Evidential Luck and Un Fair Treatment

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

‘Evidential Luck’ and (Un)Fair Treatment

Katie Steele

August 25, 2023

Abstract
Here I draw attention to a form of luck and (un)fairness in public
decisions about the deserts or entitlements of individuals. I focus on
decisions of guilt or liability in the courtroom. The issue is that the
body of supporting evidence can differ from case to case for effectively
arbitrary reasons. While it might be accepted that ‘evidential luck’ (as
we might call it) is pervasive, here I argue that it can come in surprising
forms. In particular, I argue that the problem of ‘general’ (as opposed
to ‘individualised’ or ‘specific’) evidence in the legal setting is one that
primarily concerns evidential luck and associated (un)fairness, and can be
explained in these terms. Whether or not we should exclude or downplay
‘general’ evidence in legal and other decisions thus depends on broader
questions of how and to what extent we should mitigate evidential luck
in these decision-making contexts.

1 Introduction
In a criminal trial, it is decided whether to find a defendant guilty or not guilty
given the available evidence. When would such a decision be justified? Let’s
agree that it turns on whether there is adequate ‘admissible’ evidence in support
of the hypothesis that the defendant is in fact guilty. There remains considerable
room for debate, however, about the finer details of what this means, and how
it squares with familiar legal standards of proof like ‘beyond reasonable doubt’.
This paper contributes to that debate, but raises a new consideration that I
refer to as ‘evidential luck’: luck owing to arbitrary differences in the bodies of
evidence that ground the verdicts made about defendants.
Note that while legal decisions are the focus here and are the subject of
a considerable literature, this is just one class of public decision whereby an
individual’s deserts or entitlements are not known but can only be inferred
from the evidence. Whatever is decided for these individuals, there is a risk
of not delivering on their deserts/entitlements. Other public decisions have
similar structure, such as decisions about whether patients in a state-funded

1
hospital warrant priority treatment, decisions about whether individuals qualify
for state-funded financial loans and so on. The issues raised in this paper are
arguably relevant to all such decisions. But for reasons of both exposition and
academic caution, I will focus, for the most part, on the legal setting.
The role of ‘evidential luck’ has not been adequately explored, I claim, in
thinking about the rules of evidence in the legal (and perhaps other similar)
decision-making setting(s). For instance, imagine two defendants standing trial
for petty theft, both are guilty and with similar modes of operation, but while
one was captured on film by a hidden security camera the other was not, due to
a rare malfunctioning of the camera. The bodies of evidence here differ for what
seems an arbitrary reason with respect to the defendants’ respective guilt. If so,
then to the extent that the presence or absence of the film evidence is significant
with respect to the verdicts made about the defendants, luck is involved in their
treatment and there is associated unfairness.
More must be said about what are arbitrary reasons for bodies of evidence
differing between legal trials. The case just described suggests that these are
reasons not directly to do, in some causal sense, with the defendant’s conduct
that is under investigation. Indeed, one can already discern that evidential luck,
defined along these lines, is ubiquitous. Practically every piece of evidence in a
criminal trial will depend on one or more events, for example the workings of
a camera or the movements of a witness, that are causally independent of the
deliberations or conduct of the defendant. And yet surely, we cannot rule out or
penalise every piece of evidence in a legal trial on the grounds of luck. It might
then be thought that we need not bother ourselves at all with this unfortunate
source of unfairness in the treatment of individuals.
That would be a mistake. A closer examination of this phenomenon is
important, if only for better recognising the different guises that it may take.
In particular, I argue that evidential luck can assume a form that has not been
well appreciated in the literature on legal reasoning. That form is the key to
analysing the well-known puzzle of ‘general’ or ‘statistical’ evidence, (as opposed
to ‘specific’ or ‘individualised’ evidence). The puzzle is to explain why the former
kind of evidence seems deficient or even inappropriate (I will use the umbrella
term ‘problematic’ in what follows), whereas the latter kind does not. Contrary
to other proposals in the literature, I argue that the reason we identify ‘general’
evidence as problematic has to do with evidential luck and associated unfairness.
If this is right, then whether or not we should exclude or downplay evidence in
a legal trial on the grounds that it is ‘general’ depends on the broader question
of how we should manage evidential luck in this setting.
The paper proceeds as follows. Section 2 introduces a highly idealised case
that is commonly thought to illustrate ‘general’ evidence and how it seems
problematic: the Prisoners case. It is widely thought that the problem is an
epistemic one—the evidence in Prisoners does not provide the right kind of
epistemic support for the hypothesis that the defendant is guilty. Rather than
examining the debates internal to this line of thinking, however, I simply lay the

2
groundwork for a different diagnosis. I show why the evidence brought against
the defendant in the Prisoners case is extremely unlucky for them, and that this
raises the moral issue of (un)fairness.
I develop this observation into a response to the puzzle of ‘general’ evidence
in section 3. I argue that while evidential luck abounds in legal reasoning, what
is different about the luck associated with ‘general’ evidence is that it is, for
various reasons, unusual, and much more extreme in the form that it takes than
the luck of ‘specific’ evidence. On my proposal, what is special about ‘general’
evidence is that it gives rise to a genuinely striking form of evidential luck. I will
show, however, that what is problematic about ‘general’ evidence is not unique
to it, but is the unfairness associated with any kind of evidential luck.
In section 4 of the paper, I defend my proposal as a distinctly moral account
of what is problematic about ‘general’ evidence, one that has significant advan-
tages over rival accounts. First, in section 4.1, I discuss why mine cannot be cast
as an ‘epistemic’ proposal, even though it similarly has to do with the connection
that some evidence has with the set of hypotheses at issue. I go on in section 4.2
to defend my proposal by considering other ‘moral’ alternatives. To begin with,
I claim that there are other, perhaps more important, moral problems with evi-
dence that are not to do with its ‘generality’. My proposal concerns the special
moral problem with ‘generality’, and I claim that it is easier to defend in this
regard than alternative accounts, which rest on highly contentious claims about
what individuals are owed with respect to the evidence used to judge them.
Section 5 concludes by reflecting on what follows from viewing the problem of
‘general’ evidence as one of evidential luck and associated unfairness.

2 A classic case of ‘general’ evidence revisited


Let us focus on a classic case of ‘general’ or ‘statistical’ evidence in the literature
that is typically taken to be problematic.

Prisoners. One hundred prisoners are exercising in the prison yard.


Ninety-nine of them suddenly join in a planned attack on a prison
guard; the hundredth prisoner plays no part. There is no evidence
available to show who joined in and who did not. One prisoner
is randomly selected from the yard. There is 0.99 probability that
this prisoner participated in the attack. Thus, the prisoner is found
guilty of this crime.1

Many think that it is inappropriate to find the randomly-selected prisoner guilty,


or that there is at least something problematic about this verdict, despite the
probability of the defendant’s guilt being extremely high. (It is assumed that a
case of this sort can be constructed yielding arbitrarily high probability of guilt.
1 This example is presented in Redmayne (2008), who attributes it to Nesson (1979).

3
Presumably, for instance, the intuition the there would be something wrong
with a guilty verdict in this kind of case would remain even if the prisoner
were randomly selected from a yard in which it is known that 999 out of 10000
prisoners joined the attack.) The rough diagnosis of the intuition is that the
evidence for guilt is ‘general’, in that it pertains to a class of prisoners, rather
than ‘specific’ to the defendant at hand, and hence does not support a guilty
verdict.
The Prisoners case is highly stylised, in that there is very precise evidence
about the attack, and yet none that distinguishes between the different prison-
ers. As such, one might, perhaps subconsciously, resist the terms of the case,
thinking that there was some procedural problem, say, with the reporting and
gathering of evidence. Moreover, while the description of the case does not
preclude all prisoners being prosecuted on the same terms, it may be read as
singling out an individual, again raising procedural concerns (see Steele and
Colyvan ms. for discussion of both issues).2 Notwithstanding these concerns,
the Prisoners case is in other ways a well-controlled one for examining ‘general’
evidence in the courtroom. Unlike ethnic or racial profiling evidence, for in-
stance, the class of prisoners in the yard is not per se a salient social group that
has suffered a history of oppression.
The popular reading is that the Prisoners case is indeed a challenge, a coun-
terexample even, to the notion that the strength or quality of evidence depends
purely on its probative strength. The probative strength of evidence is the extent
to which it raises the probability of a hypothesis—in this case ‘the defendant
is guilty’—relative to its complement. It is a notion of evidence that coheres
with a broader decision-theoretic or Bayesian picture of the logic of evidence.
On that view, a guilty verdict is the rational and thus correct choice just in
case the expected value of a guilty verdict is greater than the expected value of
the alternative. This will be true just in case the probability of guilt is above
some threshold. Cases like Prisoners are commonly thought to cast doubt on
this picture. Here we have extremely probative evidence that, by design, raises
the probability of guilt above the relevant threshold. And yet the evidence is
deemed deficient or problematic in some way, such that it cannot support a
guilty verdict.
Just what further epistemic ingredient is lacking in the evidence in cases like
Prisoners has been the subject of much controversy. In this article, I will offer
a novel diagnosis. I will argue that what is going on in this case is that the
evidence plausibly has a moral deficiency to do with what I will call ‘evidential
luck’. Some initial observations are useful for introducing that proposal. Thus,
in the remainder of this section I simply try to make vivid how the evidence in
the Prisoners case turns on arbitrary factors, and thus luck.
What is arbitrary about the ‘yard’ evidence (as we might call it) in the
Prisoner’s case is that it rests on a reference class or grouping that does not
2 And in the case that all prisoners are prosecuted on the same terms, the case is unusual

in that an innocent prisoner is guaranteed to be wrongly convicted.

4
reflect, in the sense of not having a causal bearing on, the question (or set of
competing hypotheses) regarding the defendant’s guilty conduct. It is simply
that 99 out of the 100 prisoners in the yard where the defendant happened to
be exercising participated in an attack. The story states that the attack was
‘sudden’, which is to say that there was no collusion amongst the prisoners.
We are to assume that neither the constitution of the group exercising nor the
exercise yard itself had any bearing on whether a given prisoner participated
in the attack. As such, the defendant may rightly be said to be judged on the
basis of other people’s causally independent behaviour, a matter of luck.
To see this more clearly, it helps to consider a modified version of the Pris-
oners case I will be assuming that the standard of proof):

Comparative Prisoners. Two hundred prisoners are exercising in


two yards (A and B), one hundred prisoners per yard. In each yard,
a group of prisoners spontaneously attacks the prison guard for that
yard. In yard A, ninety-nine prisoners participate in the attack,
while in yard B, only fifty prisoners participate in the attack. There
is no evidence available to show who joined in either attack and who
did not. And since the prisoners were randomly sorted into yards,
there is nothing about the yard in which a prisoner was located
that could be said to influence their decision to participate in an
attack. There is 0.99 probability that a prisoner randomly selected
from yard A participated in the attack and so they are found guilty
of this crime. By contrast, there is 0.50 probability that a prisoner
randomly selected from yard B participated in the attack and so
they are not found guilty of this crime.

Comparative Prisoners draws attention to the contingency of the yard evidence.


We can think of it as a variable that takes some value for a defendant. If
the defendant is from yard A, the variable takes the value of 99 guilty out of
100, while for those in yard B, it takes the value of 50 guilty out of 100. It is
explicitly stated that the prisoners were sorted randomly into the two yards. So
it is clearly arbitrary—a matter of luck—whether any given prisoner had very
many or very few guilty colleagues in their yard and was judged accordingly.
Let us focus on the predicament of the innocent individuals in yard A vis-
à-vis the innocents in yard B. For the innocents who happen to be in yard A,
the probability of false conviction is one. They (or rather the one person in
question) will all be convicted. By contrast, the innocents in yard B have zero
probability of false conviction. None will be convicted. So the false conviction
rate differs markedly for the groups that are delineated on the basis of the yard
they are in. But recall that this grouping—the yard that an individual happens
to be in—does not reflect any residing feature of the individual, nor any fact that
bears on the means, motive or opportunity for their participation in an attack.
The grouping was arbitrary in this way relative to the individual’s guilt. Indeed,

5
in this case it was decided randomly, let’s say by a coin toss. Hence the role of
luck in how the defendant is judged, something that strikes us as unfair.
Note that this analysis carries over to the Prisoners case, even though it
is not made explicit that the yard evidence could have taken a different value.
There is nonetheless a sense in which any given innocent, and indeed any given
prisoner, could have easily been drawn from a yard with a different number of
guilty colleagues. For instance, it could have been the case that fewer of the
defendant’s colleagues decided to participate in the attack, whether or not they
did. For any given prisoner, it is effectively random what are the actions of
others in the yard. And so here too the value of the yard evidence is arbitrary,
causally speaking, with respect to whether a given defendant is guilty, and
so in this sense it is a matter of luck whether the evidence is favourable or
unfavourable for the defendant.

3 But why is ‘general’ evidence noteworthy?


One might agree that there is something unlucky about the evidence in the
Prisoners case, such that its use introduces unfairness. But this does not seem
to help us understand what is especially problematic about ‘general’ evidence.
After all, as noted earlier, there is plausibly luck and associated unfairness in-
volved in any piece of evidence bearing on an individual’s guilt or liability. That
is because whether or not a piece of evidence obtains nearly always depends in
some measure on events that are effectively independent of the deliberations and
conduct of the individual in question. So any proposal that ‘general’ evidence
appears problematic since it is (un)lucky and hence unfair requires some further
qualification.
I will contend that we call some evidence ‘general’ because it gives rise to a
peculiar kind of evidential luck, such that the associated unfairness is striking
or noteworthy. I stop short of claiming that ‘general’ evidence, understood in
this way, is so problematic as to be unacceptable—that this kind of evidence
is, for reasons of (un)fairness, unable to support a legal verdict. I consider it
rather a positive upshot of my proposal that it leaves open what, if anything,
should be done about the luckiness of ‘general’ evidence. While a peculiar case
of evidential luck, it represents just one form of this general phenomenon, and
so any measure to curb the luckiness of ‘general’ evidence should be devised
with this broader perspective in mind.
To begin with, what sets ‘general’ evidence apart is that it involves a less
obvious form of what I have been calling evidential luck. We might say that it
concerns the value that a piece of evidence takes rather than the presence or
absence of the piece of evidence in the first place.3
3 This distinction is not intended to reflect some deep division; it is rather intended to

capture a difference in how we tend to think of and model bodies of evidence.

6
Return to the example given in Section 1: While one guilty defendant hap-
pened to be captured on film by a security camera, another guilty defendant
was not captured on film, simply because the security camera happened not to
be working properly. To give another example: While one guilty defendant hap-
pened to be witnessed by a person of excellent vision, another guilty defendant
was rather witnessed by a person who happened to have poor vision and was
thus unable to identify them. While one innocent defendant had an excellent
alibi, another innocent defendant was alone at the time of the crime. And so on.
In all these cases, we tend to think of the evidence in question as simply present
or absent, and the comparisons between cases highlight how this is partly the
result of arbitrary factors.
Prisoners, however, is different from these cases. Here, the arbitrariness
apparently enters at a different point—in determining the value the evidence
takes rather than its presence/absence—and so may not be recognized as the
same kind of phenomenon. We accept that the defendant was in a prison yard
and that within that yard there was some number of guilty prisoners. But
just what value this guilty proportion takes in the story depends on arbitrary
factors, causally speaking, with respect to the question of the defendant’s own
guilt—the nature of the other prisoners with whom the defendant is grouped.
The evidence in Prisoners introduces luck because it has significant probative
strength. That is the second crucial part of the peculiarity of ‘general’ evidence.
Roughly speaking, the arbitariness of the evidence does not correlate with its
probative strength. After all, one might be thinking that most evidential in-
ference relies on statistical generalisation, or the average features of a reference
class associated with the object or event in question. So what is inferred about
any object or event typically depends on the nature of other objects or events
that are grouped with it. In most cases, however, the probative strength of
evidence that turns on a reference class in this way is only probative to the
extent that the groupings are not arbitrary but are rather constructed on the
basis of some common property that has a causal bearing on the question at
hand, or set of hypotheses under consideration. example? 4 The evidence in
Prisoners is peculiar in that it is highly probative despite pertaining to what
seems an arbitrary grouping. That’s why it stands out as a separate category
of ‘general’ evidence that is problematic. (But note that on this account, ‘gen-
erality’ comes in degrees. There is no simple distinction between ‘general’ and
‘specific/individualised’ evidence; rather some evidence like that in Prisoners
is at the extreme luck or ‘general’ end of the spectrum, while other forms of
evidence are at the lesser luck or ‘individualised’ end of the spectrum.)
Let me elaborate on why the value the evidence takes in Prisoners can be
seen as arbitrary in the extreme and yet nonetheless highly probative. I will
4I use the somewhat ambiguous term ‘causal bearing’ intentionally to indicate that the
finer details are yet to be filled in. It stands in for an inclusive notion of causal relevance
rather than an actual causal connection. So for instance, the presence of an ashtray in a
person’s house has a causal bearing on whether they will develop lung cancer, even though
the two are merely related by a common cause (that the person smokes).

7
focus on evidence that appeals to a person’s characteristics. In line with my
comments above, it helps to see such evidence as effectively placing the individ-
ual in one of a set of mutually exclusive and exhaustive groups of people and/or
actions. The value the evidence takes tells us the specific group in which the
individual is placed; we should then see the individual as being picked ran-
domly from that group. If the group in question differs in relevant ways from
other competing groups that the individual might have been placed in, then
the evidence is probative. But we can further ask whether the group that the
individual is placed in is suitably representative of the individual (with respect
to the hypothesis about the individual in question).
Now, how representative a grouping is for an individual is typically impor-
tant in determining its probative strength, which, recall, is the extent to which
the evidence raises the probability of the target hypothesis concerning the in-
dividual, or more generally, the extent to which the evidence discriminates,
in probabilistic terms, between the competing hypotheses in question. For in-
stance, the group of male smokers that are similarly aged is more representative
of Smith’s probability of getting cancer in the next 5 years than is the group
of male smokers of all ages. The former group involves more of Smith’s char-
acteristics that have a causal bearing on the question of Smith getting cancer.
Hence there will be a greater difference in the cancer rate amongst the former
(age-stratified and smoking stratified) groups than there is amongst the latter
(merely smoking stratified) groups. The evidence of age and smoking status is
more probative than the evidence of smoking status alone.
In most cases, the groupings in question are hypothetical and are constructed
on the basis of (typically, causal) laws. Take the hypothesis that Smith will get
cancer in the next 5 years, and assume only the weaker evidence is available:
Smith is a smoker rather than a non-smoker. How probative is this evidence?
Think of it this way: We imagine a hypothetical group of smokers (in contrast to
non-smokers) at some given time, and we consider—by appeal (at least roughly)
to causal laws—what proportion of each of these groups are people who went on
to get cancer in the subsequent 5 years. (These proportions may be estimated by
appeal to real groups.) Here the evidence is probative—being a smoker has some
law-based connection to one’s prospects of getting cancer in the next 5 years.
The group of smokers is somewhat representative of Smith since Smith smokes.
But it is not strongly representative of Smith because the group of smokers is
still rather diverse or heterogeneous; many other causal factors besides being
a smoker play a role in the prospects for cancer in the short term, and so the
smokers in the group will vary greatly in causally relevant ways. When we use
the fact that Smith is a smoker in estimating her probability of cancer in the
next 5 years, we are effectively picking her randomly from a very diverse group.
The probability that we assign to the random pick, or Smith, getting cancer
in the next 5 years thus depends on the characteristics of a very diverse group
of smokers, some very unlike Smith. Other groupings, such as those based on
more causally relevant information—say, smokers of similar age to Smith—are
more representative of Smith in that they are less causally heterogeneous.

8
When a grouping is not representative of an individual at all with respect
to properties that have a causal bearing on the question about them that is of
interest, then the grouping or evidence is typically not probative. For instance,
consider the evidence that Smith’s favourite colour is blue. Again, we can treat
this evidence as assigning Smith to the hypothetical group of people whose
favourite colour is blue, as opposed to other groups for people with different
favourite colours. There is no law-based connection between one’s favourite
colour and one’s prospects for cancer. So each of the groups corresponding to
different favourite colours is identical and also highly diverse with respect to the
other causal factors that influence prospects for cancer.
That brings us to the peculiarity of the evidence in the Prisoners case. The
yard evidence upsets the usual relationship between the representativeness of
an evidential grouping and the probative power of that grouping with respect
to the set of competing hypotheses under consideration. After all, the yard that
a prisoner is exercising in, and in particular, who else is in that yard, has no
(as we assumed) law-based causal connection to any prisoner’s decision about
whether to join an attack. So the yard evidence or grouping is not causally
representative of whether the defendant joined the attack. And yet the yard
evidence is highly probative. The reason for this is that the grouping is not a
hypothetical one whose properties are derived in accordance with (causal) laws.
The grouping is rather an actual one—an effectively ad hoc population—about
which we happen to know the proportion of guilt and from which we know the
individual in question was randomly selected.5
So the evidence in Prisoners is particularly stark because the causal repre-
sentativeness of the grouping that the evidence imposes—the grouping by which
the defendant is judged—comes apart dramatically from its probative value. In-
deed, the evidence here has nothing to do with causally relevant features of the
defendant with respect to the hypothesis about their guilt at all—it is in this
sense arbitrary—and yet it is extremely probative. No wonder we find such
evidence perplexing.

4 A plausible and distinctly moral proposal


My proposal, laid out above, gives new significance to the phenomenon of ev-
idential luck. I claim that this phenomenon comes in different guises, some of
which have been overlooked; in particular, evidential luck and associated unfair-
ness is, surprisingly, the key to the puzzle of ‘general’ evidence. This is to say
that ‘general’ evidence is problematic for moral reasons. But one might remain
doubtful that I offer a distinctly moral diagnosis of the problem of ‘general’
5 C.f., Godfrey-Smith’s (2011) discussion of two kinds of rational inductive inference from

samples, one involving inference to the population from which the sample was randomly
selected, and the other involving projection, by appeal to causal laws, to a wider population
than was sampled.

9
evidence, and moreover a plausible moral diagnosis of that problem. In this
section, I address these doubts.

4.1 Epistemic alternatives


My evidential-luck proposal focuses on the way in which the value an evidence
variable takes together with its probative value with respect to some set of
legal hypotheses may involve much luck; the use of this evidence in a legal trial
therefore introduces a kind of unfairness. But if what is at issue is an arbitrary
relationship between evidence and hypotheses, then one might think that my
proposal could just as well be cast in terms of an epistemic worry rather than
a moral one of (un)fairness. While I claim that ‘general’ evidence is unlucky
for the defendant, albeit good evidence for their guilt, one might instead claim
that ‘general’ evidence is simply not good evidence for the defendant’s guilt
due to the luck involved. Indeed, insofar as my (un)fairness proposal rests on a
notion of arbitrary link between evidence and hypothesis, there is a counterpart
epistemic proposal that appeals to the same problematic connection between
evidence and hypothesis.6
By way of further bolstering this line of argument, it might be noted that
Prisoners is a rather unusual case, in that it involves random sampling from
an ad hoc population with known characteristics.7 Other paradigm cases of
‘general’ evidence in the literature do not seem to have this form, but rather
involve a paucity of evidence, and in these cases, the problem thus seems even
more an epistemic one as opposed to a moral one of (un)fairness. Consider, for
instance:

Blue Bus. Mrs. Brown is run down by a bus on Victoria Street; 60


percent of the buses that travel along this street are owned by the
blue bus company, and 40 percent by the red bus company. The only
witness is Mrs. Brown, who is colour-blind. Given the lack of further
information, there is 0.6 probability that Mrs. Brown was run down
by a blue bus. Hence the blue bus company is held liable.8

As in Prisoners, many have the intuition that the bus-traffic evidence is insuffi-
cient for holding the blue bus company responsible, despite the evidence being
6 The proposal could be seen as an elaboration of an epistemic proposal in the literature,

or else a competitor to the extant epistemic proposals. Those in the literature amount to
different ways of describing problematic relationships between (probability-raising) evidence
and hypotheses, whether in terms of “normic support” (Smith 2018), “safety” (Pritchard
2018), “sensitivity” (Enoch et al. 2012) or “causal connection” (Thomson 1986). For a review
of these various proposals, see Gardiner (2018).
7 There are other analogous cases, like the well-known ‘Gatecrashers’ case (due to Cohen

(1977). This case involves a rodeo event where it is known that only 10 seats out of 80 were
paid for, and there is no evidence identifying who paid. The question is whether someone
drawn randomly from the crowd can be found liable for gatecrashing.
8 This example is also found in Redmayne (2008).

10
sufficiently probative. This is not straightforwardly a case of sampling from
an ad hoc population with known characteristics. It rather seems an inference
about the perpetrator of an accident based on very little, and yet strangely
probative, causally relevant data. The issue here thus seems more apparently
an epistemic one.
I contest, however, that Blue Bus is more similar to Prisoners than first
meets the eye, or at least, it is best seen that way if treated as a paradigm case
of ‘general’ evidence. It cannot merely be a case of paucity of evidence, since
many think that similarly scant witness evidence does not raise the same issues.
Indeed the following case is typically thought a useful contrast with Blue Bus:

Blue Bus Witness. Mrs. Brown is run down by a bus on Victoria


Street; 50 percent of the buses that travel along this street are owned
by the blue bus company, and 50 percent by the red bus company.
The only witness is Mrs. Brown, who identifies the offending bus
as a blue one, and who is deemed to correctly identify blue and red
buses 60% of the time. Hence the blue bus company is found liable.9

Apparently, the evidence in cases like Blue Bus Witness is not taken to be par-
ticularly problematic; it is thought to be suitably ‘individualised’. (Any yet it
seems to have the same probative value as the evidence in Blue Bus!) Presum-
ably this is because the reference class that grounds the inference concerning
the colour of the offending bus is constructed according to causally relevant
factors—the sighting of the accident by Mrs. Brown who has given capacities
to correctly identify colour. So while the evidence for holding the blue bus
company rests on sparse causally relevant factors. one might say, it nonetheless
rests on at least some causally relevant factors in a relatively straightforward
way.10
By contrast, the evidence in Blue Bus has a more complicated form: it serves
as a base rate for the companies’ respective responsibility for bus accidents that
depends entirely on the proportion of the buses that the company is running
on the road in question. But this base rate may be seen as extremely arbitrary
for any given bus company in that it depends on the activities of the other bus
company or companies. The relative number of buses on a road may have some
causal bearing on any given accident on that road, but there are many further
causal factors, so the road-accident groups are in a sense highly heterogeneous.
Moreover, the composition of buses on the road could have been different for no
reason to do with a company’s own conduct on the road. Like Prisoners, this is
an unusual case where the evidence has significant probative strength and yet
has little to do, causally speaking, with the hypothesis in question.11 The blue
bus company is thus very unlucky in being judged according to this evidence.
9 See Redmayne (2008) for a rough presentation of this contrast case.
10 That said, there remains a tricky issue of estimating the reliability of the witness in cases
like the one at hand. Some argue that the reliability of eye-witness testimony is typically
overestimated in legal trials (Arkowitz and Lilienfeld, 2010; Buckhout 1980).
11 And yet in this case, unlike Prisoners, there is no actual random selection from a fixed

11
The question remains as to whether the luck associated with base-rate group-
ings like in Prisoners or Blue Bus presents an epistemic or rather a moral issue.
Perhaps the difference between the two is purely a matter of labelling. Not
so, I say. To begin with, there is an interpretative difference in appealing to
(un)fairness as opposed to poor epistemic support, even if the set of problematic
cases were identical either way. The different diagnoses of what is problematic
may call for different remedies. As noted above, on my proposal, ‘general’
evidence amounts to just one extreme kind of evidential luck and associated
unfairness. So any attempt to mitigate what is problematic about ‘general’ evi-
dence should presumably be part of a broader approach to mitigating evidential
and perhaps other kinds of moral luck too.
But I claim that there is an even more substantive difference between my
fairness proposal and any corresponding epistemic proposal regarding ‘general’
evidence. They have differing implications. This becomes apparent when we
generalise to applications beyond the legal domain. The epistemic version of the
proposal says that an arbitrary relationship between evidence and hypothesis
matters for any decision, or at least any for which the stakes of, say, departing
from the status quo, are sufficiently high. This is not so for the fairness proposal.
Consider the following example:

Bio-engineering. An agricultural community struggling to produce


enough to sustain their livelihoods in what seems an increasingly
hostile climate is entertaining whether to continue with their stan-
dard crop for another year, hoping for fortuitous weather, or instead
plant a genetically modified crop that thrives exclusively in hostile
weather. For various reasons the whole community must plant the
same crop. These are the only crop options. For hostile weather,
the modified crop would be a modest improvement over the standard
crop. But if the weather were to turn, then the modified crop would
be a disaster relative to the standard crop.12

According to the probability-threshold (a.k.a. standard decision theory) ap-


proach, the probability of hostile weather must be very high in this case to
warrant planting the modified crop. Precisely how high will depend on the
relative value of the four outcomes. Let’s say that, given the evidence, the
probability of hostile weather does indeed surpass this threshold.
Is it problematic for the community to pursue the option of planting the
modified crop on this basis? Well, on the relevant epistemic proposal, the an-
swer is: “it depends whether the evidence has an appropriately non-arbitrary
population, which might be reason to doubt whether the evidence in Blue Bus is really as
probative as it is taken to be. That would be a much simpler diagnosis of the epistemic
problem with the evidence in Blue Bus. I grant in my discussion, however, that the evidence
may well be taken as probative, conferring a probability of 0.6 that the blue bus company is
liable.
12 The relative value of the four possible outcomes is intended to roughly match those of

legal decisions about whether to find a defendant guilty/liable.

12
connection to the truth of the hypothesis”.13 There are no similar concerns
about the fairness of this decision, however. That is because the situation is
not one in which a single individual is in danger of being treated differently from
her fellow citizens. It is not a case in which an individual may complain of unfair
treatment relative to others. The community members are rather choosing an
option that will affect them all similarly.
The above may be contrasted with a different example that is more akin to
the legal case:

Extra care. Special extra care is available for those patients with a
given disease who have some further complication C. With the ex-
tra care, those with C are restored to a similar tolerable position as
others with the disease who do not have C. Without the care, those
with C suffer further significant health decline. The extra care is
very costly to administer, however, on a per patient basis. And if
the patient does not have C, then the extra care does not improve
their health at all. A medical team is assessing some patients. Unfor-
tunately, due to a filing mix-up, all that is known about the patients
is the percentage in the room that have condition C. In room 1, 85%
of patients have condition C, while in room 2, 50% of patients have
condition C. Unfortunately, there are no further tests available and
the efficacy of the extra care is time sensitive, so whether or not to
administer it to each patient must be decided on the spot.14

According to the probability-threshold (a.k.a. standard decision theory) ap-


proach, the probability of having C must be reasonably high to warrant ad-
ministering the extra care. Again, just how high depends on the relative social
value of the possible decision outcomes. Let’s say the probability threshold is
in the vicinity of 0.8. Then all those (randomly selected) from room 1 are given
the extra care while all those (randomly selected) from room 2 are denied the
extra care.
Is it problematic to decide on the distribution of extra care in this way? On
the fairness proposal the answer is “in a sense, yes”, since here the value the
evidence takes may be regarded arbitrary in relation to whether or not a patient
has complication C, for the same reasons as per the Prisoners case. There is
unfairness in the comparative treatment of patients. Any epistemic proposal
which similarly regards the relation between evidence and hypothesis here to be
arbitrary or otherwise inadequate will also regard the decision procedure to be
problematic, in that it is not justified given the evidence.
13 We are putting aside criticisms of the standard decision model that have to do with it

being insufficiently risk averse. In any case, greater risk aversion will simply entail a higher
probability threshold for success on the risky option as the standard for rationally pursuing
this option.
14 Again, the relative value of the four possible outcomes is intended to roughly match those

of legal decisions about whether to find a defendant guilty/liable.

13
So, sometimes fairness worries and corresponding epistemic worries about
the use of ‘general’ evidence go hand in hand (even if they invite different sorts
of remedies). Other times, however, they do not. Hence, these are very different
kinds of proposals regarding what is problematic about ‘general’ evidence in
legal and other public decision making.

4.2 Moral alternatives


My proposal is that the use of ‘general’ evidence is merely a striking case of ev-
idential luck and associated unfairness. Accordingly, what is problematic about
‘general’ evidence, on my view, is not unique to this kind of evidence after all.
We find evidential luck everywhere, in different forms and to different degrees,
in decisions about individuals’ entitlements.15 Hence I diagnose the problem of
‘general’ evidence as a moral one of unfairness resulting from evidential luck,
and therefore not unique or of special importance. But this downgrading might
be reason to find my account unsatisfactory. It might be claimed that I miss
the real moral worry about ‘general’ evidence, and that this worry is best ex-
emplified by cases that I have not considered.
Indeed, nearly all accounts of why ‘general’ evidence is morally problematic
appeal rather to ‘profile’ evidence as paradigmatic. Here is one interesting such
case taken from Di Bello and O’Neil (2020, 148):

Smuggling. The Vue brothers, of Hmong ancestry, were charged


with smuggling the opium that was sent to their residence in the
Minneapolis area. On the basis of an expert witness’s testimony, it
is derived that, in the Minneapolis area, someone who is of Hmong
ancestry is 297 times more likely to be trafficking drugs as compared
to someone who is not of Hmong ancestry. On the basis of this in
addition to other evidence, the defendants were convicted.16

Many find this verdict to be problematic to the extent that the ethnicity evidence
plays a significant role in its determination.17
To begin with, I concede that my proposal does not explain why profile
evidence stands out as particularly morally egregious. However, I will argue
15 While the evidence in Blue Bus Witness is less striking than that in Blue Bus, its use also

introduces unfairness, since the very presence of this evidence is partly a matter of luck for
the defendant (as per the security-camera examples in section 1).
16 This case description follows Di Bello and O’Neil’s (2020, 148)) account of the real court

case U.S. v. Vue, 13 F.3d 1206 (8th Circuit, 1994). Di Bello and O’Neil note that the statistic
they report derives from the testimony of an expert witness that 95% of opium smuggling cases
in the Minneapolis area involved people of Hmong ancestry, and yet such people constitute
only 6% of the population. This gives (.95/.06)/(.05/.94) = 297, for how many times more
likely it is that someone of Hmong ancestry is trafficking opium relative to other citizens.
17 Indeed, Di Bello and O’Neil (2020, 148) note that ‘the convictions were reversed on appeal,

on the grounds that the jury had been improperly invited to “put the Vues” racial and cultural
background into the balance in determining their guilt’ (U.S. v. Vue, 1213).

14
that the reason profile evidence is particularly morally egregious has to do with
the content of the evidence, rather than its being general. Moreover, I will show
that, to the extent that there is a further moral issue to do with the ‘generality’
of profile evidence, it is best captured by my proposal.
The special moral problem with public use of profile evidence, I claim, is that
its message (at the very least) harms members of the group concerned. Take
the Smuggling case. If we are to assume that there is a real causal relationship
between Hmong ethnicity and smuggling, such that the evidence is genuinely
probative, this presumably has to do with ongoing and/or past oppression and
associated limited opportunities for Hmong people across generations.18 In that
case the appeal to such profiling evidence, even when not ill-intentioned, plau-
sibly harms the Hmong community by reinforcing negative stereotypes about
the community arising from circumstances of oppression. The harm may be
regarded merely an expressive one—a reminder of the original oppression—or it
may be regarded a further perpetuation of that oppression (for discussion, see
Risse and Zeckhauser 2004 and Lever 2005; see too Hellman 2014). Either way,
this is a tangible harm associated with profile evidence that warrants treating
it differently from other sorts of evidence.
Profile evidence is thus problematic because it is socially costly. The costs are
not so much to do with how the evidence is procured (as in harmful torture);
they are incurred rather in the public use of the evidence, since the message
conveyed is harmful.19 One must then ask whether the probative value of profile
evidence is worth its social costs.20
All this is to say that what is particularly morally problematic about pro-
file evidence is not to do with its ‘generality’ per se. It is rather about the
harms/social costs of profile evidence that owe to its specific content or mes-
sage, and how that message relates to wider social events and human psychology.
One might insist, however, that there is a further moral concern about profile
evidence that applies even in cases where ‘no-one gets hurt’, and so is more
evidently related to the ‘generality’ of the evidence. It is the disrespect towards
the individual defendant that evidence based on a group profile gives rise. In
early work on this issue, Wasserman (1991, 940) puts it this way: ‘We are es-
18 Note that there may not be a real causal relationship in the first place. Many appeals to

profile evidence may well be criticised on ordinary epistemic grounds—for invoking statistics
that are compromised by reporting error, or else that cannot be extrapolated or ‘projected’ to
the case at hand. See Leslie (2017), for instance, on why we may be prone to mistakes when
it comes to social generalizations. In this discussion we are supposing the hard case—profile
evidence resting on non-accidental statistical correlation.
19 Enoch et al. (2012) suggest that there may be another subtle yet tangible harm and

thus social cost associated with the public use of profile evidence, namely, that it changes
individuals’ incentives in a socially undesirable way, since verdicts are less sensitive to the
agent’s actions. Others have challenged this account of incentives (e.g., Gardiner (2018), Di
Bello and O’Neil (2020)).
20 This is to say that tangible harms or costs of evidence, whether they be incurred in

the procurement or else the expression of the evidence, should be factored into the decision
theoretic rationale for what counts as admissible evidence in a legal setting and when the
admissible evidence warrants conviction.

15
pecially concerned about falsely attributing misconduct to a defendant based
on the frequency of similar misconduct by others or by the defendant himself.’
Later he talks of the special ‘insult to our individuality and autonomy’ (ibid,
949) that is involved in the use of such evidence.
This broad idea—that profile evidence is morally problematic because it fails
to treat people as individuals in some sense—has since been taken up by vari-
ous authors, including Duff (1998), Pundik (2016), Basu (2018 & forthcoming),
Basu and Schroeder (forthcoming), Moss (2018), Bolinger (2020) and Di Bello
and O’Neil (2020)). Yet although this idea is appealing, it has proved notori-
ously difficult to provide a concrete account of what this problem with profile
evidence really amounts to, given the assumption that the evidence in question
is probative, and more generally, that human behaviour is to some extent pre-
dictable. (Beeghly (2018), for instance, discusses the difficulty of spelling out
precisely what the failure to treat someone as an individual comes to.) In the
remainder of this section, I argue that evidential luck, as per my proposal, is one
way to fill in the further details. Moreover, I argue that my proposal has some
advantages; others rest on more contentious claims about what an individual is
owed when it comes to the inferences made about them.
The success of my account’s explanation of the apparent disrespect of profile
evidence depends on whether such evidence is a noteworthy instance of eviden-
tial luck, and thus relevantly like the evidence in Prisoners. If so, it is right to
have moral concerns about the inculpating evidence in cases like Smugglers that
go beyond its expressive harms. (And we might well say that these concerns
are about ‘respect for the individual’, even if evidential luck is not exclusive to
profile evidence.) The issue turns on whether the evidence in Smugglers involves
a grouping that is highly unrepresentative of the individual(s) with respect to
causal properties that bear on the hypotheses in question, and yet is probative.
Is the grouping of people with Hmong ancestry representative of the Vue broth-
ers with respect to their propensity to traffic opium? (Here the defendants are
not randomly drawn from an ad hoc population with known characteristics, as
in Prisoners. The evidence, if probative, rests rather on a ‘projectible’ general-
isation.) Let us assume, as we have been doing, that this is genuine evidence:
Hmong ancestry does have some causal bearing on opium trafficking (albeit via
a convoluted pathway that is mediated by unjust social structures) that is well
estimated by the given statistical data. Even then, however, the Hmong group-
ing is presumably very heterogeneous; there may be a salient sub-population
of Hmong that, due to a confluence of other causal factors, behaves very dif-
ferently from the rest of the population. This heterogeneity introduces luck:
unfortunately for them, the other subpopulations are bundled together with the
subpopulations with higher propensity for crime, and judged as one.21
21 These remarks suggest a much more complicated picture of evidential luck associated

with ‘general’ evidence (let alone other forms of evidential luck) that unfortunately cannot be
fleshed out in full here. The idea is that not all groupings based on causally relevant factors
are on a par. For instance, those groupings based on what is known as ‘INUS’ conditions
(insufficient but necessary parts of an unnececessary but sufficient cause) may be regarded

16
Can we do better in explaining what is especially morally problematic, be-
yond any tangible harms, about the use of profile evidence? I consider now a
couple of interesting alternative proposals in the literature that may appear to
do better. They can be read as substantive accounts of why profile evidence
involves grouping individuals with others in a way that is unfair for and thus
perhaps disrespectful to them. Bolinger (2020) suggests that unfair groupings
are those that are involuntary. And Di Bello and O’Neil (2020) argue that un-
fair groupings are those that are prejudiced against some individuals whatever
the particulars of any crime that occurs. (Technically speaking, Di Bello and
O’Neil are concerned with groupings that have unequal antecedent burdens on
innocents, i.e., yield unequal risks of false conviction prior to any crimes being
committed.) Both proposals plausibly explain what is especially morally prob-
lematic about profile evidence. But this virtue of these accounts is indicative
of a problem, on my view, since it means that they misidentify the paradigm
cases of ‘general’ evidence. I do not pretend to offer knockdown arguments to
this effect. I claim merely that the proposals in question, as compared to my
own, rest on much more contentious claims about what individuals are owed
with respect to evidence and inference. Hence they are harder to defend.
Start with voluntariness. It comes in degrees, which is a good thing insofar
as one thinks, as I do, that (un)fairness comes in degrees. But this notion of
(un)fairness does not seem to get the right ordering of cases. The more extreme
cases of evidence based on involuntary properties or groupings of a person do
not seem to be the more extreme cases of unfairness in the use of the evidence.
For instance, if there were some sufficient (probability-raising) biological cause
for aggressive behaviour then surely an individual’s biological profile in this
respect would be reasonable evidence in the courtroom for their propensity for
aggressive behaviour, despite this being an involuntary property or grouping.
On the other hand, there are cases of evidence arising from voluntary properties
or groupings of a person that nonetheless seem deeply unfair. For instance, we
might suppose that the defendant in Prisoners actively chose to exercise in the
yard in question, and perhaps even had full knowledge of the vandalising goals
of their colleagues. It is surely still unfair to penalise the defendant merely for
being in a yard with more rather than fewer criminal colleagues. Agents may be
wise to cultivate evidentially-favourable properties or put themselves amongst
company that looks good on average, but do we really have some right to control
the evidence about us in this way? And to the extent that agents do not exercise
control of the evidence when they could have, do they lose any claim to unfair
treatment on the basis of the evidence used against them?
Turn now to groupings that are prejudiced against individuals, whatever the
particulars of the crime. The idea is that we should remove unfairness by treat-
ing as inadmissible any evidence that upsets equal risks of false conviction, prior
highly heterogeneous and thus introduce much luck compared to those groupings based on
sufficient causes. The evidence in Prisoners involves a group that is unrepresentative in the
extreme, with respect to causally relevant factors, and so allowed us to discuss evidential luck
owing to ‘general’ evidence without getting into the thorny issues.

17
to any crimes being committed. The sort of evidence that burdens individu-
als differently at the outset is, roughly, evidence based on residing (since, say,
childhood) features of individuals that are correlated with aggression or other
behaviour related to criminality in general. Profile evidence based on involun-
tary properties of a person, like ethnicity, is the most salient kind of evidence
that fits that description. But other sorts of profile evidence based rather on
voluntary properties, like longstanding membership in some organisation such
as a rifle club, may also fit the bill. Here again, it does not seem that this
categorisation of evidence tracks our intuitions about unfair treatment. Again,
if there were a biological basis for aggression then appeal to this evidence comes
out as unfair on this proposal; and the appeal to the yard evidence in Prison-
ers does not come out as unfair (as noted by Di Bello and O’Neil (2020, 169).
More generally, it is not clear why it matters that at some point individuals had
(un)equal risks of false conviction, given that once crimes actually occur at some
later point, the risks of false conviction for defendants may be vastly different.
For one thing, it seems difficult to draw the line between the earlier life-stage
(when equality in risk profiles is important) and later life-stages (when equality
is not important) in a non-arbitrary way. What if a person at some point gets
a tattoo that is often mistaken with the logo for an aggressive gang, or takes a
job in a place that is notorious for petty crime perpetrated by people in her age
group: Do these properties unacceptably prejudice the person’s chances should
they be mistakenly convicted of a crime? When exactly should the risk of false
conviction be the same for all innocent defendants?
I maintain that my proposal is the least contentious account of the residual
moral problem to do with profile evidence owing to its ‘generality’ as opposed
to the harmful message that it conveys. The ‘generality’ problem need not
be specific to profile evidence, since it is not the primary problem with that
evidence. On my proposal, defendants have a complaint about the use of profile
and other evidence in their legal trial to the extent that it does not, roughly
speaking, reflect their conduct (more generally, it does not derive from properties
that are causally relevant to the hypotheses under consideration). This is surely
much less contentious than the alternative proposals: that individuals have a
complaint when they cannot control the evidence, or when it does not position
them well at some privileged early life-stage.

5 Concluding remarks
I have offered a proposal regarding the problem of ‘general’ evidence in the
courtroom, and plausibly any public decision setting similarly concerned with
the entitlements of individuals. What is distinct and striking about some evi-
dence classed as ‘general’, I have argued, is that the value the evidence takes de-
pends on causally arbitrary factors and yet affords significant probative strength.
While this is interesting, I claimed that it is nonetheless just one (albeit less
familiar) manifestation of the phenomenon of evidential luck. My analysis of

18
‘general’ evidence serves to draw attention to the broader phenomenon of evi-
dential luck and the different guises it may take. Any proposal to mitigate the
unfairness arising from ‘general’ evidence should accordingly be sensitive to this
bigger picture.
Even if we aimed for broad consistency in our approach to managing evi-
dential luck, however, this leaves open very different approaches. On the one
hand, one might treat ‘general’ evidence as instructive; it provides a clear view
of what is bad about evidential luck and associated unfairness. To the extent
that we want to mitigate the role of luck here (whether by downgrading or ex-
cluding ‘general’ evidence as inadmissible), we should draw the same conclusion
about other forms of evidential luck too. Alternatively, one might rather treat
our tolerance of other forms of evidential luck as instructive; this is to say that
‘general’ evidence obscures rather than clarifies the issue and that our reactions
to ‘general’ evidence are the result of bias. Accordingly, we would recognise
the widespread unfairness in the treatment of individuals due to evidential luck,
but would see this as a merely regrettable circumstance that cannot be fruitfully
changed.
The first is of course the more difficult approach in that it opens up a host of
further questions regarding how different magnitudes of evidential luck should be
measured and how evidential luck of any given magnitude ought to be mitigated.
Consider just ‘general’ evidence. What is the right ordering of evidence in terms
of its ‘generality’ and thus evidential luck? It depends on how exactly we should
measure more and less arbitrary values for evidence variables, in light of the
probative strength of the evidence. As the discussion of examples in this paper
makes clear, that is no small issue. Then there is the further question, as regards
mitigating evidential luck of varying severity goes, of how evidential luck and
associated (un)fairness to individuals should be weighed against (in)accuracy in
legal and other public decisions.
All the more reason to go with the second approach, one might think. We
should simply accept widespread evidential luck and associated unfairness in
legal and other public decision making about individual entitlements. But it is
not obvious that this is right. That the finer details of evidential luck and its
relative undesirability in decision making are difficult and messy does not seem
reason enough to ignore these considerations altogether.

References
[1] Arkowitz, H. and S.O. Lilienfeld, 2010. ‘Do the “Eyes” Have It?’ in Scientific
American Mind , 20(7) (January 2010): 68–69.
[2] Basu, R. 2018. ‘Beliefs that Wrong’, Ph.D. Thesis, University of Southern Cali-
fornia.
[3] Basu, R. forthcoming. ‘The Specter of Normative Conflict: Does Fairness Require
Inaccuracy?’, in E. Beeghly and A. Madva An Introduction to Implicit Bias:
Knowledge, Justice, and the Social Mind , New York: Routledge.

19
[4] Basu, R. and M. Schroeder. 2018. ‘Doxastic Wronging’, in B. Kim and M. Mc-
Grath (eds.), Pragmatic Encroachment in Epistemology, Routledge.
[5] Beeghly, E. 2018. ‘Failing to Treat Persons as Individuals’, Ergo, 5: 687–711.
[6] Blackstone, W. 1765. Commentaries on the Laws of England , The Legal Classics
Library.
[7] Blome-Tillmann, M. 2015. ‘Sensitivity, Causality, and Statistical Evidence in
Courts of Law’, Thought, 4: 102–112.
[8] Bolinger, R.J. 2020. ‘The Rational Impermissibility of Accepting (Some) Racial
Generalisations’, Synthese, 197(6): 2415–2431.
[9] Buckhout, R. 1980. ‘Nearly 2,000 Witnesses Can be Wrong’. Bulletin of the Psy-
chonomic Society, 16(4): 307–310.
[10] Cohen, J. 1977. The Probable and the Provable, Oxford: OUP.
[11] Di Bello, M. 2018. ‘Trial by Statistics: Is a High Probability of Guilt Enough to
Convict?’ Mind , online https://doi.org/10.1093/mind/fzy026.
[12] Di Bello, M. and C. O’Neil. 2020. ‘Profile Evidence, Fairness, and the Risks of
Mistaken Convictions’, Ethics, 130: 147–178.
[13] Duff, A. 1998. ‘Dangerousness and Citizenship’, in A. Ashworth and M. Wasik
(eds.), Fundamentals of Sentencing Theory: Essays in Honour of Andrew von
Hirsch, Oxford: OUP, pp. 141–163.
[14] Enoch, D., L. Spectre, and T. Fisher, 2012. ‘Statistical Evidence, Sensitivity, and
the Legal Value of Knowledge’, Philosophy and Public Affairs, 40(3): 197–224.
[15] Gardiner, G. 2018. ‘Legal burdens of proof and statistical evidence’, in D. Coady
and J. Chase (eds.), The Routledge Handbook of Applied Epistemology, Routledge,
pp. 179–195.
[16] Godfrey-Smith, P. 2011. ‘Induction, Samples and Kinds’, in J. Campbell,
M. O’Rourke and M. Slater (eds.), Carving Nature at Its Joints: Natural Kinds
in Metaphysics and Science, Cambridge, MA: MIT Press, pp. 33–52.
[17] Günther, M. m.s. ‘Is Sensitive Evidence Individual?’.
[18] Hellman, D. 2014. ‘Racial Profiling and the Meaning of Racial Categories’, in
A.I. Cohen and C.H. Wellman (eds.), Contemporary Debates in Applied Ethics,
2nd ed., Hoboken, NJ: Wiley-Blackwell, pp. 23–44.
[19] Leslie, S.J. 2017. ‘The Original Sin of Cognition: Fear, Prejudice, and General-
ization’, Journal of Philosophy, 114: 393-421.
[20] Lever, A. 2005. ‘Why Racial Profiling is Hard to Justify: A Response to Risse
and Zeckhauser’, Philosophy and Public Affairs, 33: 94-110.
[21] Moss, S. 2018. ‘IX—Moral Encroachment’, Proceedings of the Aristotelian Society,
118: 177-205.
[22] Nesson, C.R. 1979. ‘Reasonable Doubt and Permissive Inferences: The Value of
Complexity’, Harvard Law Review, 92: 1187–1225.
[23] Pritchard, D. 2018. ‘Legal Risk, Legal Evidence and the Arithmetic of Criminal
Justice’, Jurisprudence, 9(1): 108–119.
[24] Pundik, A. 2016. ‘Freedom and Generalisation’, Oxford Journal of Legal Studies,
37: 189–216.

20
[25] Redmayne, M. 2002. ‘The Relevance of Bad Character’, Cambridge Law Journal ,
61(3): 684–714.
[26] Redmayne, M. 2008. ‘Exploring the Proof Paradoxes’, Legal Theory, 14(4): 281–
309.
[27] Risse, M. and R. Zeckhauser, 2004. ‘Racial Profiling’, Philosophy and Public Af-
fairs, 32: 131-70.
[28] Schroeder, M. 2018. ‘When Beliefs Wrong’, Philosophical Topics, 46: 115-27.
[29] Smith, M. 2010. ‘What else justification could be’, Noûs, 44(1): 10–31.
[30] Smith, M. 2018. ‘When Does Evidence Suffice for Conviction?’, Mind , 127(508):
1193–1218.
[31] Steele, K. and M. Colyvan, ‘Meta-uncertainty and the Proof Paradoxes’, ms.
[32] Thomson, J.J. 1986. ‘Liability and Individualized Evidence’, in W.A. Parent (ed.),
Rights, Restitution and Risk: Essays in Moral Theory, Harvard MA: Harvard
University Press, pp. 225–250.
[33] Wasserman, D.T. 1991. ‘The Morality of Statistical Proof and the Risk of Mis-
taken Liability’, Cardozo Law Review , 13: 934–76.

21

You might also like