
either
directly
through
their
own
facilities
–
typically,
directly
managed
hospitals
–
or
by
paying
independent
public
structures
and
private
accredited
providers
for
services
rendered.
Patients
can
freely
choose
among
public
or
private
providers
that
are
expected
to
supply
services
of
comparable
cost
and
quality.
Patients
can
also
choose
to
be
treated
either
in
their
LHU
of
reference,
or
else-
where.
In
this
latter
case,
the
cost
of
care
will
be
paid
by
the
LHU
located
in
the
patients’
area
of
residence
thus
reinforcing
the
ten-
dency
of
hospitals
within
the
same
LHU
to
collaborate
and
exchange
resources.
Our
analysis
relies
on
both
primary
as
well
as
secondary
data
sources.
We
obtained
secondary
data
from
archival
sources
con-
tained
in
the
Regional
Hospital
Information
System
database
(SIO)
which
provides
information
on
the
residential
area
of
origins
for
each
patient
admitted
by
any
hospitals
in
the
region.
We
supplemented
official
sources
with
a
survey
designed
to
collect
information
on
a
number
of
dimensions
of
organizational
struc-
tures,
resources
and
hospital
activities.
Semi-structured
interviews
with
hospital
doctors,
managers
and
executives
were
also
con-
ducted
to
improve
our
contextual
understanding
of
processes
of
collaboration
and
competition.
Our
final
sample
includes
91
hospitals.
4.2.
Variables
and
measures
Using
publicly
available
data
on
transferred
patients
during
the
year
2003,
we
constructed
a
patient
mobility
matrix
(V
=
[vij])
of
size
91X91.
The
matrix
contains
in
each
row/column
the
hospital
sending/receiving
patients,
and
in
the
intersection
cells
(vij)
the
number
of
patients
transferred
from
the
row
to
the
column
hos-
pital.
The
overall
patient
flow
between
hospitals
is
13,178.
The
volume
of
transferred
patients
within
dyads
ranges
from
0
to
525
patients,
with
an
average
of
1.6
(standard
deviation
=
12.69).
On
average,
hospitals
transferred
patients
out
to
11
other
hospitals.
The
matrix
of
patient
transfer
relations
is
asymmetric,
since
for
any
hospital
in
the
sample
the
number
of
patients
sent
typically
differs
from
the
number
of
patients
received.
Because
we
are
inter-
ested
in
the
existence
of
network
ties,
rather
than
their
intensity,
we
dichotomized
the
matrix
by
using
the
overall
mean
number
of
transferred
patients
as
cut-off
value.
We
note
that
because
this
value
is
below
2,
various
methods
of
dichotomization
are
unlikely
to
produce
widely
different
aggregate
results.
We
have
consid-
ered
alternative
rules
to
construct
our
dependent
variable
that
include
dichotomization
using:
(i)
the
hospital
(“row”)-specific
average;
(ii)
the
average
cell
value
of
the
matrix
plus
one
stan-
dard
deviation;
(iii)
the
average
cell
value
of
the
matrix
plus
two
standard
deviations,
and
finally
a
(iv)
simple
dichotomization
(aij =
1
if
vi/=j0).
Obviously,
the
networks
that
we
obtained
vary
in
structural
properties.
Yet,
in
all
cases
we
re-estimated
the
model
that
we
report
in
the
empirical
part
of
the
paper
and
obtained
com-
parable
results.
We
have
also
tried
different
modeling
strategies.
In
particular
we
tried
a
logit
model
(with
standard
errors
cor-
rected
for
clustering)
and
a
fixed
effects
negative
binomial
model.
In
this
latter
case,
we
used
the
original
integer-valued
matrix
of
patient
mobility
and
modeled
the
strength
(rather
than
the
pres-

Author's personal copy
A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111 105
ence)
of
ties.
In
either
case,
our
basic
predictions
continued
to
hold. 1
Independent
variables.
The
main
independent
variable
of
theoretical
interest
is
spatial
multimarket
contact,
which
we
conceptualize
as
the
extent
to
which
hospitals
depend
on
sim-
ilar
resources
across
various
geographical
market
segments.
In
the
analysis
of
competitive
processes
between
organizations
the
concept
of
niche
is
frequently
used
to
specify
environmental
depen-
dencies:
two
organizations
compete
to
the
extent
that
they
rely
on
the
same
resources,
i.e.,
to
the
extent
that
their
niches
overlap
(Baum
and
Singh,
1994).
For
this
reason
the
notion
of
organizational
niche
is
frequently
used
to
summarize
the
link
between
environ-
mental
dependencies
and
competitive
conditions
(Popielarz
and
Neal,
2007).
Organizational
niches
may
be
defined
with
reference
to
various
kinds
of
resource
dependencies
(Podolny
et
al.,
1996).
Following
the
specialized
literature
on
competitive
interdependence
between
hospitals
(Sohn,
2002),
we
define
niches
in
terms
of
dependency
on
patients.
Patients
are
the
ultimate
consumers
of
health
care
ser-
vices
and
one
of
the
most
important
resources
for
hospitals
(Sohn,
2002).
Because
patients
are
spatially
distributed
and
assigned
to
defined
local
units
on
the
basis
of
their
area
of
residence,
our
mea-
sure
of
multimarket
contact
reflects
joint
dependence
of
hospitals
on
the
same
patients
across
locations.
Other
conditions
being
equal,
two
hospitals
are
multipoint
competitors
to
the
extent
that
they
depend
on
patients
living
in
the
same
areas,
i.e.,
to
the
extent
that
their
patient
pools
overlap.
The
amount
of
overlap
in
patient
pools
between
two
hospitals
is
directly
proportional
to
the
amount
of
competition
(Baum
and
Haveman,
1997):
when
two
hospitals
are
vying
for
the
same
patients,
changes
to
the
caseload
of
one
hospital
have
a
direct,
negative
impact
on
the
caseload
of
the
other
hospital
(Dexter
et
al.,
2005).
For
each
hospital
in
our
sample,
the
target
patient
pool
is
rep-
resented
by
all
the
residents
in
the
pertinent
LHU
in
which
the
hospital
is
located.
LHUs
partition
the
region
into
non-overlapping,
relatively
self-sufficient
market
areas
where
health
care
services
are
provided
by
public
and
private
accredited
hospitals.
As
in
other
Italian
regions,
LHUs
in
Lazio
play
the
role
of
payers
of
the
services
delivered
to
patients
by
health
care
providers.
Yet,
patients
may
choose
their
health
care
services
providers
regardless
of
its
specific
location,
and
hence
also
outside
their
LHU
of
residence.
Because
payments
to
hospitals
follow
the
patients,
hospitals
have
incen-
tives
to
acquire
higher
market
shares
by
attracting
patients
from
different
areas.
To
measure
overlap
in
patient
pools,
we
used
data
on
patients’
residence
and
hospitalization
provided
to
us
by
the
Public
Health
Agency
of
Lazio.
We
started
by
creating
an
origin–destination
matrix
of
hospitals
by
LHUs
of
size
91
×
12
recording
the
number
of
patients
that
every
row-hospital
i
(destination)
receives
from
every
column-LHU
k
(origin).
Following
Sohn
(2001)
the
intensity
of
com-
petition
that
hospital
i
receives
from
hospital
j
is
then
computed
as:
ωij =kWik min(pik,
pjk)
kWikpik
(1)
where
min(pik,
pjk)
indicates
the
overlap
(or
“intersection”)
in
patient
pools
between
hospital
i
and
hospital
j
in
LHU
k,
and
the
weight
wik indicates
the
proportion
of
all
patients
admitted
to
hos-
pital
i
who
come
from
LHU
k.
In
Eq.
(1)
the
numerator
expresses
the
overall
sum
of
niche
overlaps
between
hospital
i
and
j
across
all
LHUs
(market
segments),
while
the
denominator
simply
tells
the
1The
computer
output
that
we
obtained
by
using
alternative
rules
of
dichotomiza-
tion
and
alternative
models
has
been
submitted
to
the
Journal
as
an
appendix
and
may
be
obtained
from
the
authors
at
any
time
upon
request.
niche
width
of
the
i-th
hospital,
i.e.,
the
total
number
of
patients
admitted
by
hospital
i
across
all
LHUs.
The
(dyadic)
competition
coefficient,
ωij may
then
be
interpreted
as
the
proportion
of
the
patient
pool
of
a
hospital
overlapped
by
another
hospital.
The
term
min(pik,
pjk)
requires
that
ωij lies
between
0
(no
competition)
and
1
(maximum
competitive
intensity).
We
adopt
this
relational
measure
of
competition
derived
specifically
for
hospitals
as
our
measure
of
spatial
multipoint
competition
(Sohn,
2001,
2002).
This
measure
has
a
number
of
contextually
desirable
features
because
it:
(i)
reproduces
with
accuracy
essential
aspects
of
the
field
of
health
care
in
which
both
supply,
as
well
as
demand
are
highly
sensitive
to
variations
in
local
conditions
across
various
market
segments
(Gaynor
and
Haas-Wilson,
1999);
(ii)
provides
informa-
tion
on
the
strength
of
competition
at
the
dyad
level
–
the
smallest
unit
at
which
competition
can
be
measured
(Sohn,
2002;
Dexter
et
al.,
2005);
(iii)
allows
for
asymmetric
competition:
competitive
pressures
that
each
hospital
experiences
need
not
be
the
same
for
hospitals
in
a
dyad
(Sohn,
2001),
and
finally
(iv)
reflects
the
spatial
structure
of
demand
in
hospitals’
markets:
patient
pools
are
iden-
tified
on
the
basis
of
patients’
geographic
location
(LHUs).
While
specifically
developed
for
hospitals,
the
dyadic
measure
of
compe-
tition
we
adopt
is
consistent
with
other
measures
used
in
studies
of
organizational
niches
in
very
different
empirical
settings
(Podolny
et
al.,
1996).
Actor-relation
covariates.
The
effects
of
competition
are
diffi-
cult
to
observe
because
competition
is
often
indirect
(or
diffuse),
and
operates
across
multiple
levels
(Baum
and
Korn,
1996).
We
control
for
diffuse
competition
constructed
by
summing
all
com-
petition
coefficients
(ωij)
of
hospital
i
for
every
j.
The
higher
is
the
level
of
diffuse
(non-spatial)
competition
for
hospital
i,
the
stronger
will
be
the
total
competitive
pressure
to
which
hospital
i
is
exposed.
Clearly,
the
propensity
of
hospitals
to
collaborate
does
not
depend
exclusively
on
their
competitive
interdependence.
Accord-
ing
to
Ma
(1998),
for
example,
mutual
forbearance
is
more
likely
between
multipoint
competitors
with
heterogeneous
resources
and
complementary
capabilities.
We
control
for
a
variety
of
sources
of
organizational
heterogeneity
that
may
affect
the
presence
of
relations
among
multipoint
competitors.
We
include
Typology
of
assistance
to
control
for
the
effects
of
complementarity
in
the
inten-
sity
of
care
provided.
Typology
of
assistance
is
constructed
as
a
dummy
indicator
variable
taking
the
value
of
one
if
a
hospital
provides
specialized
consultative
care
(i.e.,
tertiary
care),
and
zero
otherwise
(i.e.,
secondary
care).
Other
conditions
being
equal,
we
would
expect
collaboration
to
be
more
likely
between
hospitals
offering
complementary
care.
We
include
the
Number
of
employ-
ees
to
control
for
the
possible
effect
of
differences
in
organizational
size
which
is
perhaps
the
main
source
of
structural
differentiation
between
organizations
in
general
(Blau,
1970),
and
between
hospi-
tals
in
particular
(Fennell,
1980).
For
each
hospital
in
our
sample
the
number
of
employees
includes
all
the
medical
doctors,
paramedics,
nurses,
and
administrative
staff.
We
include
the
average
percent-
age
of
beds
occupied
in
hospitals
(Occupancy
rate)
to
control
for
inter-organizational
differences
in
the
actual
utilization
of
avail-
able
capacity
(Gaynor
and
Anderson,
1995).
To
capture
differences
in
the
complexity
of
organizational
activities
we
include
the
Case
Mix
Index
(CMI)
which
is
typically
used
by
researchers,
administra-
tors,
regulators
and
policy
makers
as
an
indicator
of
the
intensity
of
hospital
complexity
and
as
an
index
of
resource
absorption
(Folland
et
al.,
1997).
We
also
control
for
differences
(or
complementarities)
in
the
typology
of
services
offered
to
patients
by
reconstructing
a
2-
mode
matrix
of
hospitals
by
the
clinical
specialties
they
contain.
We
then
computed
the
Euclidean
distance
coefficients
between
hospi-
tals
in
the
space
spanned
by
all
the
clinical
specialties.
Hospitals
that
are
farther
away
from
each
other
offer
potentially
complementary
services
and
are
more
likely
to
exchange
patients.
Other
condi-
tions
being
equal,
we
would
expect
inter-hospital
collaboration
to

Author's personal copy
106 A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111
Table
1
Descriptive
sample
statistics
(N
=
8190
dyads).
Variable
Measure
Unit
of
measurement
Mean
S.D.
Min
Max
Activity
Number
of
ties
sent
(outdegree)
Units
0.12
0.32
0
1
Multipoint
competition Patient-pools
overlap
(dyadic) Dimensionless 0.09
0.16
0
0.98
Indirect
(diffuse)
competition Sum
of
patient-pools
overlaps
Dimensionless
7.90
6.83
0.51
48.24
Typology
of
assistance
Secondary
or
tertiary
care
Binary
category
0.12
0.33
0
1
Size Number
of
employees
Units
398.25
483.90
11
2924
Available
productive
capacity
Occupancy
rate
Dimensionless
80.38
15.59
22.34
106.64
Complexity
Case
mix
index
Dimensionless
1.00
0.11
0.76
1.65
Complementarity
Euclidean
distance
Dimensionless
2.92
1.02
0
5.92
Performance Comparative
performance
index Dimensionless 1.01
0.15
0.67
1.48
Distance
Geographical
distance
Kilometers
50.77
39.50
0
222.60
Organizational
form Type
of
ownership
–
governance
structure
Dimensionless
category
–
–
1
6
LHU
membership
Shared
LHU
membership
Dimensionless
category
–
–
1
12
be
more
likely
between
hospitals
with
different
combinations
of
competences
and
resources.
We
use
the
Comparative
Performance
Index
(CPI)
as
a
general
measure
of
organizational
performance.
We
include
this
variable
to
control
for
the
possible
tendency
of
hospi-
tals
in
our
sample
to
move
patients
form
low
to
high
performance
hospitals.
In
a
recent
study,
for
example,
Iwashyna
et
al.
(2009a)
suggest
that
critically
ill
patients
should
be
systematically
trans-
ferred
to
more
capable
hospitals.
As
the
Authors
suggest
(p.
831):
“(T)ransfers
to
high-quality
hospitals
could
be
an
important
goal
when
patients’
needs
exceed
a
hospital’s
capabilities.”
The
CPI
mea-
sures
the
effectiveness
–in
terms
of
length
of
stay
–
of
a
hospital
relative
to
the
average
effectiveness
of
a
reference
set
of
hospitals
with
an
analogous
composition
of
cases
treated.
The
CPI
takes
the
value
of
1
for
organizations
whose
performance
is
aligned
with
per-
formance
of
all
other
hospitals
in
the
region.
The
CPI
takes
values
that
are
smaller
(greater)
than
one
for
hospitals
that
perform
worse
(better)
than
expected.
We
also
control
for
Geographical
distance
(in
kilometers)
between
each
hospital
in
the
sample
to
account
for
the
joint
effect
of
transportation
costs
and
clinical
risks
inherent
in
patient
transfer.
Other
conditions
being
equal,
we
would
expect
patient
transfers
to
be
more
likely
between
hospitals
that
are
closer
to
each
other.
Finally,
we
control
for
institutional
factors
that
may
influence
patterns
of
collaboration
between
hospitals.
The
variable
Organi-
zational
form
captures
the
institutional
diversity
of
hospitals
in
the
community
and
reflects
the
official
classification
adopted
by
national
health
authorities.
Organizational
form
is
a
categorical
variable
ranging
from
1
to
6
(1
=
LHU
Hospital;
2
=
Hospital
trust;
3
=
University
hospital;
4
=
Hospital
of
the
National
Institute
for
Scientific
Research;
5
=
Classified
hospital;
6
=
Private
accredited
hospital).
The
boundaries
of
these
categories
reflect
fundamen-
tal
differences
in
institutional
constraints
operating
on
hospitals,
as
well
as
broad
differences
in
forms
of
ownership
and
gover-
nance.
We
control
for
organizational
form
because
organizations
facing
similar
institutional
constraints
tend
to
be
more
similar
and
may
therefore
manage
their
dependencies
more
efficiently.
A
sec-
ond
powerful
categorical
distinction
between
hospitals
concerns
their
membership
in
the
various
LHUs.
LHU
membership
is
a
cat-
egorical
variable,
uniquely
assigning
each
hospital
to
its
reference
geographical
area
–
its
“natural”
market,
as
it
were.
As
we
discussed,
LHUs
are
also
administrative
units:
all
the
hospitals
located
in
the
same
LHU
are
coordinated
by
(and
report
to)
the
same
management
nominated
by
regional
health
authorities:
the
management
of
hos-
pitals
in
each
LHUs
is
responsible
to
the
LHU
executive
director.
Because
of
administrative,
spatial
and
financial
reasons
member-
ship
in
the
same
LHU
greatly
facilitates
exchange
relationships
between
hospitals
and
may
be
considered
also
as
a
general
proxy
for
other
forms
of
collaborations
that
we
have
not
observed
in
our
sam-
ple.
Other
conditions
being
equal,
we
would
expect
collaboration
to
be
more
likely
between
hospitals
belonging
to
the
same
LHU.
Table
1
reports
the
descriptive
statistics
of
all
the
actor-relation
variables
that
we
include
as
control
factors
in
the
empirical
model
specifications
that
we
estimate
in
the
next
section.
Given
the
prominence
of
spatial
concerns
in
our
account
of
interorganizational
network,
it
is
worth
emphasizing
that
spatial
structure
enters
our
models
through
variables
defined
along
phys-
ical
(geographical
distance
between
hospitals),
institutional
(LHU
memberships),
and
competitive
(overlap
in
patient
pools)
dimen-
sions.
The
corresponding
cofactors
are
obviously
interdependent,
but
capture
distinct
aspects
of
the
overall
space
in
which
interor-
ganizational
fields
exist.
For
example,
the
correlation
coefficient
between
geographical
distance
and
LHU
membership
is
−0.29
–
possibly
due
to
the
coexistence
of
clusters
of
LHUs
located
in
densely
packed
urban
areas
in
and
around
Rome,
and
LHUs
in
less
dense
but
wider
rural
areas.
The
correlation
between
geographi-
cal
distance
and
multimarket
contact
is
−0.24
suggesting
that
–
other
conditions
being
equal
–
competitive
pressures
generated
by
overlap
in
patient
pools
become
weaker
as
the
distance
between
hospitals
increases.
4.3.
Representing
dependencies
As
we
explain
in
the
next
section,
our
analytical
framework
is
consistent
with
the
view
that
interorganizational
networks
are
built
from
the
bottom
up
through
interdependent
acts
of
dyadic
exchange.
What
justifies
meaningful
conceptualization
of
the
relational
system
that
we
observe
as
an
interorganizational
“network”
is
the
dependence
between
hospital
dyads.
This
argu-
ment
provides
a
compelling
reason
to
adopt
innovative
Exponential
Random
Graph
Models
(ERGM)
that
specify
explicitly
local
network
dependencies
(Snijders
et
al.,
2006).
Whether
a
particular
form
of
dependence
is
present
in
a
social
system
is
typically
an
empirical
issue
but
dependencies
generated
by
relational
activities
(sending
and
receiving
ties)
and
by
triadic
closure
are
frequently
considered
defining
features
of
social
networks
(Robins
et
al.,
2005).
Building
on
Robins
et
al.
(2009)
we
include
in
our
model
specific
parameters
to
control
for
characteristics
of
the
degree
distribution,
and
for
the
tendency
toward
triadic
closure,
or
“clustering.”
To
capture
salient
features
of
the
(in
and
out)
degree
distribution
we
control
for
the
baseline
(i)
propensity
of
directed
network
ties
to
be
present
(arc),
and
(ii)
tendency
of
network
ties
to
be
reciprocated
(reciprocity).
We
also
control
for
heterogeneity
in
relational
activities
revealed
by
differences
in
the
(iii)
propensity
of
individual
organizations
to
be
selected
as
partner
by
many
others
(popularity
spread);
(iv)
propen-
sity
of
individual
organizations
to
select
multiple
others
as
partners
(activity
spread),
and
in
the
(v)
tendency
of
incoming
and
outgoing
ties
to
co-occur
(simple
connectivity).
Following
Robins
et
al.
(2009),
we
account
for
the
propensity
of
hospitals
in
our
sample
to
clus-

Author's personal copy
A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111 107
Table
2
Structural
effects.
Effects Network
substructures Network
statistics
Density 
j
yij
Reciprocity 
i<j
yijyji
Sending
and
receiving
ties
(Mixed-2-Star) 
ijk
yjiyik
Popularity
spread
(K-in-star)
n−1

k=2
(−1)kSk−in
k−2
where

=
2,
and
Sk
−in is
the
number
of
stars
with
k
edges
incident
on
the
one
node
(or
k-in-star)
Activity
spread
(K-out-Star)
n−1

k=2
(−1)kSk−out
k−2
where

=
2,
and
Sk
−
out is
the
number
of
k-out-star
Closure
n−1

k=2
(−1)kTk
k−2
where
Tkis
the
count
of
K
triangles
of
a
given
kind
(a,
b,
c
or
d)
in
the
network,
and

=
2
Generalized
Local
Multiple
Connectivity
(A2P-TDU)
n−1

k=2
(−1)kUk
k−2
where
Ukis
the
count
of
open
k-
two-paths
of
a
given
kind
(d,
e,
or
f)
in
the
network,
and

=
2
ter
by
controlling
for
tendencies
toward
(i)
transitive
closure
(path
closure);
(ii)
cyclic
closure
(generalized
exchange);
(iii)
structural
equivalence
in
outgoing
ties
(activity-based
closure);
and
(iv)
struc-
tural
equivalence
in
incoming
ties
(popularity-based
closure).
As
a
potential
counterweight
to
triadic
closure,
we
also
control
for
the
presence
of
structural
holes
in
the
form
of
multiple
open
2-paths
(multiple
connectivity).
More
specifically,
we
use
the
generalized
version
of
the
multiple
connectivity
parameter
defined
by
Robins
et
al.
(2009)
as
the
sum
of
the
network
statistics
corresponding
to
the
various
configurations
reported
in
the
last
row
of
Table
2.
Table
2
summarizes
our
discussion
on
local
dependence
struc-
tures
and
provides
the
definitions
of
the
endogenous
network
mechanisms
included
in
the
empirical
model
specification
that
we
discuss
next.
4.4.
Empirical
model
specification
and
estimation
By
considering
each
individual
network
tie
as
a
random
vari-
able,
we
link
the
dyadic
structure
of
our
data
directly
to
a
class
of
Exponential
Random
Graphs
Models
(ERGM),
also
known
as
p-star
(p*)
models
(Robins
et
al.,
2009;
Snijders
et
al.,
2006;
Wasserman
and
Pattison,
1996).
The
ERGM
used
here
has
the
general
form:
Pr(Y
=
y|X
=
x)
=1
exp 
Q
QZQ(y)
+
R
RZR(y,
x)(2)
where
(i)
Q
refers
to
possible
local
configurations
of
network
ties
(as
discussed
in
the
previous
section);
(ii)
Qis
the
parame-
ter
corresponding
to
configuration
of
type
Q;
(iii)
ZQ(y)
=yij⊂Qyij

Author's personal copy
108 A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111
Table
3
ERG
model
estimates
of
structural
and
actor-relation
effects
on
the
presence
of
network
ties
(SE
in
parentheses).
M1
M2
M3
M4
Baseline
model
parameters
Outdegree
(arc) −2.6425*(0.0503)
−1.8562*(0.0720)
−3.0923*(0.5455)
−4.3323*(0.6563)
Reciprocity 2.0150*(0.1082)
1.5998*(0.1255)
0.7598*(0.1333)
0.8076*(0.1584)
Spatial
variables
Spatial
multipoint
competition
(Patient-pools
overlap)
1.7734*(0.1680)
0.6446*(0.2121)
1.3285*(0.2967)
1.0981*(0.2404)
Geographical
distance
–
−0.0171*(0.0013)
−0.0153*(0.0015)
−0.0067*(0.0010)
LHU
membership
–
0.6909*(0.1015)
1.4222*(0.1320)
1.2624*(0.1226)
Actor-relation
effects
Diffuse
competition
(Sender
effect) –
–
−0.1955*(0.0185)
−0.1060*(0.0124)
Diffuse
competition
(receiver
effect)
–
–
−0.1162*(0.0161)
−0.0561*(0.0117)
Diffuse
competition
(homophily
effect)
–
–
0.0738*(0.0173)
0.0515*(0.0121)
Typology
of
assistance
(level
of
care)
–
–
−1.5822*(0.2324)
−1.4798*(0.2336)
Size
(N.
employees
–
homophily
effect)
–
–
0.0005*(0.0001)
0.0003*(0.0001)
Available
capacity
(occupancy
rate
–
homophily
effect)
–
–
−0.0080*(0.0029)
−0.0028*(0.0019)
Complexity
(Case
Mix
Index
– Homophily
effect)
–
–
−1.9077*(0.5271)
−0.8914*(0.4290)
Complementarity
(service
scope
–
homophily
effect)
–
–
0.0679
(0.0604)
0.0364
(0.0497)
Performance
(CPI
–
sender
effect)
–
–
1.8121*(0.3706)
0.7783*(0.2456)
Performance
(CPI
– receiver
effect)
– – 0.9177*(0.3462)
0.1846*(0.2110)
Performance
(CPI
–
homophily
effect)
–
–
−2.2772*(0.3762)
−1.1902*(0.2854)
Organizational
form
–
–
0.4692*(0.0900)
0.3935*(0.0817)
Local
dependencies
Mixed-2-Star
–
–
–
0.0260*(0.0030)
Popularity
spread
–
–
–
−0.0074
(0.1850)
Activity
spread – –
–
−0.2119
(0.2039)
Path
closure
–
–
–
0.7808*(0.1077)
Cyclic
closure –
–
–
−0.4940*(0.0276)
Activity-based
closure
–
–
–
0.3579*(0.0520)
Popularity-based
closure
–
–
–
0.3241*(0.0457)
Multiple
connectivity – – – −0.0728*(0.0141)
*p
<
0.05.
is
the
network
statistic
counting
the
frequency
of
sub-structures
Q
in
the
graph
y
(as
defined
in
Table
2),
and
(iv)

is
a
normal-
izing
quantity
included
to
ensure
that
(2)
is
a
proper
probability
distribution.
The
summation
is
taken
over
all
possible
network
con-
figurations
(Q)
included
in
a
given
model.
Finally, 
R
RZR(y,
x)
is
the
model
component
that
defines
the
actor-relation
covariates,
or
the
effects
of
the
interaction
between
network
variables
(y)
and
individual
attributes
(x)
on
the
probability
of
observing
a
tie.
The
second
summation
is
over
all
possible
configurations
R
of
ties
and
attributes.
As
it
is
usual
in
the
specification
of
ERGM,
independent
actor-relation
covariates
(summarized
in
Table
1)
enter
the
model
as
actor-relation
main
effects
(i.e.,
sender
and
receiver
effects)
as
well
as
actor-relation
homophily
(interaction)
effects
(Lusher
et
al.,
2010).
Parameter
estimation
was
conducted
with
Markov
Chain
Monte
Carlo
Maximum
Likelihood
simulation-based
techniques
(Hunter
and
Handcock,
2006;
Snijders,
2002;
Wasserman
and
Robins,
2005).
When
convergent
maximum
likelihood
parameter
esti-
mates
are
successfully
obtained,
it
is
possible
to
use
these
estimates
to
simulate
the
distribution
of
graphs
implied
by
the
model.
Any
feature
of
the
observed
graph
may
be
compared
to
the
estimated
distribution
of
that
feature
that
is
implied
by
the
model.
Goodreau
(2007)
and
Hunter
et
al.
(2008)
provide
the
statistical
argument
for
this
approach
to
comparing
structural
statistics
of
the
observed
network
with
the
corresponding
statistics
on
networks
simulated
from
the
fitted
model.
In
the
next
section
we
follow
this
simulation-
based
strategy
to
establish
the
overall
goodness
of
fit
of
our
models.
5.
Results
We
present
the
models
in
increasing
order
of
completeness.
The
first
column
in
Table
3
(M1)
reports
the
results
of
the
baseline
model
containing
only
the
effect
of
spatial
multipoint-contact
after
con-
trolling
for
basic
tendencies
to
send
and
reciprocate
network
ties.
According
to
the
estimates
in
M1
hospitals
attracting
patients
from
the
same
(multiple)
geographical
segments
of
the
market
are
signif-
icantly
more
likely
to
establish
network
ties.
Interpreted
together,
the
negative
estimate
of
the
outdegree
parameter
and
the
posi-
tive
estimate
of
the
reciprocity
parameter
suggest
that
establishing
collaborative
relations
is
costly
outside
bounds
of
reciprocity.
In
M2
we
control
for
additional
sources
of
spatial
structure
that
may
affect
the
presence
of
network
ties
like
geographical
distance
and
LHU
membership.
Transferring
patients
across
long
distances
is
costly
and
risky.
Other
conditions
being
equal,
a
hospital
is
more
likely
to
transfer
a
patient
to
a
more
proximate
partner
hospital.
Estimates
in
M2
support
this
intuition.
As
geographical
distance
increases
the
probability
of
observing
patient
transfer
relations
between
hospitals
declines
significantly.
As
expected,
membership
in
the
same
LHU
makes
inter-hospital
collaboration
more
likely
–
a
result
which
reflects
the
combined
effects
of
physical
proximity
and
administrative
dependence
implied
by
LHU
co-membership.
The
inclusion
of
additional
spatial
features
reduces
the
positive
effect
of
spatial
multipoint
contact
while
leaving
unaltered
its
statistical
significance.
In
model
M3
we
control
for
individual
differences
in
organization-specific
attributes
that
may
affect
the
tendency
of
organizations
to
collaborate.
We
add
the
sender,
receiver
and
homophily
effects
of
indirect
(or
“diffuse”)
competition.
We
do
so
to
control
for
the
level
of
non-spatial
competition
that
individ-
ual
hospitals
experience
and
that
might
affect
their
propensity
to
collaborate.
The
estimates
show
that
hospitals
exposed
to
higher
levels
of
diffuse
competition
are
less
likely
to
initiate
exchange
relations
and
to
be
selected
as
partners.
Yet,
when
the
spatial
struc-
ture
of
the
market
is
taken
into
account
hospitals
meeting
each
other
in
multiple
market
locations
are
more
likely
to
establish
network
ties.
In
M3
we
also
add
sender,
receiver
and
homophily
effects
of
organizational
performance.
We
do
so
to
control
for
the

Author's personal copy
A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111 109
Table
4
Selected
diagnostic
goodness
of
fit
statistics
(SE
in
parentheses).
Network
statistics
Actual
Restricted
model
(Col.
3
Table
3)
Full
model
(Col.
4
Table
3)
Predicted
T-statistics
Predicted
T-statistics
Arc
949
949.373(29.894)
−0.013
951.073(165.664)
−0.013
Reciprocity
188
198.158(11.513)
−0.101
188.632(52.658)
−0.012
Indegree
distribution
St.
Dev 10.037
6.999(0.336)
9.049
8.698(1.083)
1.236
Indegree
distribution
skewness 1.248
1.669(0.274)
−1.539
1.290(0.281)
−0.151
Outdegree
distribution
St.
Dev. 11.500
7.990(0.358)
9.807
8.925(1.184)
2.175
Outdegree
distribution
skewness
2.157
1.704(0.238)
1.902
1.480(0.243)
2.786
CorrCoef
in-out
degrees
0.556
0.880(0.022)
−14.916
0.786(0.078)
−2.943
Mahalanobis
distance
442.896
(196157.089)
36.599
(1339.496)
possibility
that
underperforming
hospitals
may
be
more
actively
looking
for
partners.
The
estimates
seem
to
exclude
this
possibility:
high
performing
hospitals
are
more
likely
to
initiate
collabora-
tive
relations
(sender
effect)
and
are
more
attractive
as
partners
(receiver
effect).
Because
the
largest
observed
difference
in
perfor-
mance
is
0.81
(=1.48–0.67),
the
best
performing
hospitals
in
our
sample
are
approximately
four
times
more
likely
to
be
sources
of
patient
transfer
relations
than
the
worst
performing
hospi-
tals
[exp(1.81*0.81)
=
exp(1.47)
=
4.33],
and
about
two
times
more
likely
to
be
recipients
[exp(0.92*0.81)
=
exp(0.74)
=
2.09].
Patient
mobility
is
not
explained
by
performance
differentials:
the
pres-
ence
of
network
ties
is
more
likely
to
be
observed
between
hospitals
attaining
similar
levels
of
performance
(homophily
effect
–
or
neg-
ative
effects
of
a
difference
on
the
presence
of
network
ties).
Organizations
controlling
complementary
resources
are
fre-
quently
considered
more
attractive
partners
(Gulati
and
Gargiulo,
1999).
We
account
for
the
effects
of
several
possible
sources
of
complementarity
related
to
(i)
the
quality
and
quantity
of
clini-
cal
services
offered
(typology
of
assistance
and
range
of
clinical
specialties);
(ii)
organizational
size
(number
of
employees);
(iii)
availability
of
capacity
(occupancy
rate),
and
(iv)
organizational
complexity
(CMI).
The
estimates
are
generally
consistent
with
intu-
ition.
For
example,
exchange
is
more
likely
between
hospitals
offering
different
levels
of
care
(secondary
and
tertiary
care)
and
complementary
services.
In
this
latter
case,
however,
the
corre-
sponding
effect
is
not
statistically
significant.
Hospitals
differing
in
terms
of
size
are
also
more
likely
to
collaborate,
and
so
are
hospitals
that
are
similarly
able
to
manage
their
capacity
(i.e.,
similar
occupancy
rate)
and
that
face
similar
levels
of
complex-
ity
(CMI).
Controlling
for
these
various
sources
of
complementarity
and
organization-specific
differences
leaves
unaltered
the
signifi-
cance
of
multimarket
contact.
Research
on
interorganizational
relations
in
markets
and
other
institutional
settings
reveals
that
the
formation
of
network
ties
is
affected
by
a
variety
of
local
dependencies
(Lomi
and
Pattison,
2006).
In
the
context
of
multipoint
competition,
one
way
to
interpret
such
local
dependencies
is
as
outcomes
of
attempts
to
coordinate
action
across
multiple
markets
(Fuentelsaz
and
Gómez,
2006).
In
column
M4
of
Table
3
we
control
for
endogenous
struc-
tural
tendencies
within
the
network
that
may
confound
the
casual
relation
between
organization-specific
attributes
and
the
tendency
of
organizations
to
establish
network
ties.
Because
organizations
display
a
strong
tendency
toward
forming
ties
with
their
partner’s
partners,
dependencies
induced
by
closure
mechanisms
have
been
of
particular
interest
to
students
of
interorganizational
networks
(Gulati
and
Gargiulo,
1999).
We
find
a
clear
tendency
toward
gen-
eralized
path-closure:
organizations
connected
only
indirectly
by
multiple
independent
two-paths
(i.e.,
organizations
sharing
mul-
tiple
partners)
are
more
likely
to
collaborate.
We
also
find
that
collaboration
is
significantly
more
likely
between
organizations
that
are
structurally
equivalent
(both
in
outgoing
and
incom-
ing
ties).
We
consider
these
results
consistent
with
our
main
theoretical
argument:
organizations
sharing
common
resource
dependencies
are
more
likely
to
collaborate.
In
this
case
the
rele-
vant
resources
are
capacity
and
knowledge
of
partners
rather
than
patients.
The
negative
and
significant
effect
of
multiple
connectivity
strengthens
our
conclusions:
the
parameter
estimate
correspond-
ing
to
multiple
(open)
2-paths
is
negative
indicating
a
tendency
of
the
multiple
triangles
to
close
at
the
base.
The
effect
of
cyclic
closure
is
significantly
negative,
suggesting
that
exchange
with-
out
bonds
of
direct
reciprocity
(generalized
exchange)
is
unlikely
to
occur
between
the
hospitals
in
our
sample.
Finally,
we
include
the
effects
of
popularity
(in
the
form
of
multiple
in-stars),
and
activity
(multiple
out-stars)
to
control
for
factors
that
may
con-
tribute
to
the
“spread”
of
the
indegree
and
outdegree
distributions
in
our
network
(Robins
et
al.,
2009).
In
each
case
the
correspond-
ing
estimates
are
negative
but
well
below
significance.
Considered
together,
these
results
suggest
the
presence
of
a
core-periphery
structure
that
is
sustained
by
the
closure
of
multiple
triangle
simul-
taneously
rather
than
by
underlying
differences
in
popularity
and
activity.
Controlling
for
local
dependencies
does
not
change
the
qualitative
conclusions
that
we
have
reached
about
the
effect
of
spatial
multipoint
contact
on
the
presence
of
network
ties.
Con-
ditional
on
the
rest
of
the
model,
hospitals
whose
niches
overlap
almost
completely
are
approximately
three
times
more
likely
to
be
related
by
network
ties
[(exp(0.98*1.098)
=
2.93].
How
well
does
the
model
implied
by
the
estimates
reproduce
salient
global
features
of
the
network?
The
results
of
our
attempt
to
address
this
question
are
summarized
by
the
goodness
of
fit
diagnostics
reported
in
Table
4.
We
sampled
2000
digraphs
from
the
distribution
of
random
digraphs
simulated
on
the
basis
of
the
estimates.
A
t-statistic
sum-
marizes
the
location
of
the
observed
feature
in
this
distribution.
For
effects
not
included
in
the
model
specification,
a
t-statistic
exceed-
ing
2
in
absolute
value
suggest
that
the
observed
graph
differs
from
the
distribution
implied
by
the
models
in
the
corresponding
fea-
ture.
For
effects
included
in
the
model,
a
t-statistic
below
0.1
is
typically
taken
as
evidence
of
reliability
(Robins
et
al.,
2009).
In
almost
every
case,
the
center
of
the
simulated
distributions
for
the
various
network
statistics
is
close
to
the
observed
mean,
suggest-
ing
that
the
model
implied
by
the
estimates
is
consistent
with
the
observations.
In
almost
every
case
the
full
model
accounting
for
local
dependencies
(M4)
outperforms
the
restricted
model
based
on
assumption
of
dyadic
independence
(M3).
The
only
exception
is
represented
by
the
skewness
of
the
out-degree
distribution.
The
model
also
does
not
seem
to
capture
particularly
well
the
correla-
tion
between
in
and
outdegrees
present
in
the
actual
network.
6.
Discussion
and
conclusions
Like
other
forms
of
collective
action,
organizations
may
be
understood
only
with
reference
to
their
location
in
social
and
phys-
ical
spaces
(Abbott,
1997).
Both
types
of
location
matter
because

Author's personal copy
110 A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111
they
induce
local
dependencies
that
affect
the
global
structure
of
interorganizational
networks,
communities,
and
fields
(Lomi
and
Pattison,
2006;
Laumann
and
Marsden,
1982;
Whittington
et
al.,
2009).
Perhaps
no
area
of
contemporary
organizational
research
has
benefitted
more
directly
from
this
broad
sociological
insight
than
the
study
of
multipoint
contact
where
market
structure
is
viewed
as
arising
precisely
from
local
interorganizational
depen-
dencies
of
this
sort.
The
interdependence
of
social
and
physical
dimensions
of
organizational
spaces
is
revealed
clearly
by
our
study.
In
models
controlling
for
local
network
structure
(M4),
for
example,
the
negative
effect
of
geographical
distance
on
the
presence
of
network
ties
drops
by
more
than
50
percent
when
compared
to
models
that
do
not
incorporate
elements
of
local
network
structure
(M3).
Thus
it
seems
that
the
local
dependence
structures
in
which
organizations
are
embedded
greatly
reduce
the
(negative)
impact
of
physical
distance
on
the
presence
of
net-
work
ties.
One
general
conclusion
that
this
result
supports
is
that
network
dependencies
interact
with
elements
of
organizational
environments
blurring
the
boundaries
between
exogenous
and
endogenous
mechanisms
of
tie
formation.
The
findings
of
our
study
provide
more
contextual
support
for
the
strong
version
of
the
mutual
forbearance
hypothesis
between
spatial
multipoint
competitors:
hospitals
interacting
in
multiple
geographical
segments
of
their
market
are
significantly
more
likely
to
engage
in
direct
collaboration.
This
result
is
qualitatively
consis-
tent
with,
but
also
goes
beyond
prior
studies
of
mutual
forbearance.
Extant
research
has
focused
almost
exclusively
on
possible
con-
sequences
of
collaboration
induced
by
competition
in
multiple
geographical
markets.
We
have
taken
a
different
perspective
and
concentrated
on
the
presence
of
actual
network
ties,
rather
than
on
some
of
their
possible
organization-level
consequences.
We
have
shown
that
the
network
structure
implied
by
our
empirical
estimates
reproduces
with
accuracy
salient
aspects
of
the
overall
network
in
which
hospitals
in
our
sample
are
embedded.
The
results
that
we
have
reported
add
at
least
three
important
nuances
to
our
general
understanding
of
how
competition
shapes
relations
between
organizations.
First,
we
have
shown
that
tangible
network
ties
are
more
likely
between
organizations
meeting
each
others
in
multiple
locations.
Consistent
with
Simmel
(1950)
con-
jecture,
competition
does
seem
to
facilitate
social
bonding
between
antagonists
at
the
dyadic
level,
but
not
at
the
aggregate
level.
Indi-
vidual
dyads
and
global
network
structures
are
linked
by
complex
chains
of
local
dependencies.
Characterizing
such
dependencies
accurately
is
important
because
they
reveal
basic
principles
of
net-
work
organization
(Laumann
and
Marsden,
1982;
Robins
et
al.,
2005).
We
have
shown
that
in
models
where
these
principles
are
appropriately
specified,
diffuse
competitive
uncertainty
per
se
does
not
affect
organizations’
social
orientation.
As
Baum
and
Korn
(1996)
observed,
the
paradoxical
consequence
of
this
decoupling
of
structural
levels
is
that
competition
at
one
level
of
aggregation
(overall
network
level
in
our
case)
may
coexist
with
collaboration
at
another
(dyadic
level
in
our
case).
This
result
gives
strength
to
the
view
of
economic
competition
as
a
located
fact
–
a
fact
situated
in
social
time
and
space
(Abbott,
1997).
Interorganizational
relations
that
multipoint
contact
facilitates
obey
to
structural
logics
associated
with
coherent
network
struc-
tures
–
a
result
revealed
by
the
part
of
our
models
that
specifies
endogenous
local
dependencies.
This
is
the
second
important
detail
that
our
study
adds
to
our
general
understanding
of
collaboration
between
multipoint
competitors.
Our
estimates
reveal
a
significant
tendency
of
multiple
two-paths
to
close
simultaneously,
suggest-
ing
a
tendency
for
the
overall
network
toward
a
core-periphery
structure
with
multiple
triads
appearing
in
clusters
(Robins
et
al.,
2009).
This
result
is
corroborated
further
by
the
significant
ten-
dency
that
we
detected
against
multiple
open
two-paths.
The
significantly
negative
estimates
of
parameters
associated
with
gen-
eralized
exchange
mechanisms
(cyclic
closure)
suggest
a
tendency
of
relations
between
hospitals
in
our
sample
to
take
on
a
hierarchi-
cal
character
(Lazega
and
Pattison,
1999;
Robins
et
al.,
2007).
We
note,
in
passing,
that
none
of
these
results
could
have
been
obtained
in
the
context
of
modeling
frameworks
that
assume
dyadic
inde-
pendence
among
the
observations.
On
a
more
substantive
note,
the
presence
of
local
networks
dependencies
that
our
models
document
also
suggests
that
the
effects
of
competition
in
one
local
market
do
not
depend
exclusively
on
demand
and
supply
conditions
that
are
specific
to
that
market.
The
third
implication
of
our
results,
therefore,
is
that
the
way
in
which
organizations
react
to
competition
depends
on
how
mar-
ket
segments,
or
niches,
are
themselves
connected.
This
conclusion
invites
reconsideration
of
organizations
and
markets
not
as
sepa-
rate
entities
or
levels
of
analysis,
but
rather
as
entities
connected
by
a
relation
of
mutual
constitution
(Breiger,
2002).
Organizations
define
their
identity
in
terms
of
the
market
segments
in
which
they
are
present,
but
identities
of
market
segments
are
defined
in
terms
of
the
organizations
they
contain.
Relational
settings,
or
“[P]laces
and
times
in
which
actors
meet”
(Sorenson
and
Stuart,
2008:
266),
arise
from
the
fundamental
duality
of
organizations
and
markets
(Breiger,
1974;
White
and
Eccles,
1987).
Our
study
illustrates
how
geographical
and
social
positions
that
organizations
come
to
occupy
in
such
settings
affect
the
presence
of
network
ties
between
competitors
and
hence
the
structure
of
interorgani-
zational
fields.
To
reduce
the
risk
of
over-interpreting
our
results
it
is
perhaps
useful
to
reflect
on
five
main
limitations
of
our
current
attempt
to
understand
how
multipoint
contact
affects
the
propensity
of
com-
peting
organizations
to
collaborate.
First,
we
have
concentrated
on
one
possible
dimension
of
multipoint
contact:
joint
presence
in
the
same
geographical
niches.
But
organizations
may
meet
each
other
in
a
space
with
multiple
dimensions
that
may
be
spanned,
for
example,
by
products,
projects,
price
levels,
and
characteristics
of
customers
(Baum
and
Haveman,
1997;
Vonortas,
2000).
Each
one
dimension
may
be
viewed
as
a
relevant
setting
capable
of
generating
dependencies,
and
hence
network
relations
between
participants.
Which
one,
or
which
combination,
of
these
possi-
ble
settings
is
most
relevant
to
understand
collaboration
between
competitors
remains
an
empirical
question
that
our
study
leaves
completely
open.
Second,
our
design
is
static.
As
a
consequence
we
are
unable
to
ascertain
whether
collaboration
between
competi-
tors
that
we
have
observed
is
an
outcome
of
multipoint
contact,
or
multimarket
contact
is
strategically
sought
by
organizations
that
are
linked
by
bonds
of
collaboration
(Korn
and
Baum,
1999).
Third,
we
have
selected
for
analysis
one
specific
relation.
While
our
field
experience
makes
us
confident
that
this
relation
captures
impor-
tant
dimensions
of
collaboration
between
hospitals,
it
is
possible
that
hospitals
collaborate
in
other
ways
as
well.
Exchange
of
doc-
tors,
cross
training
of
medical
staff
and
technology
transfer
all
come
to
mind
as
possible
relational
context
that
may
reveal
and
at
the
same
sustain
collaboration
between
hospitals.
Future
studies
will
have
to
pay
attention
to
the
multiplexity
that
interorganizational
collaboration
is
likely
to
involve
(Lomi
and
Pattison,
2006).
Fourth,
we
took
careful
precautions
to
ensure
the
robustness
of
our
con-
clusions
with
respect
to
alternative
ways
to
construct
of
our
basic
dependent
variable.
However,
we
cannot
rule
out
the
possibil-
ity
that
studies
of
interorganizational
relations
adopting
our
same
modeling
strategy
will
suffer
from
loss
of
information
about
the
variance
of
valued
network
ties.
For
this
reason
we
think
that
future
advances
in
the
specification
and
estimation
of
ERGM
will
have
to
dedicate
specific
attention
to
the
analysis
of
integer
and
real-valued
networks.
Fifth,
and
in
conclusion,
we
concede
that
hospitals
may
hardly
be
considered
as
a
random
sample
of
the
organizational
world
at
large,
and
that
some
may
view
patient
transfer
relations
as
unrepresentative
of
interorganizational
relations.
Yet,
under-

Author's personal copy
A.
Lomi,
F.
Pallotti
/
Social
Networks
34 (2012) 101–
111 111
standing
how
interaction
across
multiple
physical
and
social
spaces
affects
patterns
of
collaboration
between
competitors
remains
a
point
of
general
relevance
and
convergent
interest
for
students
of
organizations
and
social
networks.
References
Abbott,
A.,
1997.
Of
time
and
space.
Social
Forces
75,
1149–1182.
Baker,
W.E.,
Faulkner,
R.R.,
1993.
The
Social
organization
of
conspiracy:
illegal
net-
works
in
the
heavy
electrical
equipment
industry.
American
Sociological
Review
58,
837–860.
Barnett,
W.P.,
1993.
Strategic
deterrence
among
multipoint
competitors.
Industrial
and
Corporate
Change
2,
249–278.
Baum,
J.A.C.,
Haveman,
H.A.,
1997.
Love
thy
neighbor?
Differentiation
and
agglom-
eration
in
the
Manhattan
hotel
industry,
1898–1990.
Administrative
Science
Quarterly
42,
304–338.
Baum,
J.A.C.,
Korn,
H.J.,
1996.
Competitive
dynamics
of
interfirm
rivalry.
Academy
of
Management
Journal
39,
255–291.
Baum,
J.A.C.,
Singh,
J.V.,
1994.
Organizational
niches
and
the
dynamics
of
organiza-
tional
mortality.
American
Journal
of
Sociology
100,
346–380.
Bernheim,
B.D.,
Whinston,
M.D.,
1990.
Multimarket
contact
and
collusive
behavior.
Rand
Journal
of
Economics
21,
1–26.
Blau,
P.,
1970.
A
formal
theory
of
differentiation
in
organizations.
American
Socio-
logical
Review
35,
201–218.
Boeker,
W.,
Goodstein,
J.,
Stephan,
J.,
Murmann,
J.P.,
1997.
Competition
in
a
multi-
market
environment:
the
case
of
market
exit.
Organizational
Science
8,
126–142.
Breiger,
R.L.,
2002.
Poststructuralism
in
organizational
studies.
In:
Lounsbury,
M.,
Ventresca,
M.J.
(Eds.),
Social
Structure
and
Organizations
Revisited.
CT:
JAI
Press/Elsevier
Science,
Greenwich,
pp.
295–305.
Breiger,
R.L.,
1974.
The
duality
of
persons
and
groups.
Social
Forces
53,
181–190.
Dexter,
F.,
Wachtel,
R.E.,
Sohn,
M.-W.,
Ledolter,
J.,
Dexter,
E.U.,
Macario,
A.,
2005.
Quantifying
effect
of
a
hospital’s
caseload
for
a
surgical
specialty
on
that
of
another
hospital
using
multi-attribute
market
segments.
Health
Care
Manage-
ment
Science
8,
121–131.
Dranove,
D.,
White,
W.D.,
1994.
Recent
theory
and
evidence
on
competition
in
hos-
pital
markets.
Journal
of
Economics
and
Management
Strategy
3,
169–209.
Edwards,
C.D.,
1955.
Conglomerate
Bigness
as
a
Source
of
Power,
in
the
National
Bureau
of
Economic
Research
conference
report,
Business
Concentration
and
Price
Policy.
Princeton
University
Press,
Princeton,
NJ,
pp.
331–352.
Fennell,
M.L.,
1980.
The
effects
of
environmental
characteristics
on
the
structure
of
hospital
clusters.
Administrative
Science
Quarterly
25,
485–510.
Folland,
S.,
Goodman,
A.,
Stano,
M.,
1997.
The
Economics
of
Health
and
Health
Care,
2nd
ed.
Prentice
Hall,
Upper
Saddle
River,
NJ.
Fuentelsaz,
L.,
Gómez,
J.,
2006.
Multipoint
competition,
strategic
similarity
and
entry
into
geographic
markets.
Strategic
Management
Journal
27,
477–499.
Garnick,
D.W.,
Luft,
H.S.,
Robinson,
J.C.,
Tetreault,
J.,
1987.
Appropriate
measures
of
hospital
market
areas.
Health
Services
Research
22,
69–89.
Gaynor,
M.,
Anderson,
G.F.,
1995.
Uncertain
demand,
the
structure
of
hospital
costs,
and
the
cost
of
empty
hospital
beds.
Journal
of
Health
Economics
14,
291–317.
Gaynor,
M,
Haas-Wilson,
D.,
1999.
Change,
consolidation,
and
competition
in
health
care
markets.
Journal
of
Economic
Perspectives
13,
141–164.
Gaynor,
M.,
Vogt,
W.B.,
2003.
Competition
among
hospitals.
RAND
Journal
of
Eco-
nomics
34,
764–785.
Gimeno,
J.,
1999.
Reciprocal
threats
in
multimarket
rivalry:
staking
out
spheres
of
influence
in
the
US
airline
industry.
Strategic
Management
Journal
20,
101–128.
Gimeno,
J.,
Woo,
C.,
1999.
Multimarket
contact,
economies
of
scope,
and
firm
per-
formance.
Academy
of
Management
Journal
42,
239–259.
Gittell,
J.H.,
2002.
Coordinating
mechanisms
in
care
provider
groups:
relational
coor-
dination
as
a
mediator
and
input
uncertainty
as
a
moderator
of
performance
effects.
Management
Science
48,
1408–1426.
Goodreau,
S.M.,
2007.
Advances
in
Exponential
Random
Graph
(p*)
Models
applied
to
a
large
social
network.
Social
Networks
29,
231–248.
Greve,
H.R.,
2006.
The
intent
and
extent
of
multimarket
contact.
Strategic
Organi-
zation
4,
249–274.
Gulati,
R.,
Gargiulo,
M.,
1999.
Where
do
interorganizational
networks
come
from?
American
Journal
of
Sociology
104,
1439–1493.
Haveman,
H.A.,
Nonnemaker,
L.,
2000.
Competition
in
multiple
geographic
markets:
the
impact
of
growth
and
market
entry.
Administrative
Science
Quarterly
45,
232–267.
Hunter,
D.,
Goodreau,
S.,
Handcock,
M.,
2008.
Goodness
of
fit
of
social
network
models.
Journal
of
the
American
Statistical
Association
103,
248–258.
Hunter,
D.R.,
Handcock,
M.S.,
2006.
Inference
in
curved
exponential
family
models
for
networks.
Journal
of
Computational
and
Graphical
Statistics
15,
565–583.
Ingram,
P.,
Yue,
L.Q.,
2008.
Structure,
affect
and
identity
as
bases
of
organizational
competition
and
cooperation.
The
Academy
of
Management
Annals
2,
275–303.
Iwashyna,
T.J.,
Christie,
J.D.,
Kahn,
J.M.,
David,
A.,
2009a.
Uncharted
paths:
hospital
networks
in
critical
care.
Chest
135,
827–833.
Iwashyna,
T.J.,
Jason,
D.C.,
Moody,
J.,
Kahn,
J.M.,
Asch,
D.A.,
2009b.
The
structure
of
critical
care
transfer
networks.
Medical
Care
47,
787–793.
Kessler,
D.P.,
McClellan,
M.,
2000.
Is
hospital
competition
socially
wasteful?
The
Quarterly
Journal
of
Economics
115,
577–615.
Korn,
H.J.,
Baum,
J.A.C.,
1999.
Chance,
imitative
and
strategic
antecedents
to
multi-
market
contact.
Academy
of
Management
Journal
42,
171–193.
Krugman,
P.,
1991.
Geography
and
Trade.
MIT
Press/Leuven
UP,
London.
Laumann,
P.,
Marsden,
P.V.,
1982.
Microstructural
analysis
of
interorganizational
systems.
Social
Networks
4,
329–348.
Lazega,
E.,
Pattison,
P.,
1999.
Multiplexity,
generalized
exchange
and
cooperation
in
organizations.
Social
Networks
21,
67–90.
Levine,
S.,
White,
P.E.,
1961.
Exchange
as
a
conceptual
framework
for
the
study
of
interorganizational
relationships.
Administrative
Science
Quarterly
5,
583–
601.
Lomi,
A.,
Pattison,
P.,
2006.
Manufacturing
relations:
an
empirical
study
of
the
organization
of
production
across
multiple
networks.
Organization
Science
17,
313–332.
Lusher,
D.,
Koskinen,
J.,
Robins,
G.
(Eds.),
2010.
Exponential
Random
Graph
Models
for
Social
Networks:
Theories,
Models
and
Applications.
Cambridge
University
Press,
New
York
(Forthcoming).
Ma,
H.,
1998.
Mutual
forbearance
in
international
business.
Journal
of
International
Management
4,
129–147.
Mizruchi,
M.S.,
Marquis,
C.,
2006.
Egocentric,
sociocentric,
or
dyadic?
Identifying
the
appropriate
level
of
analysis
in
the
study
of
organizational
networks.
Social
Networks
28,
187–208.
Phibbs,
C.S.,
Robinson,
J.C.,
1993.
A
variable-radius
measure
of
local
hospital
market
structure.
Health
Services
Research
28,
313–324.
Podolny,
J.M.,
Stuart,
T.E.,
Hannan,
M.H.,
1996.
Networks,
knowledge,
and
niches:
competition
in
the
worldwide
semiconductor
industry,
1984–1991.
American
Journal
of
Sociology
102,
659–689.
Popielarz,
P.A.,
Neal,
Z.P.,
2007.
The
niche
as
theoretical
tool.
Annual
Review
of
Sociology
33,
65–84.
Prince,
J.T.,
Simon,
D.H.,
2009.
Multimarket
contact
and
service
quality:
evidence
from
on-time
performance
in
the
U.S.
airline
industry.
Academy
of
Management
Journal
52,
336–354.
Robins,
G.L.,
Pattison,
P.,
Woolcock,
J.,
2005.
Small
and
other
worlds:
global
network
structures
from
local
processes.
American
Journal
of
Sociology
110,
894–936.
Robins,
G.,
Pattison,
P.,
Wang,
P.,
2009.
Closure,
connectivity,
and
degree
distri-
butions:
Exponential
Random
Graph
(p*)
Models
for
directed
social
networks.
Social
Networks
31,
105–117.
Robins,
G.L.,
Snijders,
T.A.B.,
Wang,
P.,
Handcock,
M.,
Pattison,
P.,
2007.
Recent
devel-
opments
in
Exponential
Random
Graph
(p*)
Models
for
social
networks.
Social
Networks
29,
192–215.
Ruef,
M.,
Scott,
W.R.,
1998.
A
multidimensional
model
of
organizational
legitimacy:
hospital
survival
in
changing
institutional
environments.
Administrative
Science
Quarterly
43,
877–904.
Scott,
W.R.,
Meyer,
J.W.,
1983.
The
organization
of
societal
sectors.
In:
Meyer,
J.W.,
Scott,
W.R.
(Eds.),
Organizational
Environments:
Rituals
and
Rationality.
Sage,
Newbury
Park,
CA,
pp.
129–153.
Simmel,
G.,
1950.
Superordination
and
subordination
and
degrees
of
domination
and
freedom.
In:
Wolff,
K.
(Ed.),
The
Sociology
of
Georg
Simmel.
Free
Press,
New
York,
pp.
268–303.
Snijders,
T.A.B.,
2002.
Markov
chain
Monte
Carlo
estimation
of
exponential
random
graph
models.
Journal
of
Social
Structure
3,
1–41.
Snijders,
T.A.B.,
Pattison,
P.E.,
Robins,
G.L.,
Handcock,
M.S.,
2006.
New
specifications
for
Exponential
Random
Graph
Models.
Sociological
Methodology
36,
99–153.
Sohn,
M.-W.,
2002.
A
relational
approach
to
measuring
competition
among
hospi-
tals.
Health
Services
Research
37,
457–483.
Sohn,
M.-W.,
2001.
Distance
and
cosine
measures
of
niche
overlap.
Social
Networks
23,
141–165.
Sorenson,
O.,
Stuart,
T.E.,
2008.
Bringing
the
context
back
in:
settings
and
the
search
for
syndicate
partners
in
venture
capital
investing.
Administrative
Science
Quar-
terly
53,
266–294.
Trapido,
D.,
2007.
Competitive
embeddedness
and
the
emergence
of
interfirm
coop-
eration.
Social
Forces
86,
165–191.
Ugolini,
C.,
2001.
Mobilità
ospedaliera
e
programmazione
regionale:
alcune
rifles-
sioni.
Keiron
5,
104–115.
Van
Raak,
A.,
Paulus,
A.,
Mur-Veeman,
I.,
2005.
Why
do
health
and
social
care
providers
co-operate?
Health
Policy
74,
13–23.
Vonortas,
G.,
2000.
Multimarket
contact
and
inter-firm
cooperation
in
R&D.
Journal
of
Evolutionary
Economics
10,
243–271.
Wasserman,
S.,
Pattison,
P.E.,
1996.
Logit
models
and
logistic
regressions
for
social
networks:
an
Introduction
to
Markov
Graphs
and
p*.
Psychometrika
61,
401–425.
Wasserman,
S.,
Robins,
G.L.,
2005.
An
introduction
to
random
graphs,
dependence
graphs,
and
p*.
In:
Carrington,
P.J.,
Scott,
J.,
Wasserman,
S.
(Eds.),
Models
and
Methods
in
Social
Network
Analysis.
Cambridge
University
Press,
pp.
148–161.
White,
H.C.,
1988.
Varieties
of
markets.
In:
Wellman,
B.,
Berkowitz,
S.D.
(Eds.),
Social
Structures:
A
Network
Approach.
Cambridge
University
Press,
pp.
226–
260.
White,
H.C.,
Eccles,
R.G.,
1987.
Producers’
markets.
In:
Eatwell,
J.,
Milgate,
M.,
New-
man,
P.
(Eds.),
The
New
Palgrave:
A
Dictionary
of
Economics.
Stockton
Press,
New
York,
pp.
984–986.
Whittington,
K.B.,
Owen-Smith,
J.,
Powell,
W.W.,
2009.
Networks,
propinquity,
and
innovation
in
knowledge-intensive
industries.
Administrative
Science
Quarterly
54,
90–122.