The
Methodologies
of
Neuroeconomics
 Glenn
Harrison
 University
of
Central
Florida
 gharrison@research.bus.ucf.edu
 Don
Ross
 University
of
Cape
Town
&
University
of
Alabama
at
Birmingham
 don.ross@uct.ac.za
 Neuroeconomics
has
arisen
quickly
as
a
subfield
within
economics,
but
does
not
 exhibit
a
unified
methodology.
The
most
important
distinction
is
between
what
Ross
 (2008)
calls
“neurocellular
economics”
(NE)
and
“behavioral
economics
in
the
 scanner”
(BES).
Along
with
descriptive
statements
of
what
those
methodologies
 entail,
and
evaluations
of
some
critiques
of
these
methodologies,
our
assessment
 suggests
critical
points
in
their
application.
 1.
The
Methodological
Backlash
Against
Neuroeconomics
 For
about
a
century
after
1870,
a
dominant
methodological
trend
in
economic
 theory
was
the
progressive
severance
of
direct
psychological
commitments.
This
 process
involved
three
principal
milestones:
 • • The
generalization
of
all
forms
of
value
by
reference
to
an
abstract
concept
of
 utility.
 The
recognition
that
convexity
of
demand
need
not
be
grounded
in
a
 psychological
principle
of
diminishing
marginal
satisfaction,
but
could
be
 derived
from
anodyne
assumptions
about
substitutability
and
budget
 constraints.
 The
popularity
of
revealed
preference
theory,
which
is
ironically
about
the
 elimination
of
preferences,
conceived
as
latent
states
of
mind,
in
favor
of
 attention
to
the
logical
consistency
of
observable
choices.

    
    •
    
    In
light
of
this
history,
it
would
be
surprising
if
most
economists
were
prepared
to
 incautiously
swallow
the
suggestion
that
we
can
do
better
economics
by
examining
 people’s
brains.
And
so
it
is
not
surprising
that
the
appearance
of
several
brash,
 well‐funded,
and
sometimes
rather
messianic1
research
programs
trading
under
the
 banner
of
“neuroeconomics”
has
begun
to
provoke
a
backlash.
Harrison
(2008a)
and
 Bernheim
(2009)
both
concede
possible,
albeit
modest,
potential
contributions
that
 brain
studies
could
offer
to
economics.
Gul
and
Pesendorfer
(2008)
(GP)
concede
 nothing
at
all;
they
thereby
usefully
define
an
extreme
limit
point
with
respect
to
the
 























































 1
Witness
the
readiness
of
some
neuroeconomists,
especially
those
not
trained
as
 economists,
to
blithely
draw
revolutionary
conclusions
from
small
neuroimaging
 studies.
For
example,
Knutson
et
al
(2007)
consider
fMRI
data
from
26
subjects
 choosing
purchases,
and
on
this
basis
announce
that
economists
have
been
 mistaken
in
modeling
consumers
as
minimizing
opportunity
costs.
 1
 

    
    economist’s
assessment
of
neuroeconomics,
against
which
less
uncompromising
 positions
can
be
efficiently
developed.

 According
to
GP,
psychological
hypotheses
and
empirical
findings
might
usefully
 “inspire”
economists’
modeling
ideas,
just
as
ruminations
of
philosophers
or
 novelists
might
usefully
inspire
them,
but
economic
models
should
include
only
 variables
that
condition
what
an
agent
chooses
and
none
that
condition
how
an
 agent
chooses.
This
is
because
the
task
of
positive
economics
is
to
predict
choices
as
 functions
of
changes
in
incentives
and
opportunity
sets.
GP
do
not
intend
this
claim
 as
an
a
priori
restriction
derived
from
a
transcendental
insight
into
the
true
domain
 of
economics.
Instead,
they
believe
it
to
be
a
sociological
fact
that
most
economists
 are
not
professionally
concerned
with
variables
for
psychological
processes,
and
 will
continue
to
maintain
this
attitude
because
the
generalizations
they
seek
about
 the
influences
of
incentives
and
opportunities
on
choices
are
not
sensitive
to
 differences
in
such
variables.
 GP
presumably
do
not
think
that
choices
are
structureless
Sartrean
actes.
They
no
 doubt
suppose
that
they
are
computational
processes
of
some
sort.
They
clearly
 agree,
as
mainstream
economists,
that
the
processes
in
question
are
in
some
way
 conditioned
on
underlying
valuations
(which
may
themselves
have
resulted
from
 prior
choices).
Their
claim
is
simply
that
the
economist
does
and
should
leave
the
 processing
details
inside
black
boxes.
 Thus
a
model
that
relates
valuations
and
opportunities
to
outputs
of
choice
 processes
is
a
description
of
a
class
of
computations,
but
in
reduced
form.
Harrison
 (2008a)
points
out
that
GP’s
examples
imply
a
suggestion
that
all
economic
models
 should
be
exclusively
given
in
reduced
form,
and
should
never
be
structural.
This,
he
 observes,
is
something
one
could
only
imagine
at
great
theoretical
distance
from
the
 actual
testing
of
models
against
empirical
evidence,
particularly
experimental
 evidence.
As
the
cost
of
computing
power
has
fallen,
structural
models
have
become
 increasingly
common
in
all
areas
of
economics
except
the
most
airless
reaches
of
 theory;
this
has
in
turn
driven
a
major
surge
of
innovation
in
econometric
 techniques.
GP
cannot
deny
the
relevance
of
structural
modeling
to
economics
 unless
they
are
prepared
to
trade
in
their
empirical
sociological
standpoint
for
a
 metaphysical
one,
and
we
are
confident
that
they
would
regard
that
as
a
cure
worse
 than
the
disease.
 In
fact,
GP
allow
that
economic
models
should
include
variables
that
constrain
 opportunity
sets,
such
as
interest
rates.
But
an
interest
rate
is
a
kind
of
processing
 state
variable,
describing
trajectories
of
expected
future
transactions,
but
which
 aggregates
the
choices
of
many
agents.
What
GP
consider
alien
to
economics
about
 neural
(and
other
psychological)
processing
states,
therefore,
must
not
be
that
they
 are
processing
states,
but
that
they
are
internal
to
the
agent.
The
black
box
that
the
 economist
is
specially
forbidden
to
open
is
in
fact
the
skull
cavity.
 Taken
as
a
general
principle,
it
is
hard
to
understand
this
as
more
than
a
prejudice.
 In
particular,
it
offers
no
sound
basis
for
denying
the
validity
of
the
NE
branch
of
 2
 

    
    neuroeconomics.
This
program
is
identified
with
Glimcher
(2003)
and
colleagues,
 and
uses
the
technical
resources
of
economic
theory
to
understand
the
mechanisms
 by
which
brains
comparatively
evaluate
alternative
possible
states
that
share
no
 common
dimensions
describable
in
terms
of
sensory
modalities
or
neurochemical
 response
profiles.
Glimcher
(2009)
calls
this
“subjective
value,”
though
we
suggest
 that
“value
in
the
brain”
would
invite
fewer
misinterpretations.
The
program’s
 viability
and
potential
importance
rest
on
an
empirical
hypothesis
that
dopamine
 signals
in
the
ventral
striatum
and
medial
prefrontal
cortex
constitute
a
“common
 currency”
of
reward
that
has
many
properties
in
common
with
the
mainstream
 economist’s
concept
of
utility.
If
this
hypothesis
is
correct,
then
neuroscientists
 might
usefully
exploit
a
century’s
progress
by
economists
in
studying
utility
to
 model
valuation
in
the
brain.
 We
see
no
reason
why
GP
shouldn’t
acknowledge
the
potential
of
this
project.
Their
 claims,
after
all,
concern
what
is
and
isn’t
relevant
to
study
of
the
traditional
domain
 of
economics.
There
is
no
reason
why
they
should
resent
the
borrowing
of
economic
 theory
to
model
other
domains.
Presumably,
for
example,
they
don’t
want
to
ban
 ecologists
from
using
cost‐benefit
analysis.
Of
course,
one
might
follow
up
this
point
 by
asking
why,
if
economic
theory
should
turn
out
to
be
the
main
modeling
 technology
for
the
neural
reward
circuit,
this
shouldn’t
lead
us
to
say
that
the
 domain
of
economics
had
been
widened.2
But
this
“issue”
looks
entirely
semantic.

 A
different
way
in
which
some
theorists
have
aimed
to
carry
economics
“inside”
the
 individual
was
pioneered
by
Strotz
(1955‐56),
Schelling
(1978,
1980,
1984)
and
 Ainslie
(1992),
and
has
been
refined
in
contributions
by
Benabou
and
Tirole
(2004),
 Benhabib
and
Bisin
(2004)
and
Fudenberg
and
Levine
(2006).
All
of
this
work
 involves
explaining
behavioral
patterns
in
individual
people
as
equilibria
of
games
 amongst
sub‐personal
agents,
which
might
or
might
not
be
identified
with
 functional
modules
of
brains,
such
as
those
suggested
by
McClure
et
al
(2004).
Gul
 and
Pesendorfer
(2001)
develop
the
details
of
a
modeling
framework
that
is
 explicitly
intended
to
capture
the
main
phenomena
which
interest
the
“multiple
 self”
tradition
(procrastination,
addiction,
and
other
forms
of
intertemporal
 irresolution)
without
positing
any
sub‐personal
agents.
Thus,
whatever
view
one
 might
take
of
the
comparative
merits
of
these
approaches,
one
cannot
in
this
 instance
accuse
GP
of
resting
their
negative
case
on
arbitrary
restrictiveness
about
 what
economists
should
and
shouldn’t
get
up
to.
Note
that
Glimcher
(2009)
also
 explicitly
rejects
multiple‐selves
models,
at
least
insofar
as
the
sub‐personal
selves
 in
question
are
identified
with
neural
modules.
 We
refrain
from
declaring
for
one
of
the
dogs
in
this
fight.
Empirical
data
on
the
 dynamics
of
intertemporal
consumption
are
not
yet
sufficiently
rich
to
favor
any
one
 model
over
its
alternatives.
Instead,
we
note
an
irony
in
this
debate:
the
fact
that
GP
 make
a
methodological
fetish
out
of
choice
at
the
level
of
the
individual
personal
 agent
renders
them
bedfellows

of
precisely
the
BES
school
of
neureoconomics
they
 























































 2
This
is
the
attitude
of
Ross
(2005),
Ross
et
al
(2008),
and
Caplin
and
Dean
(2008).
 3
 

    
    attack.
BES
consists
in
repeating
protocols
that
putatively
demonstrate
human
 “irrationality”
under
neuroimaging,
and
trying
to
show
how
“anomalies”
in
rational
 choice
have
their
origins
and
explanations
in
framing
effects
that
result
from
the
 computational
processing
architecture
of
the
brain.
These
behavioral
economists
 strike
the
attitude
of
rebels
against
“mainstream”
or

“neoclassical”
economics
and
 revealed
preference
analysis,
and
promote
neuroeconomics
as
a
core
part
of
the
 alternative
program.
Camerer,
Loewenstein
and
Prelec
(2005)
provide
a
useful
 survey
of
work
in
this
area,
unfortunately
presented
as
if
it
constituted
the
whole
of
 neuroeconomics.

 We
share
the
skepticism
of
GP
about
BES.
But
we
believe
that
a
stubborn
refusal
to
 think
about
the
processes
that
support
relationships
among
incentives,
 opportunities
and
choices
is
a
self‐defeating
way
of
promoting
this
skepticism.
It
 concedes
to
the
advocate
of
BES
that
what
economics
is
fundamentally
about
are
 individual
people
arriving
at
and
applying
valuations
all
by
themselves.
This
ignores
 what
Smith
(2007)
refers
to
as
the
ecological
nature
of
economic
rationality.
 Ecological
rationality
emphasizes
the
extent
to
which
people
approximate
 consistent,
“old‐fashioned”
economic
rationality,
not
because
of
computational
 marvels
they
achieve
with
their
raw
brains,
but
by
means
of
what
the
philosopher
 Clark
(1997),
following
Hutchins
(1995),
calls
cognitive
“scaffolding.”
This
consists
 of
external
structures
in
the
environment
that
encode
culturally
accumulated
 information
and
constrain
and
channel
behavior.
Economists
are
familiar
with
social
 scaffolding
under
the
label
of
“institutions.”

 Consider,
for
example,
an
investor
deciding
how
often
she
should
churn
her
 portfolio.
One
might
imagine
her
modeling
the
financial
market,
including
a
 representative
agent
with
rational
expectations.
This
agent
gathers
time‐indexed
 data
on
prices
and
returns,
and
calculates
an
average
churn
rate,
variance
and
 signal‐response
rule
from
which
to
construct
a
policy
that
maximizes
her
expected
 lifetime
earnings.
Though
this
is
how
economists
might
aim
to
discover
the
 investor’s
policy,
they
know
this
does
not
describe
the
approach
of
the
actual,
 typical
person.
What
she
does
instead
is
exploit
the
fact
that
she
operates
within
a
 network
of
public,
normative
signals.
The
transaction
fees
charged
by
licensed
 brokers
vary
within
narrow
ranges.
These
ranges
systematically
co‐vary
with
 volatility
in
asset
prices,
dividend
rates,
and
opportunity
costs
of
investments
in
the
 assets
in
which
the
brokers
traffic.
Our
agent
might,
in
ignorance,
begin
by
suffering
 abuse
at
the
hands
of
her
broker
and
churning
too
much.
Fortunately,
her
 environment
is
likely
to
be
full
of
signals,
including
institutionally
encoded
second‐ order
signals
about
the
relative
reliability
of
various
first‐order
signals,
that
provide
 her
with
a
good
chance
of
discovering
this.
She
might
over‐react
and
for
a
time
 churn
too
little.
But
she
also
may
have
access
to
signals
about
the
earnings
of
other
 people
whose
situations
are
similar
to
hers.
If
they
consistently
do
better
than
she
 does,
it
may
occur
to
her
that
when
her
new
broker
tells
her
that
she
is
holding
 assets
for
too
long,
this
one
might
not
be
trying
to
exploit
her.
Ultimately,
our
agent’s
 behavioral
transaction
rate
might
range
close
to
what
the
economist
formally
 modeling
her
situation
would
recommend.
 4
 

    
    We
can
imagine
a
psychologist
studying
the
investor
in
detail
over
the
course
of
 years
and
predicting
her
specific
behavior
more
exactly
than
is
possible
using
the
 economist’s
“as
if”
approach.
If
all
investors
were
psychologically
identical
this
 might
be
considered
grounds
for
disinvesting
in
economists
and
using
the
savings
to
 train
more
psychologists.
But
there
are
persuasive
grounds
for
thinking
that
 psychological
variance
among
people
is
such
as
to
defeat
this
argument.
Each
 member
of
a
set
of
investors
might
use
an
idiosyncratic
learning
path
and
 idiosyncratic
representations
to
represent
the
same
thing
at
a
sufficiently
abstract
 scale
of
description
–
namely,
the
approximately
optimal
churn
rate
for
a
broad
 range
of
asset
classes
and
market
parameters.
In
that
case,
the
economists
will
be
in
 a
far
stronger
position
to
predict
out
of
sample,
and
to
offer
aggregate‐scale
 predictions,
than
any
of
the
psychologists
who
are
experts
on
particular
individuals.
 This
example
is
drastically
simplified
by
comparison
with
a
case
that
would
be
of
 genuine
scientific
interest.
GP
write
as
if
we
know,
in
advance
of
all
empirical
work,
 that
psychological
variation
among
individuals
learning
a
given
economic
 relationship
will
have
a
known
distribution,
for
which
we
can
exactly
control
in
a
 reduced‐form
model.
This
is
very
often
false,
in
which
event
work
of
psychologists
 or
sociologists
or
neuroscientists
might
provide
relevant
evidence
in
any
given
case.
 More
importantly,
our
access
to
evidence
might
be
limited
in
such
a
way
that
our
 most
practical
procedure
is
to
include
parameters
in
our
models
that
allow
the
data
 to
help
us
sort
a
population
into
sub‐populations,
each
of
whose
members
converge
 on
an
economically
distinct
solution.

 Consider,
as
a
real
instead
of
stylized
example,
the
very
phenomenon
that
has
 motivated
the
arguments
between
multiple‐self
and
single‐self
models
of
 intertemporal
consumption,
namely,
intertemporal
discounting.
Begin
by
imagining
 two
limiting
cases.
In
Case
1
each
agent
draws
a
discounting
function
from
a
random
 distribution
of
mathematically
possible
forms,
but
all
agents
are
financially
 punished
by
their
market
institutions
to
the
extent
that
they
depart
from
 exponential
discounting
at
a
common
rate
x.
Psychologically,
these
agents
are
 perfectly
idiosyncratic.
In
Case
2
all
agents
discount
identically,
but
in
a
way
that
 causes
them
all
to
manifest
systematic
intertemporal
preference
reversals.
GP
write
 as
if
Case
1
were
standard.
Here,
so
far
as
economic
prediction
is
concerned,
 psychological
processing
variables
are
pure
noise.
Advocates
of
BES
prefer
a
 methodology
that
assumes
Case
2
to
be
the
norm.
These
agents
evidently
face
some
 barrier
against
learning
to
create
institutional
scaffolding
that
could
help
them
to
 act
more
consistently.
Perhaps
it
is
their
recalcitrant
brains
that
get
in
their
way.
 Recent
empirical
work
has
shed
considerable
light
on
reality
with
respect
to
this
 example.
Hypotheses
resembling
both
limiting
case
have
been
cleanly
rejected.
 When
we
study
discounting
behavior,
or
any
choice
process,
we
have
to
worry
 about
whether
any
given
agent
uses
more
than
one
model
or
“data
generating
 process”
in
different
domains
(Harrison
and
Rutström
(2009);
Coller
et
al.
(2009)).
 When
we
model
any
behavioral
function
in
economics
we
have
to
account
 systematically
for
the
individual
heterogeneity
that
we
expect
a
priori
from
 5
 

    
    preferences,
but
which
demand
formal
econometric
methods
and
sample
sizes
 (Harrison
at
al.
(2002);
Harrison
and
Rutström
(2008)).
We
have
learned
that
 discounting
in
real
populations
is
better
described
by
structural
models
which
allow
 representation
of
heterogeneity
with
respect
to
curvature
of
utility
functions
and
 with
respect
to
discount
rates,
than
by
models
that
impose
a
common
reduced
 functional
form
on
everyone.
And
we
also
know
that
exponential
discounting
 usually
predominates
over
hyperbolic
discounting,
contrary
to
a
common
dogma
of
 behavioral
economics

(Andersen
et
al.
(2008)).
 Thus
the
most
empirically
robust
and
informative
models
of
discounting
and
related
 behaviors
include
some
processing
variables,
but
only
as
many
as
are
needed
to
 allow
us
to
estimate
a
limited
set
of
parameters
from
data.
The
process
of
discovery
 of
these
models
is
economics,
not
psychology,
because
it
does
not
discriminate
 between
different
hypotheses
that
would
be
of
major
concern
to
a
psychologist.
Do
 majorities
of
human
populations
converge
on
exponential
discounting
because
their
 brains
are
naturally
disposed
to
it,
or
because
the
social
ecologies
in
which
they
are
 immersed
train
them
to
do
so
and
induce
them
to
collectively
build
institutional
 scaffolding
that
helps
them
to
keep
on
track?

 Notice
that
if
one
were
doing
psychology,
and
were
investigating
the
question
just
 posed,
one
thing
one
might
want
to
do
is
wrench
subjects
out
of
their
scaffolded
 environments.
One
way
to
do
this
would
involve
putting
them
in
strange
 laboratories
performing
socially
novel
tasks;
or,
possibly,
to
disorient
them
still
 more,
by
making
them
lie
down
with
their
heads
in
magnets.
We
could
then
see
if
 they
discounted
in
ways
their
public
norms
would
deem
confused.
There
is
a
non‐ arbitrary
reason
to
say
one
is
doing
psychology
rather
than
economics
here:
in
the
 context
of
their
natural,
culturally‐evolved,
economies,
most
of
the
subjects
will
not
 behave
this
way.
This
points
to
a
respect
in
which
economics
is
not
merely
 methodologically
autonomous
with
respect
to
neuroscience:
economic
regularities,
 understood
as
ecological
properties
of
a
certain
kind,
can
causally
dominate
neural
 processing
properties
that
would
prevail
in
a
brain
forced
to
fend
for
itself.
 2.
The
Empirical
Methodology
of
Neuroeconomics
 One
of
the
basic
historical
divides
between
psychologists
and
economists
has
been
 the
reluctance
of
the
latter
to
embrace
latent
variables
in
their
modeling.
In
 discussions
by
both
psychologists
and
economists,

along
with
philosophers,
of
 methodologies
for
modeling
choices
of
individuals,
it
is
often
taken
for
granted
that
 the
only
grounds
for
abstemiousness
must
be
residual
behaviorism
on
the
part
of
 the
economists.
We
think
that
a
judicious
dash
of
behaviorism
is
well
advised
(Ross
 (2005)),
but
let
us
put
that
aside
here.
Even
economists
who
are
convinced
by
the
 fashionable
claim
that
behaviorism
is
a
wholly
pernicious
doctrine
should
recognize
 that
introducing
latent
variables
must
necessarily
involve
them
in
an
issue
around
 the
nature
of

“constructs”
which
their
usual
methods
allow
them
to
avoid.
 The
issue
in
question
is
the
distinction
between
“reflective”
and
“formative”
 constructs.
The
former
are
diagnosed
or
indicated
by
sets
of
observable
markers
 6
 

    
    which
are
each
supposed
to
perfectly
reflect
a
single
underlying
latent
variable.
Such
 constructs
are
based
on
factor
analysis,
used
to
discover
high‐loading
items
and
 reject
low‐loading
ones,
rather
than
structural
modeling.
Formative
constructs,
by
 contrast,
have
the
logical
character
of
dependent
variables
in
economic
models.
In
 the
context
of
research
on
pathological
gambling,
which
is
methodologically
 representative
of
much
psychological
work
on
clinical
conditions,
Schellinck
and
 Schrans
(2008)
complain
of
a
pervasive
confusion
between
reflective
and
formative
 constructs,
with
the
former
often
being
borrowed
from
diagnostic
practice
when
it
 is
the
latter
that
are
needed
for
scientific
discovery.
 This
issue
doesn’t
arise
for
economists,
for
two
reasons.
First,
their
unobservable
 explanatory
variables
are
often
axiomatized
by
reference
to
behavioral
conditions.
 Caplin
(2008)
has
drawn
attention
to
the
potential
value
of
this
for
neuroeconomics,
 and
Caplin
and
Dean
(2008)
have
begun
to
do
something
about
it.
More
prosaically,
 however,
the
use
of
standard
econometric
techniques
forces
economists
to
work
as
 if
all
of
their
lists
of
independent
variables
are
formative
constructs,
because
distinct
 variables
in
models
may
not
be
perfectly
correlated
with
one
another.
This
tends
to
 have
the
effect
of
making
it
hard
to
find
models
that
both
fit
wide
ranges
of
data
and
 are
econometrically
tractable.
Though
this
has
sometimes
been
a
subject
for
 complaint,
we
think
it
has
had
a
salutary
influence
on
economics.
For
one
thing,
it
 has
forced
a
steady
improvement
in
the
depth
and
sophistication
of
econometric
 techniques
and
theory.
This
is
a
good
reason
why
neuroeconomists
should
wish
to
 remain
economists.
 They
cannot
have
it
both
ways,
however.
If
maintenance
of
modeling
discipline
 recommends
the
economists’
club
when
key
target
variables
are
unobservable
as,
 we
will
argue,
they
are
in
neuroeconomics,
fMRI
nothwithstanding,
then
 econometric
restrictions
must
be
treated
seriously.
We
do
not
generally
find
this
in
 BES,
however,
and
we
have
an
hypothesis
as
to
why
not:
fMRI
data
are
frequently
 treated
as
if
they
were
first‐order
observations
rather
than
products
of
chains
of
 statistical
inference.
We
illustrate
this
general
point
with
three
types
of
 methodological
problems
that
have
arisen
in
recent
empirical
neuroscience
and
 neuroeconomics.
 
 2.1
Data
Versus
Estimates

    
    The
unit
of
neuroeconomic
analysis
is
a
spatial
location
in
the
brain
emitting
signals
 per
unit
of
time.
So
in
a
neuroeconomist’s
typical
fMRI
data
set
a
few
brains
 contribute
many
observations
at
each
point
in
time,
and
in
a
time‐series.

Therefore,
 statistical
issues
arise
both
for
inferences
about
single
brains,
and
for
inferences
 about
pooled
samples
of
brains.
To
understand
the
significance
of
the
former,
one
 need
only
review
the
typical
list
of
estimation
methods
employed;
for
example,
see
 Rabe‐Hesketh
et
al
(1997,
pp.
217‐226).
The
inferential
problem
here
is
simply
that
 point
estimates
from
one
stage
are
taken
as
data
in
the
next
stage,
and
then
there
is
 a
long
chain
of
such
inferences.
The
implication
is
that
the
standard
errors
of
 estimates
at
later
stages
tend
to
overstate
the
precision
of
estimations,
and
later
 7
 

    
    estimates
may
be
completely
inconsistent.3
This
is
compounded
by
the
inferential
 can
of
worms
involved
in
pooling
across
brains.
There
are
many
different
ways
to
 normalize
brains,
as
one
can
imagine
and
hope,
and
these
matter
for
inference.4

 The
end
result
is
that
the
statistical
modeling
of
neural
data
is
a
sequential
mixture
 of
limited‐information
likelihood
methods,
frequently
cobbled
together
by
ad
hoc

 methods:
“MacGyver
econometrics,”
to
borrow
the
label
of
Harrison
(2008a).
 Recognizing
this
fact
is
not
meant
as
a
way
of
invalidating
the
clinical
or
research
 goals
of
such
exercises,
but
as
a
reminder
of
the
extremely
limited
extent
to
which
 modeling
and
estimation
errors
are
likely
to
be
correctly
propagated
throughout
the
 chains
of
inferences
based
on
fMRI
data.
The
end
result
is
often
a
statistical
test
in
 which
left
hand
side
and
right
hand
side
variables
are
themselves
estimates,
often
 from
a
long
chain
of
estimates,
and
are
treated
as
if
they
are
first‐order
data.
This
 encourages
significant
understatement
of
standard
errors
on
estimate
of
effects,
 implying
significant
overstatement
of
statistically
significant
differential
activation.
 Consider
an
example
of
this
problem
from
Glimcher,
Kable
and
Louie
(2007).
In
one
 instance
a
correlation
is
calculated
that
refers
to
an
econometrically
estimated
 discounted
utility
function
from
standard
behavioral
data.
The
parameters
of
this
 function
have
standard
errors
when
estimated.
But
when
used
to
predict
activity
of
 the
brain,
when
used
as
a
right‐hand‐side
correlate,
the
standard
errors
disappear:
 These
individually
measured
indifference
curves
permitted
us,
for
 each
subject,
to
model
the
discounted
utility
of
each
delayed
option
 presented
to
our
subjects
in
the
brain
scanner.
With
this
behavioral
 measurement
in
hand,
we
could
then
ask
whether
any
activity
in
the
 brain
of
these
subjects
was
correlated
with
the
discounted
utility
of
an
 option
under
consideration.
We
found
that,
in
each
of
our
subjects,
the
 activity
of
the
brain
in
three
areas
typically
associated
with
option
 valuation
[...]
showed
a
clear
correlation
with
this
behaviorally
 derived
function.
Put
another
way,
brain
activity
measured
in
[three
 areas]
…
had
many
of
the
properties
of
that
subject’s
discounted
 utility
function
(p.
143).

 We
flag
this
problem
precisely
because
we
are
partial
to
the
underlying
algorithmic
 hypotheses
about
discounting
behavior.
It
is
most
important
to
be
critical
of
stories
 one
finds
most
plausible,
since
others
can
be
counted
on
for
the
rest.

 2.2
Reverse
inference
 The
other
general
problem
is
related,
and
is
known
in
the
literature
as
the
reverse
 inference
problem.
This
arises
when
activations
in
regions
of
the
brain
are
 























































 3
This
point
has
been
made
in
the
neuroscience
literature
by
Vul,
Harris,
 Winkielman
and
Pashler
(2009),
and
in
a
critique
of
the
neuroeconomics
literature
 by
Harrison
(2008a).
 4

Harrison
(2008a,
p.

312ff)
provides
references
to
the
literature.
 8
 

    
    presumed
to
identify
the
activation
of
a
(labelled)
cognitive
process.
Poldrack
 (2006)
explains
the
process
well,
and
provides
some
well‐cited
illustrations
of
the
 extent
of
the
problem.
 There
is
much
debate
in
neuroscience
about
the
selectivity
of
cognitive
processes.
 D’Esposito
et
al
(1998)
give
an
early
discussion
of
the
reverse
inference
problem
in
 this
context.
They
“conclude
that
human
lateral
prefontal
cortex
supports
processes
 in
addition
to
working
memory.
Thus,
reverse
inference
of
the
form
‘if
prefontal
 cortex
is
active,
working
memory
is
engaged’
is
not
supported”
(p.
274).
Claims
of
 this
sort
are
subject
to
debate
and,
of
course,
refinement
through
different
designs,
 instrumentation
and
methods
of
statistical
inference.
It
is
not
hard
to
find
such
 debates
in
neuroscience,
but
they
tend
to
be
glossed
in
the
very
neuroeconomics
 work
that
should
be
using
the
sophisticated
modelling
strategies
of
economics
to
 develop
tests
of
them.
 Kahneman
(2009,
pp.
523‐524)
provides
insight
into
the
tendency
in
the
 neuroeconomics
literature
to
avoid
discussing
awkward
statistical
issues,
such
as
 those
posed
by
the
reverse
inference
problem.
He
poses
the
problem
that
 researchers
face
as
follows,
and
in
exactly
the
way
that
is
relevant
for
the
 accumulation
of
knowledge
in
this
area,
when
one
must
evaluate
a
favored
story:
 High
correlations
between
well‐identified
psychological
and
neural
 measures
are
the
exception,
not
the
norm.
In
most
experiments
…
the
 correspondence
between
psychological
terms
and
neural
measures
is
 more
equivocal,
and
the
interpretation
of
imaging
results
is
tricky.
 Poldrack
[…]
has
drawn
attention
to
the
problem
of
"reverse
 inference,"
which
arises
when
people
infer
a
specific
a
psychological
 process
from
activity
in
a
particular
region
‐‐
for
example,
when
 activity
in
dorsal
striatum
is
interpreted
as
an
indication
that
people
 enjoy
punishing
strangers
who
have
behaved
unfairly
[…].
There
is
 indeed
a
problem,
because
activity
in
dorsal
striatum
is
not
perfectly
 correlated
with
enjoyment:
many
other
circumstances
produce
 activity
in
that
region,
and
there
is
no
assurance
that
it
will
be
active
 whenever
the
individual
experiences
pleasure.
In
spite
of
this
 difficulty,
the
result
and
its
proposed
interpretation
is
just
what
a
 general
psychologist
(not
a
neuroscience
specialist)
would
order.
It
is
 surprising
but
plausible,
and
it
drives
thinking
in
new
directions.
The
 more
difficult
test,
for
a
general
psychologist,
is
to
remember
that
the
 new
idea
is
still
a
hypothesis
which
has
passed
only
a
rather
low
 standard
of
proof.
I
know
the
test
is
difficult,
because
I
fail
it:
I
believe
 the
interpretation,
and
do
not
label
it
with
an
asterisk
when
I
think
 about
it.
And
I
will
be
sorry
if
it
is
disproved,
but
will
have
no
difficulty
 in
accepting
its
demise
‐‐
it
would
join
a
long
list
of
defunct
once‐ cherished
ideas.

 An
example
of
a
neuroeconomist
acknowledging
the
importance
of
the
problem
is
 Phelps
(2009,
pp.245‐247).
She
points
out
that
the
amygdala
and
the
insular
cortex,
 9
 

    
    two
regions
often
flagged
as
“proving”
that
emotions
play
a
role
in
economic
 decision‐making
have
been
identified
as
playing
other,
arguably
non‐emotional
 roles.
She
correctly
concludes
(p.
247)
that
“[a]lthough
reverse
inference
is
a
 powerful
technique
for
generating
hypotheses
and
ideas
that
inspire
additional
 studies
or
measurements,
its
use
as
a
primary
technique
for
determining
a
role
for
 the
emotions
is
questionable.”
In
light
of
this,
we
don’t
quite
see
what
licenses
her
 general
conclusion
about
the
role
of
neuroscience
in
demonstrating
an
empirical
 role
for
emotions
in
economic
decision‐making:
“Although
emotion
was
considered
 an
important
variable
in
economic
decision
making
prior
to
neuroeconomics

[…],
 the
recent
growth
in
this
field
has
highlighted
a
role
for
emotion
in
economic
 choice.”
(p.234).
We
do
not
quarrel
with
the
highly
plausible
conclusion.
We
simply
 can’t
find
new
evidence
for
it
coming
from
neuroeconomics,
even
in
a
careful
survey
 precisely
intended
to
summarize
that
contribution.
 The
problem
afflicts
BES
to
a
greater
extent
than
it
does
NE,
partly
because
the
 former
is
more
reliant
on
fMRI
work
with
humans
whereas
the
empirical
 foundations
of
the
latter
lie
in
single‐cell
recordings
in
rats.
Indeed,
we
find
a
nice
 example
of
leading
NE
researchers
invoking
the
problem
against
a
frequently
cited
 flagship
BES
result.
Glimcher
(2009,
pp.
518‐519)
criticizes
the
claim
of
McClure
et
 al
(2004)
that
an
identifiable
part
of
the
brain
implements
the
β
part
of
the
quasi‐ hyperbolic
discounting
model
and
another
part
of
the
brain
implements
the
δ
part
 of
that
model.
He
initially
questions
the
anatomical
viability
of
the
proposed
 localization
of
brain
activity,
and
the
lack
of
correspondence
to
well‐studied
animal
 data.
But
his
most
telling
criticism
is
that
two
of
the
regions
that
have
been
 identified
as
being
associated
with
"emotional"
decision‐making
in
these
particular
 tests,
the
basal
ganglia
and
the
medial
prefrontal
cortex,
have
also
been
shown
by
 others
to
be
associated
with
traditionally
“rational”
functions
such
as
the
encoding
 of
monetary
and
primary
rewards,
and
the
expression
of
ordinal
preference.

 2.3
Sorting
Out
Shadows
 Another
issue
of
significance
in
recent
debate
over
the
use
of
fMRI
imaging
is
the
 problem
of
“pre‐emptive
blood
flow”
identified
by
Sirotin
and
Das
(2009)
and
 summarized
by
Leopold
(2009).
The
phenomenon
was
detected
by
independently
 and
directly
measuring
blood
flow
and
neural
activity
in
monkeys.
The
inferential
 problem
arises
if
there
is
blood
flow
activity
in
anticipation
of
some
event
even
if
the
 event
does
not
occur.
Some
mismatch
between
blood
flow
and
neural
activity
is
 expected,
but
the
difference
here
is
several
orders
of
magnitude
beyond
the
 customary.
One
conjecture
is
that
this
is
a
type
of
“priming”
activity
in
the
brain,
so
 that
blood
is
ready
and
available
in
the
expectation
that
it
will
be
used.
But
since
the
 temporal
connection
between
stimulus
and
response
lies
at
the
heart
of
almost
all
 neuroeconomic
methods,
this
finding
is
extremely
troublesome.
It
is
particularly
 likely
to
be
a
problem
for
neuroscientific
research
involving
humans,
who
can
learn
 vast
networks
of
anticipation
from
their
cultures.
 
 10
 

    
    3.
Conclusion
 BES‐style
neuroeconomics
is
plagued
by
two
problems.
First,
it
inherits
from
 behavioural
economics
an
over‐emphasis
on
individual
“inboard”
choice,
divorced
 from
socially
scaffolded
context,
as
the
basic
subject
matter
of
economics.

Second,
 statistical
problems
that
are
debated
in
the
neuroscience
literature,
and
which
are
 familiar
to
econometricians,
simply
have
to
be
resolved
before
claims
of
typical
BES
 strength
can
be
regarded
as
justified.

Advocates
of
BES
cannot
simply
put
them
to
 one
side
as
“footnote
qualifications”
while
they
get
on
with
telling
interesting
 stories,
a
tendency
we
find
in
their
literature.5
 We
reject
the
view
that
neural
data
are
irrelevant
to
economics
as
needlessly
 isolationist.
But
we
also
reject
the
free‐disposability
view
that
any
data
is
useful
data
 until
proven
otherwise,
implying
that
we
should
just
collect
it
anyway
and
decide
 later
if
it
was
useful.
That
is
a
poor
model
for
advancement
of
study
in
any
field.
We
 welcome
NE
as
a
potential
contributor
to
formal
modelling
of
the
processes
by
 which
agents
make
economic
decisions,
though
we
emphasize
that
this
project
is
in
 its
infancy
and
depends
on
an
empirical
hypothesis
that
might
turn
out
to
be
wrong
 –
that
there
is
a
common
currency
of
valuation
in
the
brain.
We
advise
 neuroeconomists
to
remember
that
and
why
there
is
a
border
between
economics
 and
psychology,
however
often
they
commute
across
it.


    
    























































 5
For
example,
see
Fox
and
Poldrack
(2009,
p.166),
who
mention
the
reverse
 inference
problem
developed
so
well
by
Poldrack
(2006),
but
simply
move
on
 without
a
blink.
 11
 

    
    References
 Ainslie,
G.
(1992).
Picoeconomics.
Cambridge:
Cambridge
University
Press.
 Andersen,
S.;
Harrison,
G.W.;
Lau,
M.I.,
and
Rutström,
E.E.
(2008).
Eliciting
Risk
and
 Time
Preferences.
Econometrica
76(3):
583‐618.
 Bénabou,
R.,

and
Tirole,
J.
(2004).
Willpower
and
personal
rules.
Journal
of
Political
 Economy
112:
848‐886.
 Benhabib,
J.,
and
Bisin,
A.
(2004).
Modeling
internal
commitment
mechanisms
and
 self‐control:
A
neuroeconomics
approach
to
consumption‐saving
decisions.
Games
 and
Economic
Behavior
52:
460‐492.
 Bernheim,
B.D.
(2009).
On
the
potential
of
neuroeconomics:
a
critical
(but
hopeful)
 appraisal.
American
Economic
Journal:
Microeconomics.
Forthcoming
 Camerer,
C.,
Loewenstein,
G.,
and
Prelec,
D.
(2005).
Neuroeconomics:
how
 neuroscience
can
inform
economics.
Journal
of
Economic
Literature
43:
9‐64.
 Caplin,
A.
(2008).
Economic
theory
and
psychological
data:
Bridging
the
divide.
In
A.
 Caplin
and
A.
Schotter,
eds.,
The
Foundations
of
Positive
and
Normative
Economics:
A
 Handbook,
pp.
336‐371.
Oxford:
Oxford
University
Press.
 Caplin,
A.,
and
Dean,
M.
(2008).
Dopamine,
Reward
Prediction
Error,
and
Economics.
 Quarterly
Journal
of
Economics,
forthcoming.
 Caplin,
A.,
and
Schotter,
A.
(2008)
Eds.,
The
Foundations
of
Positive
and
Normative
 Economics:
A
Handbook,
Oxford:
Oxford
University
Press.
 Clark,
A.
(1997).
Being
There.
Cambridge,
MA:
MIT
Press.
 Coller,
M.;
Harrison,
G.W.,
and
Rutström,
E.E.
(2009).
Latent
process
heterogeneity
 in
discounting
behavior.
Working
Paper
09­04,
Department
of
Economics,
College
of
 Business
Administration,
University
of
Central
Florida,
2009.

 D’Esposito,
M.;
Ballard,
D.;
Aguirre,
G.
K.,
and
Zarahn,
E.
(1998).
Human
Prefrontal
 Cortex
Is
Not
Specific
for
Working
Memory:
A
Functional
MRI
Study.
NeuroImage
8:
 274‐282.
 Fox,
C.R.,
and
Poldrack,
R.A.
(2009).
Prospect
Theory
and
the
Brain.
In
P.
Glimcher,
C.
 Camerer,
E.
Fehr
and
R.
Poldrack,
eds.,
Neuroeconomics:
Decision
Making
and
the
 Brain,
pp.
145‐173.
London:
Elsevier.
 Fudenberg,
D.,
and
Levine,
D.K.
(2006).
A
dual
self
model
of
impulse
control.

 American
Economic
Review
96:
1449‐1476.
 Glimcher,
P.
(2003).
Decisions,
Uncertainty
and
the
Brain.
Cambridge,
MA:
MIT
Press.

    
    12
 

    
    Glimcher,
P.
(2009).
Choice:
towards
a
standard
back‐pocket
model.
In
P.
Glimcher,
 C.
Camerer,
E.
Fehr
and
R.
Poldrack,
eds.,
Neuroeconomics:
Decision
Making
and
the
 Brain,
pp.
503‐521.
London:
Elsevier.
 
 Glimcher,
P.W.;
Kable,
J.,
and
Louie,
K.
(2007)
Neuroeconomic
Studies
of
Impulsivity:
 Now
or
Just
as
Soon
as
Possible?
American
Economic
Review
(Papers
&
Proceedings)
 97(2):
142‐147
 Gul,
F.,
and
Pesendorfer,
W.
(2001).
Temptation
and
self‐control.
Econometrica
69:
 1403‐1436.
 
 Gul,
F.,
and
Pesendorfer,
W.
(2008).
The
case
for
mindless
economics.
In
A.
Caplin
 and
A.
Schotter,
eds.,
The
Foundations
of
Positive
and
Normative
Economics:
A
 Handbook,
pp.
3‐39.
Oxford:
Oxford
University
Press.
 Harrison,
G.W.
(2008a).
Neuroeconomics:
A
critical
reconsideration.
Economics
and
 Philosophy
24:
303‐344.
 
 Harrison,
G.W.
(2008b).
Neuroeconomics:
A
rejoinder.
Economics
and
Philosophy
24:
 533‐544.
 
 Harrison,
G.W.
(2009).
The
behavioral
counter‐revolution.
Journal
of
Economic
 Behavior
and
Organization
71:
Forthcoming.
 
 Harrison,
G.W.,
Harstad,
R.M.,
and
Rutström,
E.E.
(2004).
Experimental
Methods
and
 Elicitation
of
Values.
Experimental
Economics
7(2):
123‐140.
 Harrison,
G.W.;
Lau,
M.I.,
and
Williams,
M.B.
(2002).
Estimating
Individual
Discount
 Rates
for
Denmark:
A
Field
Experiment.

American
Economic
Review
92(5):
1606‐ 1617.
 Harrison,
G.W.,
and
Rutström,
E.E.
(2008).
Risk
Aversion
in
the
Laboratory.
In
J.C.
 Cox
and
G.W.
Harrison,
eds.,
Risk
Aversion
in
Experiments
Bingley,
UK:
Emerald.
 Harrison,
G.W.,
and
Rutström,
E.E.
(2009).

Representative
Agents
in
Lottery
Choice
 Experiments:
One
Wedding
and
A
Decent
Funeral.
Experimental
Economics,
12(2):
 133‐158.
 
 Hutchins,
E.
(1995)
Cognition
in
the
wild.
Cambridge,
MA:
MIT
Press.
 Kahneman,
D.
(2008).
Remarks
on
neuroeconomics.
In
P.
Glimcher,
C.
Camerer,
E.
 Fehr
and
R.
Poldrack,
eds.,
Neuroeconomics:
Decision
Making
and
the
Brain,
pp.
523‐ 526.
London:
Elsevier.
 
 Knutson,
B.,
Rick,
S.,
Wimmer,
G.E.,
Prelec,
D.,
&
Loewenstein,
G.
(2007).
Neural
 predictors
of
purchases.
Neuron,
53:
147‐156.
 Leopold,
D.A.
(2009).
Pre‐emptive
blood
flow.
Nature
457:
387‐388.
 13
 

    
    McClure,
S.,
Laibson,
D.,
Loewenstein,
G.,
and
Cohen,
J.
(2004).
Separate
neural
 systems
value
immediate
and
delayed
monetary
rewards.
Science
306:
503‐507.
 Phelps,
E.A.
(2008).
The
study
of
emotions
in
neuroeconomics.
In
P.
Glimcher,
C.
 Camerer,
E.
Fehr
and
R.
Poldrack,
eds.,
Neuroeconomics:
Decision
Making
and
the
 Brain,
pp.
523‐526.
London:
Elsevier.
 
 Poldrack,
R.A.
(2006).
Can
Cognitive
Processes
by
Inferred
from
Neuroimaging
 Data?
Trends
in
Cognitive
Science
10:
59‐63.
 Poldrack,
R.A.,
and
Wagner,
A.D.
(2004).
What
Can
Neuroimaging
Tell
Us
About
the
 Mind?
Insights
from
Prefontal
Cortex.
Current
Directions
in
Psychological
Science
 13(5):
177‐181.
 Rabe‐Hesketh,
S.;
Bullmore,
E.T.,
and
Brammer,
M.J.,
“The
Analysis
of
Functional
 Magnetic
Resonance
Images,”
Statistical
Methods
in
Medical
Research,
1997,
6,
215‐ 237.
 Ross,
D.
(2005).
Economic
Theory
and
Cognitive
Science:
Microexplanation.
 Cambridge,
MA:
MIT
Press.
 Ross,
D.
(2008).
Two
Styles
of
Neuroeconomics.
Economics
and
Philosophy
24:
473‐ 483.
 Ross,
D.,
Sharp,
C.,
Vuchinich,
R.,
and
Spurrett,
D.
(2008).
Midbrain
Mutiny.
 Cambridge,
MA:
MIT
Press.
 Schellinck,
T.,
and
Schrans,
T
(2008).
Screening
With
Meaning:
New
Directions
in
the
 Development
of
Gambling
Risk
Measures.
 http://www.assissa.eu/easg/thursday/1400‐ses4/schellinck_tony.pdf
.
Accessed
1
 June
2009.
 Schelling,
T.
(1978).
Economics,
or
the
art
of
self‐management.
American
Economic
 Review.
68:
290‐294.
 Schelling,
T.
(1980).
The
intimate
contest
for
self‐command.
Public
Interest.
60:
94‐ 118.
 Schelling,
T.
(1984).
Self‐command
in
practice,
in
policy,
and
in
a
theory
of
rational
 choice.
American
Economic
Review.
74:
1‐11.
 Sirotin,
Y.G.,
and
Das,
A.
(2009).
Anticipatory
haemodynamic
signals
in
sensory
 cortex
not
predicted
by
local
neuronal
activity.
Nature
457:
475‐479.
 Smith,
V.
(2007).
Rationality
in
Economics.
Cambridge:
Cambridge
University
Press.
 Strotz,
R.H.
(1955‐56).
Myopia
and
Inconsistency
in
Dynamic
Utility
Maximization.
 The
Review
of
Economic
Studies
23(3):
165‐180.

    
    14
 

    
    Vul,
E.;
Harris,
C.;
Winkielman,
P.,
and
Pashler,
H.
(2009).
Puzzlingly
High
 Correlations
in
fMRI
Studies
of
Emotion,
Personality,
and
Social
Cognition.
 Perspectives
on
Psychological
Science.
4(3):
274‐290.

    
    15
 


Add Comment

To add a comment, you need to login or signup (you can signup without leaving this page)