1

~11,000
words



Re‐Thinking
the
Criminal
Standard
of
Proof:
Seeking
Consensus
about
the
Utilities
of

Trial
Outcomes



Larry
Laudan1

Harry
Saunders2





For
many
people,
utility
assessments
for
trial
outcomes
and

threshold
probabilities
are
not
well
developed
and
closely

articulated.—Terry
Connolly3





Summary/Abstract:
 For
 more
 than
 a
 half‐century,
 evidence
 scholars
 have
 been

exploring
whether
the
criminal
standard
of
proof
can
be
grounded
in
decision
theory.


Such
grounding
would
require
the
emergence
of
a
social
consensus
about
the
utilities

to
 be
 assigned
 to
 the
 four
 outcomes
 at
 trial.
 Significant
 disagreement
 remains,
 even

among
legal
scholars,
about
the
relative
desirability
of
those
outcomes
and
even
about

the
formalisms
for
manipulating
their
respective
utilities.

We
attempt
to
diagnose
the

principal
 reasons
 for
 this
 dissensus
 and
 to
 suggest
 ways
 in
 which
 a
 broadly
 shared

evaluation
 might
 be
 forged,
 both
 with
 respect
 to
 the
 appropriate
 equations
 for

defining
 the
 standard
 of
 proof
 and
 with
 respect
 to
 the
 appropriate
 utilities
 to

associate
with
the
various
trial
outcomes.
Where
consensus
cannot
be
forged,
we
hold

that
remaining
differences
can
probably
be
finessed.
We
also
suggest
ways
to
elicit
the

utilities
of
individuals
on
these
matters
so
as
to
avoid
the
usual
flaws
of
such
surveys.


Along
 the
 way,
 we
 note
 a).
 the
 disproportionate
 role
 that
 the
 Blackstone
 ratio
 of

errors
 continues
 to
 play
 in
 appraisals
 of
 the
 utilities
 of
 trial
 outcomes
 (despite
 its

unintelligibility
in
the
context
of
utilities)
and
b).
the
persisting
belief
–for
which
there

is
 no
 theoretical
 basis—that
 every
 plausible
 assignment
 of
 utilities
 will
 inevitably

result
 in
 a
 very
 high
 standard
 of
 proof.
 
 Finally,
 we
 examine
 some
 of
 the
 technical

features
associated
with
a
proposed
rank
ordering
of
the
utilities
of
trial
outcomes.






Introduction



Ever
 since
 the
 abandonment
 of
 trials
 by
 ordeal
 in
 the
 thirteenth
 century,
 Western

legal
 systems
 –whether
 Roman
 or
 common
 law—have
 shared
 the
 view
 that
 a
 just

finding
of
guilty
requires
a
very
high
degree
of
confidence
in
guilt
on
the
part
of
the

trier
 of
 fact.
 
 Accordingly,
 through
 the
 late
 Middle
 Ages
 and
 early
 modern
 era,

certainty
 or
 full
 proof
 was
 required
 to
 convict
 the
 accused.
 
 By
 the
 late
 seventeenth

century,
 this
 overly
 exacting
 demand
 had
 given
 way
 in
 the
 English‐speaking
 world,



























































1
Instituto de Investigaciones Filosóficas, Universidad Autónoma de México.
2
Managing Director, Decision Processes Incorporated.
3
Terry Connally et al., JUDGMENT AND DECISION MAKING (2nd ed.), Cambridge University Press (2000), p.
239.
 2

largely
 under
 the
 influence
 of
 Locke
 and
 other
 empiricist
 theorists
 of
 knowledge,
 to

the
 idea
 that
 high
 probability
 or
 moral
 certainty
 was
 the
 right
 threshold
 for
 a

conviction.
 By
 the
 early
 nineteenth
 century,
 Roman
 law
 had
 signed
 on
 to
 this

weakening
 of
 the
 traditional
 standard
 as
 well,
 with
 an
 “intimate
 conviction
 of
 guilt”

replacing
certainty
or
“full
proof”
as
the
standard
in
criminal
trials.





One
of
the
key
ideas
driving
the
belief
that
the
standard
must
be
very
demanding
was

the
 thought
 that
 falsely
 convicting
 someone
 of
 a
 felony
 (and
 thus,
 in
 those
 days,

condemning
 them
 to
 death)
 was
 an
 unacceptably
 egregious
 harm
 and
 that
 the
 best

way
to
prevent
it
was
to
set
the
standard
of
proof
at
the
limits
of
what
mere
mortals

could
achieve
by
way
of
proof
about
matters
of
fact.




From
early
on,
scholars
understood
that
an
exacting
standard
of
proof,
while
holding

false
convictions
to
a
minimum,
would
exact
an
abundant
harvest
of
false
acquittals.

This
is
a
price
that
they
thought
society
should
be
willing
to
pay.
William
Blackstone

famously
 encapsulated
 this
 idea
 in
 the
 aphorism
 that
 “it
 is
 better
 that
 ten
 guilty

persons
escape,
than
that
one
innocent
suffer.”4

Demanding
moral
certainty
(or,
later:

proof
 beyond
 a
 reasonable
 doubt),
 Blackstone
 and
 many
 of
 his
 successors
 evidently

thought,
would
respect
and
reproduce
that
result:
to
wit,
at
least
ten
false
acquittals

for
every
false
conviction.5
Prominent
French
theorists
took
the
idea
one
step
further,

calculating
that
it
would
be
unjust
to
have
a
trial
that
exposed
an
innocent
person
to
a

greater
 risk
 of
 false
 conviction
 than
 he
 ran,
 in
 his
 daily
 life,
 of
 dying
 prematurely

within
 the
 next
 twenty‐four
 hours.
 
 Using
 this
 touchstone,
 Condorcet
 surmised
 that

the
 standard
 of
 proof
 in
 a
 criminal
 trial
 should
 be
 a
 confidence
 of
 guilt
 exceeding

144,767/144,768
or
99.9993%;
otherwise,
the
legal
system
would
be
unjust.6
He
was

evidently
 not
 overly
 concerned
 about
 how
 one
 might
 assemble
 a
 proof
 of
 guilt
 that

was
this
powerful.



We
 have
 to
 fast
 forward
 to
 the
 middle
 of
 the
 20th
 century
 before
 this
 idea
 of

quantifying
 the
 standard
 of
 proof
 takes
 on
 a
 more
 realistic
 form.
 
 Modern
 evidence

scholars
 of
 a
 quantitative
 bent
 have
 pursued
 the
 idea
 that
 one
 could
 define
 the

criminal
 standard
 of
 proof
 much
 more
 precisely
 and
 less
 ambiguously
 than
 “moral

certainty”
or
“proof
beyond
a
reasonable
doubt”
does
by
representing
it
as
a
specific

probability
 or
 clearly‐delimited
 range
 of
 probabilities.
 
 More
 than
 that,
 they
 have


























































4
4 William Blackstone, COMMENTARIES 352 (1769).
5
Recent writers on this subject would be quick to point out that this move from a demanding standard of
proof to any claim whatever about the actual frequency of ensuing errors is a very perilous hypothesis. (See
below as well as Michael DeKay, “The Difference between Blackstone-like Error Ratios and Probabilistic
Standards of Proof,” 21 LAW AND SOC. INQUIRY, 95 (1996) and Ronald Allen, “Clarifying the Burden of
Persuasion and Bayesian Decision Rules,” 4 INTL. J. EVID. & PROOF, 246 (2000)) Among other things, the
frequency of error will depend on the (unknowable) distribution of guilt and innocence among those who are
brought to trial.
6
For details on the rivalry among French intellectuals in the eighteenth century to see who could offer the
most preposterous definition of the standard of proof, see Lorraine Daston, Classical Probability in the
Enlightenment (Princeton: Princeton University Press, 1988), chapter six. For Condorcet’s proposal in
particular, see his Essai sur l’application de l’analyse à la probabilité des decisiones rendues à la pluralité
des voix (Paris, 1785), p. cix.
 3

maintained
 that
 we
 can
 derive
 the
 appropriate
 probabilistic
 standard
 from
 a

consideration
of
the
costs
or
utilities
of
the
pertinent
outcomes
at
trial.





One
 of
 the
 first
 attempts
 to
 derive
 the
 standard
 of
 proof
 from
 utilities
 is
 associated

with
 John
 Kaplan,
 whose
 widely
 cited
 paper,
 “Decision
 Theory
 and
 the
 Factfinding

Process”
 (1968),
 offered
 “a
 formula
 for
 determining
 the
 probability
 necessary
 to

return
 a
 verdict.”7
 
 He
 arrived
 at
 the
 conclusion
 that
 the
 criminal
 standard
 of
 proof

would
be
given
by

, (1)
where
 
is
the
threshold
or
standard
of
proof,
 
is
the
disutility
of
a
false
acquittal

and
 is
the
disutility
of
a
false
conviction.
In
effect,
Kaplan’s
proposal
was
that
the

standard
of
proof
was
a
function
of
the
ratio
of
the
(dis)utilities
of
the
two
erroneous

outcomes
of
a
trial.

If
we
could
fix
the
value
of
that
ratio,
he
thought,
we
would
have
a

unique
and
precise
value
for
the
point
at
which
a
criminal
conviction
was
rational.



One
 trouble
 with
 Kaplan’s
 proposal
 
 ‐‐as
 subsequently
 pointed
 out
 by
 Tribe8,

Lillquist9,
and
several
others10‐‐
is
that
it
is
an
incomplete
application
of
the
theory
of

rational
decisions
to
the
problem
at
hand.

There
are
not
only
two
outcomes
at
trial

but
 four.
 Kaplan’s
 analysis
 ignores
 the
 utilities
 of
 correct
 verdicts.
 
 Unless
 those

positive
 outcomes
 have
 negligible
 utility
 
 ‐‐which,
 as
 we
 shall
 see
 below
 is
 wildly

improbable—their
exclusion
from
the
calculation
of
the
standard
of
proof
vitiates
this

way
of
conceiving
the
problem.



Not
 long
 after
 the
 Kaplan
 proposal,
 Laurence
 Tribe
 produced
 (in
 1971)
 a
 more

plausible
way
of
relating
trial
outcomes
to
the
theory
of
utilities.11

Specifically,
Tribe

argued
 that
 if
 we
 take
 all
 four
 utilities
 seriously
 into
 account,
 we
 could
 locate
 the

threshold
of
proof
at
that
point
of
indifference,
where
the
product
of
the
utilities
of
a

conviction
 times
 its
 probability
 was
 the
 same
 as
 the
 utility
 of
 an
 acquittal
 times
 its

probability.
Tribe’s
formula
for
the
threshold
of
proof
thus
becomes:

, (2)
where
 is
the
utility
of
a
true
acquittal,
 
is
the
utility
of
a
false
acquittal,
 
is
the

utility
 of
 a
 true
 acquittal
 and
 
 is
 the
 utility
 of
 a
 false
 conviction.
 Clearly,
 Tribe’s



























































7
John Kaplan, “Decision Theory and the Factfinding Process,” 20 STAN L REV 1065 at 1071ff. (1967-68).
8
Laurence
Tribe,
“Trial
by
Mathematics:

Precision
&
Ritual
in
the
Legal
Process,
84
HARV.
L.
REV.
1329,


1375
(1971).

9
Erik Lillquist, “Recasting Reasonable Doubt: Decision Theory and the Virtues of Variability,” 36 U.C.
Davis L. Rev., 85 (2002).
10
See especially, Richard Friedman, “Standards of Persuasion and the Distinction between Fact and Law,” 86
NW. U. L. REV. 916 (1992).
11
Laurence Tribe, op. cit. Interestingly, Tribe believes that Kaplan already incorporated all four utilities. He
also correctly credits Alan Cullison, along with Kaplan, with having formulated the four-utility model. See
Alan Cullison, “Probability Analysis of Judicial Fact-Finding: A Preliminary Outline of The Subjective
Approach,” U. TOL. L. REV. 538 (1969).
 4

version
explicitly
incorporates
the
utilities
that
Kaplan
had
ignored.
 12
Equally
clearly,

Tribe’s
version
reduces
to
Kaplan’s
only
if
the
utilities
of
a
true
conviction
and
a
true

acquittal
 are
 zero.
 
 What
 (2)
 suggests
 is
 that
 if
 society,
 presumably
 through
 its

legislators,
 can
 reach
 agreement
 about
 assigning
 plausible
 values
 to
 the
 costs
 and

benefits
 of
 the
 four
 possible
 outcomes
 at
 trial,
 then
 we
 will
 have
 a
 nonarbitrary

mechanism
 for
 setting
 a
 standard
 of
 proof
 that
 makes
 us
 indifferent
 as
 between

acquitting
or
convicting
someone
accused
of
a
crime.




Many
 evidence
 scholars
 continue
 to
 propose
 variants
 of
 the
 Kaplan
 approach
 of

deriving
the
standard
of
proof
from
only
two
utilities
rather
than
Tribe’s
four,
despite

powerful
 arguments
 from
 DeKay
 and
 Lillquist
 about
 the
 dangers
 of
 ignoring
 the

utilities
of
positive
outcomes.





We
believe
that
the
case
for
grounding
the
standard
of
proof
on
(2)
rather
than
on
(1)

is
overwhelming.

We
also
believe,
however,
that
far
too
little
care
has
been
paid
by

scholars
 to
 the
 question
 of
 how
 we
 might
 figure
 out
 what
 utilities
 ordinary
 citizens

would
and
should
assign
to
those
outcomes.

In
this
paper,
we
have
four
aims:

a).
to

review
succinctly
the
arguments
already
in
the
literature
about
the
relative
merits
of

(1)
 and
 (2);
 b).
 
 to
 strengthen
 the
 case
 for
 (2)
 by
 elaborating
 additional
 reasons
 for

regarding
(1)
as
founded
on
fundamental
confusions
about
the
nature
of
utilities
and

the
 theory
 of
 decisions;
 c).
 to
 explain
 why
 (1),
 despite
 its
 evident
 lack
 of
 merit,

continues
 to
 enjoy
 a
 large
 following
 in
 the
 field;
 and
 d).
 to
 attempt
 to
 remedy
 the

inattention
 to
 methods
 of
 empirically‐ascertaining
 the
 utilities
 involved
 in
 (2)
 by

proposing
a
wholly
new
way
of
eliciting
those
utilities.



Calculating
the
Standard
of
Proof
from
Societal
Utilities




We
begin
by
reprising,
in
somewhat
more
detail
than
Tribe
gave,
how
the
standard
of

proof
can
be
derived.
The
following
diagram
illustrates
the
decision
faced
by
a
juror:



























































12
It is crucial to point out that Tribe, having cleaned up the utilities-based formula for the standard of proof,
proceeded to reject the enterprise categorically because of its putative conflict with basic value commitments
of the administration of criminal justice. For instance, it explicitly admits that we are prepared to convict
persons even when there is some non-zero probability that they are innocent, an admission that Tribe believes
would undermine the acceptance of the results of trial by the general population. It strikes us as curious that
Tribe thinks that the unwashed multitudes cannot comprehend that mistakes at trial are inevitable and that,
accordingly, we must take intellectually dishonest steps to conceal this information from them.
 5





Figure
1.
Decision
Faced
by
a
Trial
Juror



In
this
diagram
 is
the
probability
of
guilt
of
the
defendant
as
perceived
by
the
juror.


 Given
the
following
four
societal
utilities:


The
expected
societal
utility
of
conviction
as
a
function
a
juror’s
perceived
probability

of
guilt
 is:

(3)
Similarly,
 the
 expected
 societal
 utility
 of
 acquittal
 as
 a
 function
 of
 this
 perceived

probability
is:

(4)
Given
 the
 four
 societal
 utilities,
 the
 quantified
 standard
 of
 proof
 is
 calculated
 as
 the

perceived
 probability
 of
 guilt
 that
 makes
 society
 indifferent
 between
 conviction
 and

acquittal,
that
is,
the
probability
that
makes
these
two
expected
utilities and 

equal:

(5)

where .
 This
 is
 formally
 equivalent
 to
 the
 Tribe


proposal
 (2),
 as
 can
 be
 seen
 by
 dividing
 the
 numerator
 and
 denominator
 by

.



 6

This
result
says
that
society
would
like
individual
jurors
to
vote
to
convict
if
and
only

if
 their
 perceived
 probability
 of
 the
 defendant’s
 guilt
 is
 greater
 than
 or
 equal
 to ,

and
would
like
them
to
vote
acquittal
if
and
only
if
their
perceived
probability
of
guilt

is
less
than
this
threshold.



Relationship
between
the
Standard
of
Proof
and
Societal
Utilities




By
plotting
the
expected
utilities
 and
 
against
perceived
probability
of
guilt,


we
 can
 illustrate
 this
 calculation
 and
 show
 how
 the
 four
 utilities
 and

determine
the
desired
societal
threshold
of
proof:






Figure
2.
Determining
the
Desired
Societal
Standard
of
Proof



The
green
line
is
the
expected
societal
utility
 
of
an
acquittal
(Not
Guilty
verdict)

as
a
function
of
the
perceived
probability
of
guilt
 
and
the
red
line
is
the
expected

societal
 utility
 of
 a
 conviction
 (Guilty
 verdict)
 as
 a
 function
 of
 the
 perceived

probability
 of
 guilt.
 
 The
 intersection
 point
 of
 these
 two
 lines
 shows
 the
 desired

standard
of
proof .



Note
 that
 if
 the
 juror
 perceives
 the
 probability
 of
 guilt
 of
 the
 defendant
 to
 be ,

equation
 (3)
 shows
 the
 expected
 societal
 utility
 of
 a
 conviction
 to
 be ;
 that
 is,

society
will
deem
a
conviction
to
be
entirely
inappropriate
if
the
juror
perceives
it
to

be
certain
that
the
defendant
is
innocent.
This
is
shown
in
the
lower
left
of
Figure
2
as

the
point ,
the
intersection
of
the
Guilty
line
with
the
axis .

Similarly,
if

the
 juror
 perceives
 the
 probability
 of
 guilt
 of
 the
 defendant
 to
 be ,
 equation
 (4)

shows
the
expected
societal
utility
of
an
acquittal
to
be ;
that
is,
society
will
deem

an
 acquittal
 to
 be
 entirely
 unjustified
 if
 the
 juror
 perceives
 it
 to
 be
 certain
 that
 the

defendant
is
guilty.

This
is
shown
in
the
upper
right
of
Figure
2
as
the
point .


 7

The
 other
 utilities and can
 be
 likewise
 derived
 from
 the
 other
 intersection

points
shown.



An
Erroneous
Criticism
of
the
Four‐utility
Model



Tribe13
 argues
 that
 the
 utilities
 of
 trial
 outcomes
 are
 themselves
 functions
 of
 the

perceived
probability
of
guilt,
making
the
fully
correct
version
of
the
standard
of
proof

formula
 intractable
 and
 of
 little
 use.
 
 DeKay14
 evidently
 accepts
 this
 claim
 that

outcome
 utilities
 depend
 on
 the
 probability,
 and
 creates
 an
 elaborate
 argument
 that

purports
 to
 show
 it
 does
 not
 create
 difficulties
 with
 the
 standard
 of
 proof
 formula

given
certain
assumptions
he
argues
are
reasonable
about
the
functional
form
of
this

relationship.



Confusion
on
this
issue
appears
to
arise
from
a
misconception
of
utility
theory.

Tribe

summarizes
 the
 perceived
 problem
 as
 “…the
 trier
 might
 justly
 regard
 as
 worse
 the

erroneous
 conviction
 of
 a
 man
 to
 whose
 guilt
 he
 had
 attached
 a
 probability
 of
 just

over
4/7
than
the
erroneous
conviction
of
one
whose
guilt
had
seemed
to
be
virtually

certain.”
 
 
 
 This
 is
 essentially
 identical
 to
 DeKay’s
 statement
 of
 the
 problem:
 “…the

disutility
 associated
 with
 acquitting
 a
 truly
 guilty
 defendant
 might
 be
 perceived
 as

greater
if
the
subjective
probability
of
guilt
equaled
0.70
at
the
time
of
acquittal
than
if

the
subjective
probability
of
guilt
equaled
0.30
at
the
time
of
acquittal.”



Both
 these
 authors
 confuse
 outcome
 utilities
 with
 expected
 utilities.
 
 The
 utilities

associated
 with
 the
 outcomes,
 ,
 reflect
 the
 societal
 value
 of
 these

outcomes
 given
 that
 they
 happen
 with
 absolute
 certainty.
 
 In
 contrast,
 the
 expected

utilities
 of
 each
 verdict,
 and
 ,
 are
 functions
 of
 the
 perceived
 probability
 of

guilt
 as
shown
in
equations
(3)
and

(4).

Figure
2
shows
graphically
the
expected

utility
 of
 each
 verdict
 as
 a
 function
 of
 the
 juror’s
 perceived
 probability
 of
 guilt
 (and

just
 to
 be
 precisely
 clear,
 this
 is
 the
 juror’s
 posterior
 assessment
 of
 the
 likelihood
 of

guilt
–
after
hearing
all
the
evidence,
arguments
of
the
attorneys,
and
views
of
his
or

her
fellow
jurors).





In
 Figure
 2
 it
 is
 seen
 that
 the
 expected
 utility
 of
 a
 guilty
 verdict
 is
 an
 increasing

function
of
 ,
and
the
expected
utility
of
a
not
guilty
verdict
is
a
decreasing
function

of
 .

Both
functions
are
linear.





Consider
the
first
line,
the
expected
utility
of
a
guilty
verdict.

This
line
says
that
the

higher
 the
 juror’s
 perceived
 probability
 of
 guilt,
 the
 higher
 is
 the
 expected
 utility
 of

voting
 conviction.
 
 Similarly,
 the
 not
 guilty
 line
 says
 that
 the
 higher
 the
 juror’s

perceived
 probability
 of
 guilt,
 the
 lower
 is
 the
 expected
 utility
 of
 voting
 acquittal.


Together,
they
entail
that
the
upper
v‐shaped
envelope
of
these
two
lines
depicts
the


























































13
Laurence Tribe, op.cit., n. 168 (1971).
14
Michael DeKay, “The Difference between Blackstone-Like Error Ratios and Probabilistic Standards of
Proof,” Law & Doc. Inquiry 95(n.21) at 114 n. 46 (1996).
 8

societal
 utility‐maximizing
 decision
 alternative.
 
 For
 perceived
 probability
 of
 guilt


below
the
point
where
they
intersect,
 ,
societal
expected
utility
is
maximized
by
a

not
 guilty
 verdict;
 for
 perceived
 probability
 of
 guilt
 above
 this
 threshold,
 societal

expected
utility
is
maximized
by
a
guilty
verdict.



So
 here
 is
 where
 the
 Tribe/DeKay
 argument
 falls
 down:
 It
 is
 absolutely
 true,
 to

paraphrase
 DeKay,
 that
 the
 disutility
 to
 society
 of
 acquitting
 a
 defendant
 will
 be

greater
for
 
than
for
 .

This
is
saying
nothing
more
nor
less
than
that

the
green
line
in
Figure
2
is
sloped
negatively
with
respect
to
 :
society
has
a
lower

expected
utility
for
an
acquittal
that
is
more
likely
(as
perceived
by
their
designees,
the

jurors)
to
be
a
false
acquittal
than
a
true
one.

Perfectly
sensible.



Still,
 our
 paraphrase
 of
 DeKay
 is
 missing
 something
 important.
 
 DeKay
 slips
 in
 a

descriptor
 of
 the
 acquittal
 characterizing
 it
 as
 the
 acquittal
 of
 “a
 truly
 guilty

defendant.”
 
 Yet
 the
 societal
 utility
 of
 acquitting
 a
 truly
 guilty
 defendant
 is
 entirely

independent
 of
 the
 probability
 of
 guilt
 perceived
 by
 the
 trial
 juror.
 
 A
 truly
 guilty

person
 has
 been
 acquitted.
 
 Society
 bears
 this
 cost,
 and
 the
 cost
 is
 the
 same

irrespective
of
the
trial
jury’s
perceptions.

It
is
the
cost
of
the
outcome.



Tribe
 likewise
 slips
 in
 a
 key
 word:
 “erroneous.”
 
 
 He
 then
 argues
 that
 an
 erroneous

conviction
may
be
“justly
regarded
as
worse”
if
the
perceived
probability
is
just
above

Tribe’s
threshold
of
4/7
than
if
it
is
nearly
1.

But
again,
the
societal
utility
is
actually

the
 same
 in
 either
 case.
 
 The
 outcome
 is
 a
 defendant
 falsely
 convicted.
 
 And
 society

bears
the
full
cost
of
this.

Independent
of
jury
beliefs.

What
Tribe’s
argument
plays

on
is
the
greater
regret
a
juror
might
feel
if,
upon
somehow
learning
the
truth
that
a

defendant
is
in
fact
innocent,
he
or
she
remembered
their
assigned
probability
of
guilt

as
having
been
lower
rather
than
higher.

But
this
has
no
bearing
on
the
actual
societal

utility
associated
with
this
event.

The
appropriately‐applied
societal
utility
in
Tribe’s

hypothetical
is
the
societal
utility
of
the
particular
outcome
of
a
false
conviction.



Tribe
and
DeKay
seem
to
be
running
together
semantic
and
epistemic
questions
that

are
 quite
 distinct.
 
 The
 costs
 of
 falsely
 convicting
 or
 falsely
 acquitting
 someone
 are

wholly
 independent
 of
 how
 probable
 their
 guilt
 appears.
 Even
 if
 defendant’s
 guilt

appears
 overwhelmingly
 likely
 to
 the
 jurors,
 and
 they
 therefore
 convict
 him,
 such
 a

conviction
(if
the
defendant
is
factually
innocent)
is
every
bit
as
costly
as
if
they
had

convicted
him
when,
on
the
evidence,
he
appeared
to
be
innocent.



The
 arguments
 of
 Tribe
 and
 DeKay
 in
 no
 way
 alter
 the
standard
 of
 proof
 as
 derived

from
equating
the
two
utilities
in
equation
(2).

If,
to
use
DeKay’s
example,
we
assume

,
 and
 if
 the
 standard
 of
 proof
 is
 as
 showing
 in
 Figure
 2
 (i.e.,
 slightly
 less

than
 0.70),
 we
 see
 that
 the
 juror
 should
 vote
 acquittal,
 since
 this
 alternative
 has
 a

higher
expected
utility
than
conviction.

If
instead
the
juror’s
perceived
probability
is

,
 while
 it
 is
 true
 an
 acquittal
 in
 this
 case
 delivers
 lower
 expected
 societal

utility
than
if
 ,
acquittal
nonetheless
delivers
higher
expected
societal
utility

 9

than
 conviction.
 
 If
 ,
 the
 situation
 reverses
 and
 a
 higher
 expected
 societal

utility
is
delivered
by
conviction.





In
 other
 words,
 the
 model
 fully
 accounts
 for
 the
 fact
 that
 societal
 expected
 utilities

may
depend
strongly
on
 .

But
these
are
expected
utilities.

There
is
no
need
to
adjust

the
outcome
utilities
to
reflect
any
dependence.



The
case
is
made
even
clearer
by
considering
two
extreme
cases.

If
a
juror’s
perceived

probability
of
guilt
is
zero,
it
can
be
seen
from
equation
(3)
that
the
expected
utility
of

conviction
 is
 simply
 .
 
 That
 is,
 if
 the
 juror
 is
 absolutely
 certain
 the
 defendant
 is

innocent,
society
would
like
him
or
her
to
behave,
in
their
decision
making,
as
if
the

utility
 of
 voting
 conviction
 is
 precisely
 the
 utility
 society
 places
 on
 a
 certain
 false

conviction;
 and
 society
 would
 like
 him
 or
 her
 to
 behave
 as
 if
 the
 utility
 of
 voting

acquittal
is
precisely
the
utility
society
places
on
a
certain
true
acquittal.

Similarly,
for

,
the
expected
utilities
of
conviction
and
acquittal
are
precisely
the
raw
outcome

utilities
 and
 .


This
is
entirely
consistent
with
the
assertion
that
the
outcome

utilities
 reflect
 the
 societal
 value
 of
 these
 outcomes
 given
 that
 they
 happen
 with

absolute
certainty.



In
 summary,
 Tribe’s
 assertion
 that
 a
 proper
 accounting
 would
 require
 making
 the

outcome
utilities
a
function
of
perceived
probability
of
guilt
is
an
erroneous
one,
and

the
 fully
 comprehensive
 formula
 for
 calculating
 the
 standard
 of
 proof
 
 remains
 that

shown
in
equation
(2).



Despite
 a
 broad,
 but
 by
 no
 means
 universal,
 acquiescence
 in
 this
 principled

mechanism
 for
 setting
 the
 standard
 of
 proof
 non‐arbitrarily,
 there
 are
 precious
 few

serious
discussions
in
the
literature
of
what
the
values
for
these
four
utilities
might
or

should
be.

We
will
examine
some
of
those
shortly.

For
now,
it
is
important
to
note

that
 the
 majority
 of
 scholars
 who
 have
 conjectured
 about
 these
 utilities
 continue
 to

focus
 their
 attention
 on
 only
 two
 of
 the
 four
 outcomes:
 false
 convictions
 and
 false

acquittals.15





There
 are
 doubtless
 many
 reasons
 why
 this
 is
 so.
 
 Among
 them:
 a).
 it
 simplifies
 the

calculation,
requiring
consensus
only
on
the
costs
of
two
of
the
four
outcomes;
b).
it

speaks
to
the
centuries‐long
tendency
to
suppose
that
the
only
outcomes
at
trial
that

really
 matter
 are
 the
 erroneous
 verdicts16;
 and
 c).
 it
 obviates
 the
 need
 to
 reach
 a


























































15
This tactic of giving short shrift to the utilities of correct outcomes is succinctly summed up by Schauer and
Zeckhauser when they write: “[W]e believe that a properly conceived regret matrix, which measures
comparative utility losses from different mistakes, fully captures the proper comparison between two
decisions.” (Frederick Schauer and Richard Zeckhauser, “On the Degree of Confidence for Adverse
Decisions,” 25 J LEGAL STUD, 27 at 34n.9 (1996). Or, stressing the same point: “It is well understood [sic]
that a legal system’s choice among these standards [of proof] is an exercise in trading off the harms that flow
from different types of error.” (Ibid. at 34) The same simplification can be seen in the influential evidence text
by Richard Lempert and Stephen Saltzburg, A MODERN APPROACH TO EVIDENCE (2nd ed., 1982).
16
A search among scholarly articles on LexisNesis vividly illustrates the syndrome. While ‘false conviction”
and ‘wrongful conviction’ yield 497 hits, ‘true’ and ‘rightful conviction’ yield a total of 83. While ‘false’ or
 10

consensus
 about
 the
 specific
 utilities
 for
 the
 erroneous
 verdicts,
 since
 it
 is
 sufficient

merely
 to
 decide
 what
 the
 ratio
 of
 their
 (dis)utilities
 is.
 The
 last
 point
 is
 especially

salient
because
the
legal
profession
continues
to
believe
(wrongly,
in
our
opinion)
that

Blackstone
already
settled
that
question
a
long
time
ago.



Blackstonian
Fantasies



We
will
argue
in
this
section
that
focusing
exclusively
on
the
utilities
of
the
erroneous

outcomes
 is
 not
 just
 an
 oversimplification
 of
 a
 complex
 situation
 but
 a
 seriously

distorting
 mistake.
 Indeed,
 it
 commits
 numerous
 errors
 at
 the
 same
 time
 ‐‐some

already
identified
by
other
writers,
others
not‐‐
and
at
several
distinct
levels.

We
will

attempt
to
disentangle
several
of
them.





a).
 The
 Blackstone
 thesis,
 which
 has
 long
 been
 a
 key
 article
 of
 faith
 among
 legal

scholars,
does
not
have
anything
to
say
about
the
utilities
of
the
outcomes
of
a
trial,

strictly
 speaking.
 
 Rather,
 as
 one
 can
 see
 by
 its
 very
 formulation
 ("Better
 that
 ten

guilty
 persons
 escape
 than
 that
 one
 innocent
 suffer,"17),
 it
 addresses
 itself
 to
 the

relative
 frequency
 with
 which
 errors
 of
 false
 conviction
 and
 false
 acquittal
 should

occur.
 
 It
 tells
 us
 that
 a
 necessary
 condition
 for
 an
 acceptable
 system
 of
 criminal

justice
 is
 that
 there
 are
 at
 least
 ten
 false
 acquittals
 for
 every
 false
 conviction.
 
 This

proportion,
of
course,
has
become
known
as
the
Blackstone
ratio.





Blackstone
 did
 not
 fathom
 that
 there
 is
 no
 way
 of
 guaranteeing
 in
 advance
 that
 this

ratio
of
mistakes
will
ever
occur
in
any
long,
random
sequence
of
criminal
trials.

This

is
 because
 the
 ratio
 of
 false
 acquittals
 to
 false
 convictions
 not
 only
 depends
 on
 the

standard
of
proof
that
we
employ
but
also
on
such
uncontrollable
variables
as
the
de

facto
proportion
of
truly
innocent
and
truly
guilty
defendants
who
are
brought
to
trial,

the
 robustness
 of
 the
 evidence
 presented
 by
 the
 two
 sides,
 and
 the
 perspicacity
 of

particular
juries
in
evaluating
that
evidence.18





However,
 many
 scholars
 have
 thought
 that,
 with
 a
 subtle
 twist
 in
 the
 content
 of

Blackstone’s
thesis,
we
can
circumvent
that
problem.

Instead
of
reading
Blackstone
as

specifying
an
acceptable
frequency
of
errors,
they
instead
suppose
that
the
Blackstone

ratio
addresses
the
values
that
should
drive
the
selection
of
the
verdict
in
a
criminal

trial.

Specifically,
they
think
that
the
Blackstone
thesis
is,
or
could
be,
saying
that
the

disutilities
 of
 a
 false
 conviction
 are
 ten
 times
 greater
 than
 the
 disutilities
 of
 a
 false

acquittal.
 19
If
that
were
what
he
is
saying,
and
if
we
likewise
were
to
accept
Kaplan’s












































































































































































‘wrongful’ acquittal produces 194 hits, ‘true’ or ‘rightful’ acquittal yields only 21. In sum, the ratio of
references to erroneous verdicts to references to true verdicts is just shy of 7:1. Blackstone would presumably
be pleased.
17
4 William Blackstone, COMMENTARIES 358.
18
See Larry Laudan, “The Elementary Epistemic Arithmetic of Criminal Justice,” 5 EPISTEME, 282 (2008).
19
It should be noted, as we will argue below, that this way of speaking of ratios may not even be intelligible.
Just as it would be silly to say that it is half as cold this afternoon (4 degrees F) as it was this morning (wen it
was 2 degrees F, so it will generally be opaque what we mean by saying that a utility of twenty is ten times
greater than a utility of two.
 11

simplified
 utility
 calculation,
 then
 we
 would
 have
 all
 we
 needed
 to
 calculate
 a

“rational”
standard
of
proof.

Doing
the
requisite
calculation
(while
assuming
that
the

disutilities
of
a
false
conviction
are
ten
times
worse
than
those
of
a
false
acquittal),
we

discover
that
the
standard
of
proof
for
criminal
trials
should
be
about
91%.




One
of
the
attractions
of
this
approach
has
been
its
intuitive
appeal;
it
shows
that
we

have
 been
 right
 in
 using
 a
 very
 demanding
 standard
 like
 proof
 beyond
 a
 reasonable

doubt.
 
 Prior
 to
 the
 emergence
 of
 all
 these
 calculations,
 of
 course,
 no
 one
 had
 ever

suggested
specifically
that
BARD=91%
but
it
is
clear
that
both
BARD
(which
is
usually

glossed
 by
 scholars
 as
 falling
 within
 the
 90‐95%
 range)
 and
 a
 standard
 of
 91%

apparent
 guilt
 are
 close
 neighbors.20
 
 In
 sum,
 a
 utility
 calculation
 –viewed
 from
 this

perspective‐‐exonerates
what
we
believed
all
along;
to
wit,
that
a
careful
appraisal
of

the
respective
costs
of
the
two
errors
at
trial
would
show
that
the
standard
must
be

demanding
 and
 much
 more
 prone
 to
 one
 sort
 of
 error
 than
 the
 other.
 
 No
 one
 quite

shouted,
“Long
live
the
status
quo!”
when
these
results
emerged.
Still,
the
congruence

between
what
the
legal
system
has
long
held
as
sacred
and
the
reduction
of
the
utility

calculation
to
the
only
outcomes
recognized
in
the
Blackstone
ratio
has
done
nothing

but
reinforce
the
conventional
wisdom.



b).
 This
 myopic
 focus
 on
 the
 expected
 losses,
 and
 thus
 on
 the
 ratio
 of
 erroneous

verdicts
 as
 the
 only
 determinant
 of
 the
 standard
 of
 proof,
 is
 really
 quite
 stunning

when
 one
 thinks
 about
 it.
 Specifically,
 we
 are
 asked
 to
 put
 wholly
 to
 one
 side
 any

consideration
 of
 the
 respective
 utilities
 associated
 with
 correct
 outcomes
 at
 trial.

Having
defined
the
standard
exclusively
in
terms
of
the
supposedly
acceptable
ratio
of

errors,
 we
 happily
 discount
 altogether
 the
 benefits
 of
 true
 acquittals
 and
 true

convictions.

That
would
appear
to
be
an
egregious
miscalculation.

After
all,
it
is
likely

that
true
convictions
have
important
deterrent
effects
and
it
is
indisputable
that
they

have
 important
 incapacitative
 consequences.
 
 Likewise,
 acquittals
 for
 the
 truly

innocent
would
seem
to
have
important
moral,
practical
and
symbolic
consequences.


Leaving
such
weighty
matters
out
of
the
calculation
of
the
standard
of
proof
would
be

a
bad
joke,
were
it
not
so
ubiquitous.
Clearly,
something
has
gone
badly
awry
in
this

on‐going
preference
for
(1)
over
(2)
as
the
definition
of
the
standard
of
proof.





Author
 uTC
 uFC
 uFA
 uTA
 Implied
SoP
 Rank
Ordering
of
Outcomes

Tribe
 1
 0
 .5
 .67
 57%
 uTC>uTA>uFA>uFC

Lillquist
 1
 ‐10
 ‐1
 0
 83%
 uTC>uTA>uFA>uFC

Milanich
 .9
 0
 .1
 1
 55%
 uTA>uTC>uFA>uFC

Nagel
 10
 ‐100
 ‐10
 100
 95%
 uTA>uTC>uFA>uFC



Figure
3.
Estimates
of
Utilities
of
Outcomes
by
Various
Scholars21


























































20
See below, where we point out that most existing empirical studies of the meaning of proof beyond a
reasonable doubt [hereafter: BARD] that assess utilities suggest that jurors locate it at a much lower level than
academic lawyers generally think they should.
21
The relevant sources are: Tribe, op. cit.; Lillquist, op. cit.; Patricia G. Milanich, Decision Theory and
 12





To
 get
 a
 quick
 and
 dirty
 sense
 of
 just
 how
 misleading
 it
 can
 be
 to
 leave
 the
 positive

utilities
 out
 of
 the
 calculation,
 consider
 three
 examples
 deriving
 from
 the
 recent

literature.
In
the
article
already
cited,
Tribe
explored
the
following
utilities
for
the
four

outcomes:

uTC=1
 
 uFC=0
 
 uFA=.5

 uTA=.67.

The
standard
of
proof
that
would
follow
(using
(2))
from
Tribe’s
utilities
is
57%,
even

though
 the
 utility
 of
 a
 false
 conviction
 proposed
 by
 Tribe
 is
 sharply
 lower
 than
 the

utility
of
a
false
acquittal.
Years
later,
Erik
Lillquist
explored
a
not
implausible
ranking

of
these
utilities
that
would
appear
to
be
fully
Blackstone
compliant:

uTC=1
 
 uFC=‐10
 uFA=‐1

 uTA=0.22

Using
the
same
formula
as
before,
the
standard
of
proof
unique
to
this
set
of
utilities
is

83%.
 
 To
 jack
 the
 standard
 up
 into
 the
 range
 of
 90%,
 we
 would
 have
 to
 set
 the

disutility
 of
 a
 false
 conviction
 at
 about
 ‐20,
 thereby
 doubling
 “the
 Blackstone
 ratio.”


One
 further
 example
 should
 secure
 the
 point.
 
 Suppose
 that,
 recognizing
 that
 a
 true

conviction
is
a
very
desirable
outcome
and
that
a
false
conviction
is
a
very
undesirable

one,
we
consider
this
set
of
assignments:



uTC=10
 uFC=‐10
 uFA=‐1

 uTA=0.

The
 only
 change
 here
 from
 the
 previous
 set
 is
 an
 increase
 in
 the
 utility
 of
 a
 true

conviction.

Now
we
have
a
standard
of
proof
of
about
48%,
even
though
the
disutility

of
a
false
conviction
is,
in
the
spirit
of
Blackstone,
vastly
greater
than
the
disutility
of
a

false
 acquittal.
 
 What
 this
 example
 shows,
 as
 would
 indefinitely
 many
 others,
 is
 that

the
neo‐Blackstonian
strategy
of
assigning
much
higher
disutilities
to
false
convictions

than
to
false
acquittals
need
not
guarantee
that
the
appropriate
standard
of
proof
will

be
 very
 high,
 provided
 that
 the
 standard
 takes
 into
 account
 prima
 facie
 plausible

utilities
for
true
outcomes.




Indeed,
 the
 situation
 is
 even
 more
 extreme
 than
 these
 examples
 suggest.
 
 It
 is

provable
 that,
 even
 if
 we
 make
 the
 disutility
 of
 a
 false
 conviction
 ten
 times
 greater

then
the
disutility
of
a
false
acquittal,
any
desired
standard
of
proof
between
0
and
1

can
 be
 generated
 by
 an
 appropriate
 assignment
 of
 utilities
 to
 the
 two
 positive

outcomes.23
 
 In
 short,
 the
 use
 of
 the
 Blackstone
 ratio
 is
 not
 a
 sufficient
 condition
 for

guaranteeing
 a
 demanding
 standard
 of
 proof.
 
 Neither
 is
 the
 Blackstone
 ratio
 a

necessary
condition
for
a
tough
standard,
since
standards
of
proof
greater
than
90%

can
be
generated
when
the
uFC=
2
x
uFA.

For
instance,
the
set
(uTC=1,
uFC=‐2,
uFA=‐










































































































































































Standards of Proof, 5 LAW & HUM. BEHAV. 87, 93 (1981);
and
Stuart Nagel et al., Decision Theory and
Juror Decision-Making, in THE TRIAL PROCESS
22
Lillquist, op cit., at 109.
23
Given any set of four utilities (provided uFC<uFA), which generates any particular standard of proof
(however low), it is possible to convert that set into one in which uFC=10 x uFA, while preserving the
standard of proof at its prior level. For instance, by multiplying each utility in the set by ten and then adding
a (defined as a= {10 x uFA)-uFC)/9} to each utility –both operations permitted by the rules governing
manipulation of utilities—we obtain a set of utilities that is Blackstone compliant and that produces a standard
of proof equal to the standard derived from the prior, unBlackstonian utilities.
 13

1,
 and
 uTA=20)
 generates
 a
 standard
 of
 about
 92%.
 Even
 if
 we
 assign
 the
 same

disutilities
to
false
acquittals
and
false
convictions,
it
is
possible
to
have
a
standard
of

proof
of
90%
(for
instance,
uTC=1,
uFC=uFA=‐2,
and
uTA=25).





For
 such
 reasons,
 we
 must
 lay
 to
 one
 side
 the
 idea
 that
 we
 can
 read
 off
 the

appropriate
standard
of
proof
from
a
knowledge
(supposing
we
had
such
knowledge)

of
 what
 society
 takes
 to
 be
 an
 appropriate
 value
 for
 the
 Blackstone
 ratio.
 Assigning

much
 higher
 disutilities
 to
 false
 convictions
 than
 to
 false
 acquittals
 in
 and
 of
 itself

does
nothing
to
locate
the
standard
close
to
1.0.


We
have
to
take
all
four
utilities
into

account;
 fixing
 the
 value
 for
 any
 pair
 of
 them
 –which
 is
 what
 Kaplan’s
 (1)
 urges
 on

us—leaves
the
standard
drastically
underdetermined.





c).
 It
 is
 bad
 enough
 that
 respecting
 the
 Blackstone
 ratio
 does
 not
 guarantee
 a

demanding
standard
of
proof
(once
we
factor
in
plausible
values
for
the
utilities
of
the

positive
 outcomes).
 Worse
 still,
 there
 is
 a
 serious
 conceptual
 error
 going
 on
 in
 the

rush
 to
 embrace
 some
 version
 or
 other
 of
 the
 Kaplan
 formula
 for
 the
 standard
 of

proof.
 
 It
 is
 clear
 that
 formula
 (1)
 invites
 us
 to
 consider
 a
 ratio
 of
 two
 utilities.
 
 The

trouble
is
that
utilities,
like
temperatures,
do
not
readily
lend
themselves
to
division

by
 other
 utilities.
 
 Consider,
 by
 way
 of
 illustration,
 the
 four
 utilities
 proffered
 by

Lillquist.
 Here,
 the
 ratio
 formed
 by
 dividing
 the
 disutility
 of
 a
 false
 acquittal
 by
 the

disutility
of
a
false
conviction
appears
to
be
1:10.

However,
it
is
a
feature
of
all
utility

sets
 that
 operations
 like
 multiplication
 and
 addition
 can
 be
 performed
 on
 each

member
of
the
set
without
changing
the
matrix.

So,
if
we
add
20
to
each
of
Lillquist’s

utilities,
we
get
the
following
configuration
of
utilities:

uTC=21
 uFC=10
 uFA=19
 uTA=20.

Substituted
 into
 the
 Tribe
 formula
 for
 the
 standard
 of
 proof,
 these
 utilities
 yield

precisely
 the
 same
 standard
 as
 Lillquist’s
 original
 values
 (83%).
 But
 note
 what
 has

happened
to
the
erstwhile
“Blackstone
ratio.”
A
“ratio”
of
1:10
has
become
a
ratio
of

almost
1:2.

Yet,
from
the
point
of
view
of
decision
theory,
this
set
of
utilities
and
the

original
Lillquist
set
are
formally
indistinguishable.



The
general
point
is
that
whether
the
utilities
of
a
false
acquittal
and
a
false
conviction

exhibit
an
order
of
magnitude
difference
between
them
is
of
no
significance
whatever

for
 the
 calculations
 of
 utility
 theory,
 since
 that
 proportion
 can
 be
 changed
 at
 will
 by

suitable
 additions
 and
 multiplications.
 The
 moral
 is
 that
 ratios
 of
 utilities
 are
 very

ephemeral
things.

We
should
be
leery
of
any
definition
of
a
standard
of
proof
that,
like

Kaplan’s,
invokes
such
ratios
since
such
ratios
themselves
are
ill‐defined.




Still,
even
though
ratios
of
utilities
are
suspect
creatures,
ratios
of
differences
between

select
utilities
are
not.

That,
indeed,
is
a
key
attraction
of
equation
(2).
As
we
can
see

in
 the
 denominator,
 there
 is
 a
 ratio
 of
 two
 utility
 differences.
 The
 numerator
 of
 that

ratio
signifies
the
differences
between
the
utilities
associated
with
guilty
defendants;

the
 denominator
 is
 formed
 by
 differences
 in
 the
 utilities
 of
 what
 can
 befall
 innocent

defendants.24

This
ratio
of
differences
is
absolutely
fundamental
for
reasons
we
shall


























































24
Multiplication of, or additions to, the utilities making up this ratio do not alter the value of the ratio itself.
 14

see
shortly.

For
now,
it
is
sufficient
to
note
that
ratios
of
utility
differences
don’t
suffer

from
 the
 problems
 that
 ratios
 of
 utilities
 do
 and
 that
 the
 smaller
 this
 ratio
 of

differences
 is
 (that
 is,
 the
 larger
 the
 denominator
 relative
 to
 the
 numerator),
 the

higher
the
standard
of
proof
will
be.



Thus
 far,
 we
 have
 suggested
 that
 there
 is
 no
 clear
 connection
 between
 the

preoccupations
 driving
 the
 classical
 Blackstone
 ratio
 and
 the
 task
 of
 settling
 on
 a

plausible
set
of
utilities.

That
doesn’t
yet
exhaust
the
errors
in
the
use
of
equation
(1)

to
define
the
standard
of
proof
but
it
does
show,
we
think,
that
one
important
motive


for
finding
(1)
attractive
–namely,
that
it
seems
readily
to
incorporate
something
like

the
Blackstone
concerns—is
a
chimera.



d).
There
is,
however,
one
option
open
to
those
who
are
committed
to
the
idea
that
the

standard
 should
 be
 derived
 from
 the
 ratio
 of
 errors.
 
 We
 have
 stressed
 that
 the

utilities
 of
 positive
 outcomes
 should
 not
 be
 ignored.
 
 There
 would
 appear
 to
 be
 one

way
 in
 which
 the
 proponents
 of
 (1)
 could
 accommodate
 this
 concern.
 
 The
 ratio

formulated
in
(1)
is
a
ratio
of
the
disutilities
of
the
two
erroneous
outcomes.
There
is

nothing
 whatever
 that
 prevents
 us
 from
 taking
 the
 positive
 attributes
 of
 the
 correct

outcomes,
 stripping
 them
 away
 from
 the
 positive
 outcomes,
 and
 assigning
 their

contraries
to
the
two
negative
outcomes
that
figure
in
(1).

Consider
an
example.

One

prominent
virtue
of
a
true
conviction
is
that
the
guilty
party
receives
his
just
deserts.

We
 could
 count
 that
 among
 the
 utility‐enhancing
 attributes
 of
 a
 true
 conviction;

alternatively,
we
could
say
that
falsely
acquitting
a
guilty
defendant
fails
to
give
him

his
just
deserts
and
count
that
as
a
disutility‐enhancing
feature
of
false
acquittals.





If
 we
 applied
 this
 strategy
 to
 all
 the
 traits
 of
 true
 verdicts,
 transforming
 them
 from

utility‐enhancing
 properties
 of
 those
 true
 verdicts
 into
 disutility‐enhancing
 traits
 of

false
verdicts,
we
could
try
to
arrive
at
a
situation
in
which
the
true
verdicts
–stripped

of
all
their
assets,
as
it
were—come
to
exhibit
a
utility
of
zero
while
the
disutilities
of

false
 verdicts
 continue
 to
 rise.
 
 If
 this
 strategy
 were
 successful,
 we
 could
 reduce

equation
 (2)
 to
 equation
 (1)
 since
 the
 ratio
 of
 differences
 in
 (2)
 would
 simply
 be

transformed
 to
 the
 ratio
 uFA/uFC,
 as
 in
 (1).
 
 This
 tactic
 quickly
 becomes
 unglued,

however.

It
is
easy
enough
to
say
that
we
will
debit
“failure
to
deliver
just
deserts”
to

a
 false
 acquittal
 instead
 of
 crediting
 the
 delivery
 of
 just
 deserts
 to
 a
 true
 conviction.

The
problem
arises
as
soon
as
we
realize
that
it
is
not
only
false
acquittals
that
fail
to

give
 just
 deserts
 to
 the
 truly
 guilty.
 
 So
 do
 true
 acquittals
 and
 false
 acquittals.
 
 We

could
 readily
 add
 to
 the
 liabilities
 of
 a
 false
 acquittal
 that
 it
 too
 fails
 to
 give
 just

deserts.
Still,
how
do
we
capture
in
this
Kaplan‐esque
scenario,
the
fact
that
a
positive

outcome
 
 ‐‐a
 true
 acquittal—fails
 to
 deliver
 an
 important
 benefit
 associated
 with
 a

true
conviction?
The
only
reasonable
way
to
do
that
is
to
reduce
the
utility
of
a
true

acquittal
relative
to
a
true
conviction
(or,
alternatively,
raise
the
utility
of
the
latter).


But
either
modification
would
undermine
the
whole
enterprise,
since
the
object
of
this

maneuver
 was
 to
 render
 true
 convictions
 and
 true
 acquittals
 as
 neutered
 and
 thus

dispensable,
each
possessing
a
utility
of
zero.





 15

For
such
reasons,
we
conclude
that
there
is
no
avoiding
taking
the
utilities
of
all
four

outcomes
into
account
in
shaping
a
rational
standard
of
proof.
It
is
the
ratio
of
utility

differences
rather
than
the
ratio
of
the
utilities
of
the
erroneous
verdicts
that
should

be
our
focus.



Making
a
Stab
at
Fixing
the
Utilities
of
the
Outcomes



If
we
are
minded
to
try
to
determine
what
all
four
utilities
are,
there
are
at
least
two

different
 general
 directions
 in
 which
 we
 might
 go.
 
 One
 of
 them
 would
 be
 purely

descriptive:
 we
 could
 ask
 citizens
 or
 jurors
 or
 legislators
 questions
 that
 would
 elicit

what
 utilities
 they
 implicitly
 assign
 to
 these
 outcomes.
 
 The
 other
 tactic
 would
 be

much
 more
 normative;
 it
 would
 involve
 trying
 to
 settle
 what
 utilities
 it
 would
 be

rational
to
assign
to
these
outcomes,
given
a
careful
assessment
of
what
consequences

are
 associated
 with
 each.
 We
 think
 a
 combination
 of
 both
 these
 approaches
 is

preferable.

Ultimately,
it
is
(we
suppose)
legislators
who
have
to
make
the
decision

about
 what
 utilities
 to
 assign
 to
 these
 outcomes,
 for
 they
 are
 the
 ones
 who
 bear
 the

responsibility
for
setting
the
standard
of
proof.

It
manifestly
should
not
be
judges
or

jurors
 who
 are
 left
 to
 make
 such
 decisions.
 Still,
 legislators
 cannot
 responsibly

discharge
that
task
until
and
unless
they
have
reflected
carefully
upon
the
costs
and

benefits
of
the
four
outcomes.




To
 that
 end,
 we
 think
 it
 would
 be
 helpful
 to
 identify
 the
 kinds
 of
 factors
 that
 might

shape
a
rational
agent’s
assignment
of
utilities
to
the
outcomes
and
thus
those
crucial

appraisals
of
difference.

We
will
do
so
in
terms
of
figure
4.






Consequence
 TC
 FC
 TA
 FA

Felon
gets
just
deserts
 Yes↑
 No↓
 No↓
 No↓

Crime
reduction
by
incapacitation
 Yes↑
 No↓
 No↓
 No↓

Crime
reduction
by
deterrence
 Yes↑
 Yes↑
 No↓
 No↓

Grievous
harm
to
innocent
defendant
 No↑
 Yes↓
 No↑
 No↑

Closure
for
victims
 Yes↑
 Yes↑
 No↓
 No↓

Modifiable
on
appeal
 Yes↓
 Yes↑
 No↑
 No↓

Truth
and
justice
served
 Yes↑
 No↓
 Yes↑
 No↓

Partial
restoration
of
defendant’s
reputation
 No↑
 No↓
 Yes↑
 Yes↓



Figure
4.
Some
Salient
Consequences
of
the
Four
Outcomes



(The
 upward
 arrows
 indicate
 benefits
 that
 can
 be
 associated
 with
 the
 outcome
 in

question;
 downward
 arrows
 point
 to
 costs.)
 Focus,
 first,
 on
 the
 two
 elements
 in
 the

numerator
 of
 the
 ratio
 in
 (2):
 true
 convictions
 and
 false
 acquittals.
 True
 convictions

have
 a
 host
 of
 attractions.
 Justice
 is
 done;
 victims
 are
 given
 closure;
 the
 interests
 of

truth
 are
 served;
 bad
 people
 get
 their
 just
 deserts;
 crimes
 are
 prevented
 by

incarcerating
 wrong‐doers,
 many
 of
 whom
 are
 already
 repeat
 offenders;
 and
 crimes

by
third
parties
are
possibly
deterred
by
the
example
set.
By
contrast,
false
acquittals

 16

possess
 precious
 few
 redeeming
 features.
 
 On
 the
 other
 hand,
 they
 don’t
 have
 many

terrible
features
either.

It
is
true
that
a
false
acquittal
means
that
a
guilty
person
will

not
receive
his
just
deserts,
will
be
released
back
onto
the
streets
to
commit
additional

crimes,
 and
 that
 victims
 will
 not
 get
 any
 sense
 of
 closure.
 
 But
 we
 have
 already

included
 the
 avoidance
 of
 these
 features
 among
 the
 benefits
 of
 a
 true
 conviction.


Counting
 them
 again
 under
 the
 costs
 of
 a
 false
 acquittal
 would
 be
 an
 unacceptable

case
of
double
counting,
and
so
should
be
firmly
resisted.





One
 could
 argue,
 of
 course,
 that
 these
 negative
 features
 would
 as
 appropriately
 be

included
under
the
costs
of
a
false
acquittal
as
under
the
benefits
of
a
true
conviction.

The
 fact
 is
 that
 it
 doesn’t
 matter
 in
 which
 column
 we
 count
 them,
 since
 what
 we’re

after
 here
 is
 the
 difference
 between
 the
 utilities
 of
 a
 true
 conviction
 and
 those
 of
 a

false
acquittal.

Those
differences
will
not
be
affected
by
where
we
locate
one
or
other

of
these
traits.
In
any
event,
we
can
see
powerful
reason
to
expect
that
the
difference

in
 their
 respective
 utilities
 will
 be
 rather
 large.
 Since
 this
 difference
 forms
 the

numerator
 of
 the
 ratio
 in
 the
 denominator
 of
 (2),
 it
 should
 already
 be
 clear
 that
 we

should
not
expect
the
value
of
that
ratio
to
be
very
small.

And
if
it
is
not,
then
clearly

the
resulting
standard
of
proof
is
not
apt
to
be
very
high.




Let
us
turn
now
to
the
denominator
of
the
ratio:
the
difference
in
the
utilities
of
true

acquittals
 and
 false
 convictions.
 
 True
 acquittals
 clearly
 have
 advantages:
 truth
 and

justice
 are
 served;
 an
 innocent
 defendant’s
 reputation
 is
 at
 least
 partially
 restored.


But
such
verdicts
do
nothing
to
help
control
crime
or
to
salve
the
wounds
of
victims.

False
convictions,
by
contrast,
have
almost
nothing
to
be
said
in
their
favor
and
a
great

deal
 to
 be
 said
 against
 them.
 
 On
 the
 positive
 side,
 it
 might
 be
 claimed
 that,
 being

convictions,
 they
 help
 to
 deter
 future
 criminals
 and,
 if
 undiscovered
 as
 counterfeit,

they
 can
 give
 a
 sense
 of
 closure
 to
 victims.
 On
 the
 other
 hand,
 they
 come
 at
 an

enormous
price.

Above
all,
they
involve
depriving
an
innocent
person
of
his
liberty,

his
reputation
and
perhaps
his
self‐respect.

They
also
mean
that
the
real
felon
is
left

at
liberty
to
continue
his
life
of
crime.
If
a
true
conviction
seems
the
most
valuable
of

all
the
verdicts,
a
false
conviction
obviously
is
the
worst
of
the
four
outcomes.
Since
a

true
acquittal
is
a
positive
outcome
but
not
a
stunning
success,
it
seems
reasonable
to

suppose
that
the
difference
in
utilities
between
a
true
acquittal
and
a
false
conviction

will
likewise
be
large.



What
 does
 all
 this
 mean
 for
 the
 standard
 of
 proof?
 
 Both
 differences
 appear
 to
 be

nontrivial.
But
we
must
focus
on
the
ratio
of
those
differences.

Defenders
of
the
idea

that
 the
 standard
 of
 proof
 must
 be
 very
 demanding
 (>90%)
 have
 to
 hold
 that
 the

difference
 in
 the
 utilities
 in
 the
 denominator
 is
 at
 least
 ten
 times
 greater
 than
 the

difference
of
utilities
in
the
numerator.


That
strikes
us
as
both
arbitrary
and
a
priori.

It
 is
 unfounded
 in
 anything
 resembling
 an
 assessment
 of
 societal
 representatives’

beliefs
 about
 the
 utilities
 of
 various
 trial
 outcomes.
 Even
 if
 you
 hold
 that
 a
 false

conviction
 has
 a
 disutility
 three
 or
 four
 times
 greater
 than
 the
 utility
 of
 a
 true

conviction,
this
gets
us
nowhere
near
what
would
be
needed
to
justify
a
standard
like

proof
beyond
a
reasonable
doubt
(assuming
the
latter
to
be
>90%),
since
the
absolute

 17

utilities
 of
 a
 true
 acquittal
 and
 a
 false
 acquittal
 are
 not
 likely
 to
 differ
 by
 a
 vast

amount.




Return,
briefly,
to
the
examples
we
have
already
cited
of
scholars
who
have
explored

in
print
specific
utilities
for
the
four
outcomes.

If
we
look
at
the
ratio
of
differences
in

their
 respective
 proposals,
 they
 work
 out
 as
 follows:
 Tribe,
 0.75;
 Lillquist,
 0.2;

Milanich,
0.8;
and
Nagel,
0.1.

All
agree
that
uFC
is
less
than
the
other
outcomes
and

that
the
uFA
is
less
than
the
utility
of
a
true
conviction
or
a
true
acquittal.

In
terms
of

rank
ordering
of
the
outcomes,
the
only
disagreement
surfaces
about
whether
a
true

acquittal
or
a
true
conviction
has
the
higher
utility.

All
that
said,
it
remains
true
that

the
 ratios
 of
 the
 differences
 (and
 thus
 the
 standard
 of
 proof
 entailed)
 range
 all
 over

the
 map.
 
 Nagel’s
 and
 Lillquist’s
 utilities
 put
 the
 ratio
 of
 the
 differences
 rather
 low,

thus
ensuring
a
relatively
high
standard
of
proof
(95%
and
83%
respectively).

Tribe

and
 Milanich,
 by
 contrast,
 are
 committed
 to
 a
 relatively
 high
 value
 for
 the
 ratio
 of

differences
and
thus
to
a
much
lower
standard
of
proof
(respectively
57%
and
55%).




Given
such
discrepancies
among
estimates
of
the
ratio
of
differences
by
legal
scholars

(a
 phenomenon
 that
 shows
 up
 alike
 in
 studies
 of
 ordinary
 jurors
 or
 citizens),
 one

might
 conclude
 that
 there
 simply
 are
 profound
 and
 irreconcilable
 disagreements

about
 what
 the
 criminal
 standard
 of
 proof
 should
 be,
 deriving
 from
 disagreements

about
the
underlying
utilities
of
the
outcomes
at
trial.

We
think
it
much
too
early
to

draw
that
pessimistic
conclusion,
though
neither
would
we
wish
to
rule
it
out.



What
we
propose,
in
its
place,
is
a
more
serious
empirical
investigation
as
to
what
the

relevant
utilities
of
the
general
population
are.
The
design
of
such
a
study
would
have

to
 be
 relatively
 subtle,
 since
 the
 ordinary
 citizen
 hasn’t
 a
 clue
 what
 a
 util
 is.
 Worse,

since
 everyone
 can
 scale
 his
 own
 utilities,
 translations
 between
 the
 utility
 sets

proposed
by
different
people
require
careful
interpretation.

(For
instance,
Tribe
and

Lillquist
both
give
a
true
conviction
a
utility
of
1.0.

Still,
in
Tribe’s
case,
1.0
represents

the
 highest
 point
 on
 the
 scale:
 nothing
 could
 be
 more
 valuable;
 while
 in
 Lillquist’s

case,
a
utility
of
1.0
suggests
a
modest
positive
advantage
and
nothing
more.)








Several
researchers
have
conducted
experiments
to
figure
out
what
utilities
ordinary

citizens
 would
 assign
 to
 each
 of
 the
 four
 outcomes
 at
 trial.25
 Nagel
 reports

respondents
 adopting
 utilities
 that
 yield
 a
 standard
 of
 proof
 of
 55%.
 
 Francis
 Dane

reports
juror
utilities
compatible
with
a
standard
of
proof
of
about
51%.26
As
Lillquist

properly
 stresses:
 “relying
 on
 values
 assigned
 by
 ‘society’
 for
 the
 utility
 functions,

appears
to
lead
to
a
much
lower
standard
of
proof
than
is
conventionally
assumed.”27



























































25
Probably the best know study of this sort is Stuart Nagel et al., “Decision Theory and Juror Decision-

Making” in THE TRIAL PROCESS.
[LL:
fill
in
bibliographic
data]

26
Francis C. Dane, “In Search of Reasonable Doubt: A Systematic Examination of Selected Quantification
Approaches,” 9 LAW & HUM. BEHAV. 141, at 152 (1985).

27
Lillquist, op. cit., at 146.
 18

As interesting
as
such
work
is,
we
think
it
possible
to
fix
the
standard
of
proof
without

inquiring
 explicitly
 and
 in
 a
 one‐by‐one
 fashion
 about
 the
 utilities
 that
 jurors
 or

legislators
would
assign
to
each
of
the
four
utilities.

Indeed,
the
structural
features
of

equation
(2)
enable
us
to
short‐circuit
the
process.
Instead
of
eliciting
empirical
data

about
individual
utility
assignments,
we
can,
as
(2)
makes
clear,
infer
what
someone’s

implicit
 standard
 of
 proof
 should
 be
 once
 we
 know
 what
 values
 he
 assigns
 to
 the

differences
 in
 the
 relevant
 outcomes
 specified
 in
 (2).
 
 In
 short,
 we
 needn’t
 inquire

directly
about
someone’s
utility
assignments
if
we
know
what
values
the
two
sets
of

differences
 take
 on.
 
 One
 possibility
 would
 be
 to
 solicit
 those
 differences
 directly,

asking
 for
 instance:
 “How
 much
 would
 you
 pay
 if
 we
 could
 transform
 a
 given
 false

acquittal
 into
 a
 true
 conviction?”
 
 (This
 would
 give
 us
 the
 value
 of
 the
 numerator
 of

the
 ratio
 of
 differences.)
 A
 similar
 question
 would
 elicit
 information
 about
 the
 other

difference
and
together
the
two
answers
would
reveal
the
standard
of
proof
to
which

that
individual’s
implicit
utilities
commit
him.





Still,
there
is
an
even
cleaner
solution
to
the
problem.
We
think
one
could
get
at
the

answer
 to
 our
 conundrum
 by
 posing
 a
 single
 question,
 for
 its
 answer
 implicitly

embodies
 all
 four
 utilities.
 As
 (2)
 makes
 clear,
 the
 standard
 of
 proof
 to
 which
 one
 is

committed
 is
 specified
 once
 we
 know
 the
 value
 of
 the
 ratio
 of
 differences
 in
 the

denominator
in
(2).

It
would
seem
we
could
elicit
the
size
of
that
ratio
directly
simply

by
posing
the
following
question:


“You
 have
 just
 been
 given
 $100,000,
 tax‐free.
 
 You
 can
 keep
 it

all,
or
you
can
spend
some
or
all
of
it
in
the
following
way:


A
 felon
 has
 just
 been
 acquitted
 of
 a
 homicide
 that
 he
 actually

committed.
 
 In
 the
 next‐door
 courtroom,
 an
 innocent
 person

has
just
been
sentenced
to
prison
for
15
years
for
a
murder
he

didn't
commit.



There
is
an
infallible
appellate
Wizard
who
can

always
 identify
 the
 guilty
 and
 the
 innocent.
 By
 examining
 a

criminal
 case,
 he
 can
 with
 certainty
 determine
 whether
 the

defendant
is
guilty
or
innocent
and
he
has
the
legal
authority
to

change
the
verdict.



You
know
nothing
about
either
of
the
defendants.

You
can
use

some
 of
 your
 $100,000
 to
 pay
 the
 Wizard
 to
 convert
 the
 false

acquittal
to
a
true
conviction,
and
you
can
use
some
of
it
to
pay

the
 Wizard
 to
 convert
 the
 false
 conviction
 into
 an
 acquittal.


What
 is
 the
 maximum
 amount
 you
 would
 pay
 the
 Wizard
 to

convert
the
false
acquittal
to
a
true
conviction?

And
what
is
the

maximum
 you
 would
 pay
 the
 Wizard
 to
 convert
 the
 false

conviction
into
an
acquittal?”?



This
question
is
structured
to
take
advantage
of
the
fact
that
it
reflects
the
value
one

puts
 on
 the
 difference
 in
 utilities
 between
 false
 verdicts
 and
 true
 ones.
 The

respondent’s
answers
give
us
his
valuation
of
the
ratio
of
differences

in
(2).

From
the

mean
 of
 the
 answers
 of
 multiple
 respondents,
 we
 could
 compute
 the
 standard
 of

 19

proof,
 at
 least
 in
 the
 case
 of
 trials
 for
 homicide.28
 The
 value
 of
 the
 ratio
 will
 likely

differ
substantially
for
different
people;
until
such
time
as
this
(or
a
similar)
question

is
 posed
 to
 legislators,
 a
 value
 for
 the
 standard
 of
 proof
 of
 90%
 or
 95%
 is
 wholly

capricious.





It
 is
 vital
 to
 stress
 that
 this
 question
 does
 not
 ask
 the
 respondent
 to
 determine
 the

relative
costs
of
a
false
conviction
and
a
false
acquittal.

It
asks,
rather,
what
are
the

differences
 in
 utility
 between
 two
 false
 verdicts
 and
 their
 respective
 true

counterparts.
 
 Even
 if
 one
 believes
 that
 a
 false
 conviction
 is
 ten
 times
 worse
 than
 a

false
acquittal,
it
will
not
follow
that
one
should
be
willing
to
“pay”
ten
times
more
to

convert
a
false
conviction
into
a
true
acquittal
than
one
would
pay
to
convert
a
false

acquittal
 into
 a
 true
 conviction,
 for
 those
 “payments”
 should
 reflect
 not
 only
 the

disutilities
of
false
verdicts
but
also
the
positive
utilities
of
true
ones.



Rank­Ordering
the
Outcomes.
Even
while
(we
have
argued)
it
is
not
necessary
to
know

what
 specific
 utilities
 respondents
 would
 assign
 to
 which
 outcomes,
 it
 would
 be

interesting
 to
 know
 whether
 a
 normative
 consensus
 could
 be
 forged
 about
 how
 the

utilities
of
those
outcomes
should
be
ranked
in
terms
of
their
desirability.

With
such

knowledge,
we
would
have
a
set
of
plausible
constraints
or
restrictions
to
impose
on

respondents’
specific
assessments
of
utility.




We
 have
 already
 argued
 that
 true
 verdicts
 are
 preferable
 to
 false
 ones,
 so
 we
 can

already
say
that
specific
utilities
must
obey
these
relations:

uTC>uFC

uTC>uFA

uTA>uFC

uTA>uFA

It
 is
 also
 clear
 that
 there
 is
 a
 very
 broad
 consensus
 that
 false
 acquittals,
 while

undesirable,
are
clearly
preferable
to
false
convictions.29
We
share
that
sense,
since
a

false
conviction,
especially
for
a
serious
crime,
visits
enormous
undeserved
harm
on

an
 innocent
 defendant.
 
 More
 than
 that,
 a
 false
 conviction
 fails
 to
 bring
 to
 book
 the

true
perpetrator
of
the
crime
and
thereby
makes
it
highly
unlikely
that
he
will
never

be
 punished.
 While
 false
 acquittals
 clearly
 have
 their
 drawbacks
 
 ‐‐not
 the
 least
 of

which
is
the
release
of
a
guilty
person
back
onto
the
streets—they
are
less
undesirable

than
false
convictions.



Indeed,
the
only
point
about
which
there
appears
to
exist
controversy
has
to
do
with

the
relative
utilities
of
true
convictions
and
true
acquittals.
We
believe
that
the
correct

relation
 here
 is
 .
 
 Still,
 as
 Figure
 3
 shows,
 there
 is
 controversy
 about
 this
 in

the
 literature,
 and
 various
 scholars
 would
 reverse
 the
 inequality
 sign
 here,
 insisting



























































28
Although we have, for ease of exposition, written in this paper as if there were a single standard of proof
that would apply alike to all crimes, we deliberately leave that question open to empirical research. It may
turn out that the acceptable ratios of differences for (say) shoplifting, drug possession, aggravated assault and
homicide are quite different. For further discussion, see below.
29
Judging by the attributes shown in Figure 4, false acquittals have precious few redeeming features.
 20

that
the
utility
of
a
true
acquittal
is
greater
than
that
of
a
true
conviction.
Tribe
argues,

as
 we
 will,
 in
 favor
 of
 ,30
 whereas
 Milanich
 argues
 for
 its
 reverse, ,

asserting
 that
 “acquitting
 the
 innocent
 is
 the
 expressed
 highest
 value
 of
 our
 legal

system.”31




a).
The
subtleties
involved
here
can
be
brought
into
sharper
focus
by
considering
the

following
two
very
different
questions
aimed
at
eliciting
relative
preference
between

the
 two
 outcomes.
 
 The
 first
 is,
 “If
 there
 were
 two
 different
 trials
 underway,
 one
 of

which
 involves
 a
 defendant
 who
 is
 actually
 guilty
 and
 the
 other
 a
 defendant
 who
 is

actually
 innocent,
 would
 I
 rather
 see
 the
 guilty
 defendant
 convicted
 or
 the
 innocent

defendant
 acquitted?”
 
 For
 most
 respondents,
 the
 answer
 to
 this
 question
 will

probably
reflect
the
sentiment
expressed
by
Milanich.

That
is,
were
one
able
to
assure

the
 outcome
 of
 only
 one
 of
 these
 trials,
 most
 would
 probably
 choose
 to
 have
 the

innocent
defendant
acquitted.

This
suggests
a
utility
ranking .


 But
 now
 consider
 the
 second
 question,
 “Would
 I
 rather
 have
 society
 bring
 to

trial
 an
 innocent
 person
 and
 then
 acquit
 that
 person
 or
 have
 society
 bring
 to
 trial
 a

guilty
 person
 and
 then
 convict
 that
 person?”
 
 It
 can
 be
 seen
 that
 the
 answer
 to
 this

question
will
almost
certainly
be
opposite
the
answer
to
the
first
one.

For
most,
the

clear
winner
would
be
to
bring
to
trial
a
guilty
person
and
then
convict
that
person.


This
is
consistent
with
the
restriction
 
.


 While
 this
 highlights
 the
 tension,
 neither
 of
 these
 questions
 actually
 gives
 us

sufficient
 clarity.
 
 The
 problem
 with
 the
 first
 question
 is
 that
 it
 surreptitiously

introduces
as
pertinent
considerations
that
are,
in
fact,
ancillary
to
it
and
thus
need
to

be
distinguished
from
it.

To
illustrate,
consider
two
extreme
cases:

First,
suppose
it

were
 the
 case
 that
 we
 knew
 all
 juries
 performed
 perfectly
 and
 without
 error
 in

acquitting
innocent
defendants
and
imperfectly
in
convicting
guilty
defendants.

Then

there
 would
 be
 no
 value
 to
 causing
 the
 innocent
 defendant
 to
 be
 acquitted
 and
 we

would
 undoubtedly
 prefer
 to
 assure
 that
 the
 guilty
 defendant
 is
 convicted.
 
 This

suggests
a
utility
ranking .
Secondly,
suppose
juries
were
perfect
in
convicting

guilty
 defendants
 and
 imperfect
 in
 acquitting
 innocent
 defendants.
 
 In
 this
 case
 we

would
 choose
 to
 have
 the
 innocent
 defendant
 acquitted,
 indicating
 that
 my
 utility

ranking
is .

The
core
problem
here
is
that,
cast
this
way,
the
utilities
depend

on
 prior
 beliefs
 about
 the
 performance
 of
 juries.
 
 This
 is
 arguably
 unknowable,
 and

could
 even
 change
 with
 time
 and
 locale.
 
 It
 would
 be
 far
 preferable
 to
 have
 a
 utility

ranking
that
avoids
this
complication
and
somehow
reflects
society’s
pure
preferences

on
the
outcomes
themselves,
independent
of
considerations
as
to
their
likelihood.





 The
 problem
 with
 the
 second
 question
 is
 even
 subtler.
 
 While
 it
 avoids
 the

complicating
considerations
of
the
first
question,
it
embeds
in
it
a
certain
“sunk
cost”

consideration
with
respect
to
the
task
juries
are
asked
to
perform.

That
is,
by
the
time

society
 calls
 upon
 a
 jury
 to
 make
 a
 decision
 to
 convict
 or
 acquit
 a
 defendant,
 that

defendant
has
already
suffered
the
cost
of
being
charged
and
brought
to
trial.

In
the


























































30
Tribe, 84 Harv. L. Rev. at 1379-80.
31
Patricia G. Milanich, “Decision Theory and Standards of Proof,” 5 LAW & HUM. BEHAV. 87, at 93 (1981).
 21

case
of
an
innocent
defendant,
that
cost
can
be
severe
even
given
acquittal
–
given
the

BARD
standard,
an
acquittal
tells
the
public
only
that
there
was
insufficient
evidence

to
 convict,
 not
 that
 this
 person
 has
 been
 wrongly
 brought
 to
 trial,
 so
 acquittal
 can

carry
 with
 it
 a
 serious
 stigma,
 not
 to
 mention
 the
 expenses
 required
 to
 mount
 a

defense.
 
 And
 here
 is
 the
 problem
 with
 this:
 in
 Decision
 Theory,
 correct
 decision

making
 requires
 the
 decision
 maker
 to
 consider
 only
 those
 things
 that
 the
 decision

can
affect
going
forward,
and
to
ignore
those
events
that
have
already
happened,
and

in
particular
any
costs
or
benefits
associated
with
those
events.




 Avoiding
the
“sunk
cost
fallacy”
can
be
most
easily
illustrated
by
reference
to
a

business
 example.
 
 
 In
 this
 example,
 a
 business
 executive
 has
 made
 a
 $500
 million

acquisition,
 which
 at
 the
 time
 looked
 to
 be
 a
 prudent
 way
 to
 increase
 earnings

substantially.

A
few
years
down
the
road,
things
have
not
gone
as
well
as
expected.


An
analysis
shows
the
net
present
value
of
expected
future
cash
flows
from
this
entity

to
now
be
$200
million.

The
executive
is
approached
by
someone
who
offers
to
buy

the
 entity
 for
 $300
 million.
 
 Psychologically,
 the
 executive
 is
 heavily
 disinclined
 to

accept
the
offer
–
she
knows
she
will
be
required
to
take
a
write‐down
(i.e.,
publicly

admit
to
a
bad
outcome
for
a
major
decision
she
made).

She
is
inclined
to
say,
“I
don’t

want
to
take
a
penny
less
than
I
paid
for
it.”

But
this
latter
is
not
among
the
options

available
 to
 her.
 
 Her
 decision
 is
 a
 choice
 between
 two
 alternatives:
 one,
 continue

operating
 the
 entity
 at
 a
 present
 value
 of
 $200
 million
 to
 her
 shareholders;
 or
 two,

accept
a
$300
million
payment
now
for
her
shareholders.

The
clear
choice
is
to
accept

the
offer,
the
goal
being
to
maximize
value
given
the
options
available
going
forward.


The
$500
million
initial
outlay
is
irrelevant
to
this
decision.

And
it
would
be
irrelevant

whether
the
$500
million
were
$700
million
or
$10
billion.


 The
jury
decision
has
similar
characteristics.

By
the
time
a
jury
is
empanelled,

several
important
decisions
have
already
been
made,
including
decisions
on
whether

to
remand
into
custody
or
release
with
bail,
whether
to
bind
over
for
trial
or
dismiss,

whether
to
charge
the
individual
or
not,
and
whether
to
arrest
the
individual
or
not.


Each
is
a
separate
decision
carrying
its
own
outcomes
and
its
own
considerations
of

the
societal
utilities
and
disutilities
of
these
outcomes,
but
the
important
point
is
that

for
 the
 jury
 decision
 itself
 society
 should
 not
 allow
 jurors
 to
 consider
 any
 of
 these

sunk
costs
if
the
objective
is
to
maximize
social
utility.

Instead,
the
outcome
utilities

jurors
 should
 use
 (indirectly,
 as
 argued
 here,
 by
 adhering
 to
 a
 societally‐specified

standard
of
proof),
should
ignore
all
the
sunk
costs
that
the
defendant
has
brought
to

the
courtroom,
including
jail
time,
bond
cost,
attorneys’
fees,
and
the
stigma
of
being

brought
to
trial.

This
is
in
conflict
with
the
second
question,
which
is
cast
in
terms
of

being
 brought
 to
 trial
 to
 begin
 with,
 “Would
 I
 rather
 have
 society
 bring
 to
 trial
 an

innocent
 person
 and
 then
 acquit
 that
 person
 or
 have
 society
 bring
 to
 trial
 a
 guilty

person
 and
 then
 convict
 that
 person?”
 
 That
 question
 unfortunately
 brings
 into

consideration
what
to
a
jury
will
be
the
sunk
cost
of
bringing
the
person
to
trial,
and

so
cannot
form
a
basis
for
assessing
the
correct
societal
utilities.



b).
Hence,
if
such
questions
don’t
carry
us
to
the
nub
of
the
issue,
we
need
to
consider

a
 different
 strategy
 for
 resolving
 the
 question
 of
 the
 relative
 ranking
 of
 true

convictions
 and
 true
 acquittals.
 
 We
 conjecture
 that
 those
 minded
 to
 rank
 true

acquittals
 as
 more
 valuable
 than
 true
 convictions
 do
 so
 because
 they
 are,
 in
 effect,

 22

double‐counting
 the
 costs
 and
 benefits,
 thereby
 exaggerating
 the
 gaps
 between
 the

utilities
that
we
are
trying
to
estimate.


If
we
look
at
Table
3,
true
acquittals
appear
to

be
much
less
valuable
than
true
convictions.

They
don’t
bring
closure
to
victims;
they

don’t
visit
his
just
deserts
on
the
real
felon;
they
don’t
deter
crime
or
incapacitate
the

guilty.
 
 They
 don’t
 even
 provide
 a
 robust
 exoneration
 to
 the
 acquitted
 defendant.


While
truth
and
justice
have
triumphed
(and
that
is
never
to
be
underestimated),
that

virtue
does
not
mark
off
a
true
acquittal
from
a
true
conviction.





Why,
 then,
 might
 anyone
 suppose
 that
 a
 true
 acquittal
 is
 more
 valuable
 than
 a
 true

conviction,
which
exhibits
multiple,
impressive
virtues?

The
most
plausible
answer
to

this
 question
 goes
 as
 follows:
 
 when
 someone
 is
 asked
 to
 judge
 how
 valuable
 a
 true

acquittal
 is,
 he
 may
 well
 say
 to
 himself:
 “Well,
 if
 a
 truly
 innocent
 person
 is
 not

acquitted,
 the
 only
 alternative
 is
 that
 he
 is
 falsely
 convicted.
 
 Since
 that
 outcome
 is

horrendous,
a
true
acquittal
must
be
a
very
good
thing
as
it
prevents
something
very

bad.”

But
this
inference
is
precisely
an
instance
of
double
counting.

A
false
conviction

is
indeed
very
costly;
it
sends
to
jail
an
innocent
person,
depriving
him
of
his
liberty,

his
reputation,
and
any
prospect,
at
least
in
the
short
and
medium
term,
for
a
normal

life.
That
is
why
almost
anyone
would
agree
that
the
disutility
of
a
false
conviction
is

very
 high.
 
 Still,
 since
 we
 have
 already
 packed
 all
 those
 nasty
 features
 into
 our

appraisal
of
the
disutilities
of
a
false
conviction,
we
cannot
consistently
backload
our

fears
about
that
outcome
into
our
appraisal
of
other
trial
outcomes.

It
makes
no
sense

whatever
 to
 say
 that
 a
 true
 acquittal
 is
 highly
 desirable
 chiefly
 because
 its

counterpart,
a
false
conviction,
is
so
otiose.





Consider
 a
 homely
 example:
 
 suppose
 one
 is
 contemplating
 a
 skiing
 holiday.
 
 The

alternative
is
to
stay
home.

It
might
be
pertinent,
in
drawing
up
the
pros
and
cons
of

this
 decision,
 to
 factor
 in
 the
 possibility
 that
 one
 might
 break
 one’s
 neck
 on
 the
 ski

slopes.

One
could
either
count
that
as
a
reason
not
to
go
skiing
or
as
a
reason
to
stay

home.
 
 But
 if
 we
 count
 it
 both
 as
 a
 reason
 not
 to
 go
 skiing
 and
 as
 a
 reason
 to
 stay

home,
 we
 are
 distorting
 the
 real
 nature
 of
 the
 choice
 between
 the
 alternatives.
 
 We

make
the
gap
between
the
respective
utilities
of
the
two
options
much
wider
than
it

should
be.
Returning
to
our
example
of
the
valuation
of
a
true
acquittal,
one
likewise

has
two
choices:
one
could
favor
a
true
acquittal
as
the
best
possible
outcome
on
the

grounds
 that
 it
 would
 avoid
 all
 the
 nasty
 consequences
 of
 a
 false
 conviction
 or
 one

could
leave
all
the
nasty
consequences
in
the
debit
column
of
a
false
conviction.

But

one
cannot
do
both.

That
is,
we
can
either
give
a
very
high
utility
to
a
true
acquittal

(thus
robbing
a
false
conviction
of
all
those
horrors
that
give
it
such
a
high
disutility)

or
we
can
give
a
high
disutility
to
false
convictions
and
judge
true
acquittals
by
what

they
 are
 rather
 than
 what
 they
 are
 not.
 
 Interestingly,
 many
 scholars
 who
 have

addressed
these
issues
have
apparently
sought
to
do
both.
They
have
held
both
that
a

false
 conviction
 is
 the
 most
 otiose
 of
 outcomes
 and
 that
 a
 true
 acquittal
 is
 the
 most

desirable
 of
 outcomes,
 chiefly
 –we
 conjecture‐‐because
 the
 latter
 precludes
 a
 false

conviction.
Our
reply
is
that
if
the
principal
virtue
of
a
true
acquittal
were
simply
that

it
avoids
a
false
conviction,
then
one
can
either
rank
true
acquittals
at
the
top
of
the

scale
 of
 desirable
 outcomes
 (thereby
 precluding
 oneself
 from
 being
 able
 to
 put
 false

convictions
in
the
cellar
of
the
ratings)
or
one
can
accept
that
true
acquittals
are
fairly

 23

inert,
low
utility/low
disutility
verdicts
and
that
false
convictions
are
the
outcomes
to

be
most
avoided.




For
 such
 reasons,
 we
 believe
 that
 the
 rational
 rank
 ordering
 of
 the
 outcomes
 is
 as

follows:

uTC>uTA>uFA>uFC.




An
 important
 aside:
 we
 have
 so
 far
 been
 somewhat
 cavalier
 in
 our
 treatment
 of
 the

required
 utilities
 (or
 more
 precisely,
 the
 utility
 differences)
 in
 regard
 to
 the
 exact

nature
 of
 the
 crime
 involved,
 beyond
 characterizing
 it
 as
 “serious.”
 
 
 The
 objection

might
legitimately
be
raised
that
assessed
utility
differences
may
depend
on
the
exact

crime
under
consideration
and
will
likely
differ
as
between
a
homicide
and
an
armed

robbery.
 
 Moreover,
 if
 the
 methodology
 is
 purported
 to
 be
 applicable
 to
 all
 criminal

cases,
it
is
not
difficult
to
put
forward
a
conjecture
that,
for
many
respondents,
their

expressed
 utility
 differences
 will
 differ
 as
 between
 a
 homicide
 and
 shoplifting,
 thus

calling
into
question
the
methodology.



But
in
fact,
this
is
a
strength
of
the
methodology,
not
a
weakness.

If
the
objective
is
to

provide
 more
 rational
 standards
 of
 proof
 to
 jurors,
 the
 methodology
 can
 readily

accommodate
different
standards
of
proof
for
different
classes
of
crime.

By
specifying

the
crime
(and
associated
punishment)
in
advance
of
assessing
the
utility
differences,

we
open
the
door
to
a
natural
association
of
classes
of
crimes
with
different
standards

of
proof.

It
is
easy
to
speculate
that
society
may
prefer
to
require
a
different
standard

of
 proof
 for
 a
 capital
 crime
 than
 for
 shoplifting.
 
 While
 it
 cannot
 be
 so
 claimed
 in

advance,
 actual
 assessments
 of
 utility
 differences
 from
 legislators
 may
 point
 to
 the

need
 for
 different
 standards
 of
 proof
 for
 different
 classes
 of
 crime.
 
 This
 would
 be
 a

natural
outcome
of
the
methodology,
not
a
limitation
of
it.

Some
Formal
Rationality
Restrictions
on
Societal
Utilities:



We
have
presented
the
case
that
utilities
themselves
are
not
necessary
to
establishing

a
 standard
 of
 proof
 for
 trial
 juries,
 but
 rather
 a
 more
 readily‐assessed
 difference
 in

utilities.

However,
if
the
problem
is
viewed
more
generally
as
establishing
standards

of
proof
for
the
justice
system,
including
decisions
regarding
arrest
or
don’t
arrest,
file

charges
or
don’t,
bind
over
for
trial
or
dismiss,
and
remand
into
custody
or
release,
it

will
 likely
 be
 the
 case
 that
 outcome
 utilities
 will
 need
 to
 be
 assessed
 rather
 than

simple
differences.

Accordingly,
it
behooves
us
to
comment
on
plausible
restrictions

that
should
be
placed
on
these
outcome
utilities.



While
 citizens
 and
 legislators
 are
 free
 to
 assign
 whatever
 subjective
 utilities
 to
 the

four
 outcomes
 they
 find
 appropriate,
 that
 activity
 should
 nonetheless
 be
 subject
 to

some
 formal
 constraints.
 For
 instance,
 we
 would
 want
 to
 exclude
 any
 utility

assignments
that
allowed
for
the
value
of
the
standard
of
proof
to
take
on
a
negative

value
 or
 that
 allowed
 for
 probabilities
 greater
 than
 1.0.
 
 
 Likewise,
 we
 would
 find
 it

bizarre
if
someone
assigned
higher
utilities
to
false
verdicts
than
they
did
to
true
ones.



 24

We
 have
 already
 claimed
 that
 no
 one
 who
 would
 dispute
 that
 true
 verdicts
 are

preferable
 to
 false
 ones.
 
 So,
 we
 can
 be
 sure
 that
 ,
 ,
 ,

,
and
that
 .




These
 four
 inequalities
 assure
 that
 a
 standard
 of
 proof
 will
 always
 exist,
 and
 that
 it

will
 be
 between
 0
 and
 1.32
 They
 also
 insure
 that
 the
 societal
 appeal
 of
 each
 verdict

changes
in
a
sensible
way
with
respect
to
the
juror’s
perceived
probability
of
guilt, .


That
 is,
 these
 restrictions
 imply
 that
 the
 societal
 expected
 utility
 of
 a
 guilty
 verdict

increases
 as
 the
 juror’s
 perceived
 probability
 of
 guilt
 goes
 up,
 and
 that
 the
 societal

expected
 utility
 of
 a
 not
 guilty
 verdict
 decreases
 as
 juror’s
 perceived
 probability
 of

guilt
goes
up.

More
formally,
with
these
two
restrictions,

and .

This
 result
 relies
 on
 the
 fact
 that
 the
 two
 expected
 utility
 functions
 are
 linear
 and

strictly
monotonic.

However,
these
restrictions
do
not
alter
the
feasible
range
of .



This
can

be
seen
by
considering
Figure
5,
below.33































































32
Existence
Theorem


If
both
 and ,
there
exists .





A
 formal
 proof
 is
 not
 required
 as
 the
 result
 is
 evident:
 by
 looking
 at
 Figure
 2
 it
 is

apparent
that
if has
strictly
larger
value
than (consider
the
left‐hand
intercepts

at )
 and has
 strictly
 larger
 value
 than (consider
 the
 right‐hand
 intercepts

at ),
 the
 two
 lines
 must
 cross
 somewhere
 between and .
 
 This
 would

hold
 even
 if
 the
 two
 lines
 were
 highly
 non‐linear,
 provided
 only
 that
 the
 functions

determining
 these
 lines
 were
 continuous.
 
 Here,
 the
 expected
 utility
 lines
 are
 both

linear
and
continuous.




Note
 that
 given
 these
 two
 restrictions,
 and ,
 the
 value
 of is
 not

restricted
at
all,
except
to
be
bounded
by
 .

Note
also
that
these
conditions

are
a
sufficient
(but
not
necessary)
condition
for
the
existence
of
p*.



33
This figure is essentially identical to Figure 3 of DeKay, op. cit.
 25





Figure
5.
Relationship
between
Utility
Differences
and
the
Standard
of
Proof



Figure
5
shows
how
the
standard
of
proof
depends
on
the
ratio
of
 to
 .


If
 one
 visualizes
 shrinking
 the
 distance
 (indicated
 as
 the
 right‐hand
 brace)

relative
to
 (indicated
as
the
left‐hand
brace),
one
can
see
that
this
increases

the
threshold
standard
of
proof
 ;
similarly,
shrinking
the
left‐hand
brace
relative

to
 the
 right‐hand
 brace
 decreases
 the
 standard
 of
 proof
 .
 
 It
 can
 be
 seen
 that

appropriate
reductions
in
the
two
utility
differences
can
lead
to
a
very
large
range
for

the
 value
 of
 .
 
 Thus,
 as
 we
 have
 argued
 previously,
 there
 is
 no
 foundation
 for

arguing
 that
 the
 standard
 of
 proof
 must
 necessarily
 be
 anything
 like
 the
 value
 of

around
90%
commonly
proposed
by
other
researchers.



A
 cautionary
 theoretical
 note:
 The
 core
 of
 what
 we
 are
 trying
 to
 accomplish
 is

bringing
rationality
and
clarity
to
the
standard
of
proof
in
criminal
law
using
decision

and
 utility
 theory.
 It
 is
 argued
 elsewhere
 that
 this
 is
 foundational
 to
 securing
 equal

protection
 and
 due
 process
 and
 that
 it
 can
 be
 practically
 accomplished.34
 
 
 But

achieving
this
goal
requires
establishing,
directly
or
indirectly,
society’s
utilities
on
the

four
outcomes.


 However,
 there
 is
 a
 profound
 conceptual
 and
 logical
 problem
 with
 this

requirement
 having
 to
 do
 with
 the
 meaning
 of
 “societal
 utilities.”
 
 As
 famously

demonstrated
 by
 Arrow35
 in
 his
 Impossibility
 Theorem,
 there
 is
 no
 way
 to

consistently
 and
 with
 surety
 obtain
 the
 overall
 utility
 of
 any
 group
 on
 any
 set
 of

outcomes
based
on
the
utilities
of
the
individuals
in
that
group.

Practically
speaking,

society
 tries
 to
 accomplish
 this
 goal
 of
 reflecting
 societal
 utilities
 by
 means
 of

representative
 governance
 and
 various
 voting
 procedures.
 
 If
 the
 case
 is
 compelling



























































34
Harry D. Saunders, “Quantifying Reasonable Doubt: A Proposed Solution to an Equal Protection Problem,”
The Berkeley Electronic Press, Paper 881, http://law.bepress.com/expresso/eps/881 (2005).
35
Ken Arrow, “A Difficulty in the Concept of Social Welfare,” JOURNAL OF POLITICAL ECONOMY 58(4), 328-
346 (1950).
 26

enough
to
replace
the
current
standard
with
a
better,
clearer
one,
a
similar
approach

will
need
to
be
adopted
to
establish
these
societal
utilities.


 As
argued
elsewhere36
the
constitutionally‐correct
forums
for
this
in
the
United

States
 are
 the
 State
 Legislatures
 and
 the
 US
 Congress
 (where
 federal
 crimes
 are

concerned).
 
 A
 future
 study
 will
 address
 methods
 for
 facilitating
 the
 assessment
 of

societal
 utilities
 from
 State
 legislators
 and/or
 their
 designates
 and
 facilitating
 the

creation
 of
 a
 consensus
 recommendation
 leading
 to
 a
 quantified
 standard
 of
 proof

that
reflects
as
accurately
as
is
feasible
the
societal
utilities
of
that
particular
State.


 

Conclusions




This
analysis
is
based
on
the
presumption
that
standards
of
proof
can
be
quantified
in

criminal
law.

The
analysis
demonstrates
that
while
the
magnitude
of
a
utility‐based

standard
 of
 proof
 is
 highly
 dependent
 on
 societal
 utilities
 of
 the
 four
 possible

outcomes
of
a
criminal
trial,
given
all
the
rational
restrictions
on
the
ranking
of
these

societal
utilities
there
are
no
inherent
restrictions
on
the
magnitude
of
that
standard

beyond
 .
It
also
shows
that,
even
if
we
impose
on
the
utility
structure
a
set
of

unabashedly
Blackstonian
sensibilities
about
the
respective
costs
of
a
false
conviction

and
a
false
acquittal,
and
add
a
sop
to
those
who
think
true
acquittals
are
better
than

true
 convictions,
 it
 does
 not
 follow
 that
 the
 criminal
 standard
 of
 proof
 need
 be

anywhere
near
the
high
levels
associated
with
proof
beyond
a
reasonable
doubt
in
the

academic
literature.37






























































36
Harry D. Saunders, op.cit.
37
One final caution: The fact that we have spoken throughout this paper as if the standard of proof were best
expressed as a probability should not necessarily be taken to mean that we are advocating that it should be so
formulated for consumption by fact-finders. Some, including one of the authors of this paper, have voiced the
suspicion that qualitative, verbal standards of proof may be more accessible and intelligible to jurors than
quantitative ones are. If true, the height of the standard should be set by using probabilities and then
‘translated’ into a verbal expression. Others, including the other author of this paper, have argued that explicit
quantification of the standard is the only way to overcome serious equal protection and due process problems,
and to avoid a logical conundrum faced by jurors cognizant of widely disparate interpretations of verbal
statements of the standard. But either way, we offer that the analysis of this paper should help point a way
forward to a more rational, and honestly articulated, standard of proof in criminal law.