Punishment

Punishment involves the deliberate infliction of suffering on a supposed or actual offender

for an offense such as a moral or legal transgression. Since punishment involves inflicting
a pain or deprivation similar to that which the perpetrator of a crime inflicts on his victim, it
has generally been agreed that punishment requires moral as well as legal and political
justification. While philosophers almost all agree that punishment is at least sometimes
justifiable, they offer various accounts of how it is to be justified as well as what the
infliction of punishment is designed to protect – rights, personal autonomy and private
property, a political constitution, or the democratic process, for instance. Utilitarians
attempt to justify punishment in terms of the balance of good over evil produced and thus
focus our attention on extrinsic or consequentialist considerations. Retributivists attempt a
justification that links punishment to moral wrongdoing, generally justifying the practice on
the grounds that it gives to wrongdoers what they deserve; their focus is thus on the
intrinsic wrongness of crime that thereby merits punishment. “Compromise” theorists
attempt to combine these two types of theories in a way that retains their perceived
strengths while overcoming their perceived weaknesses. After discussing the various
attempts at justification, utilitarian and retributive approaches to determining the amount of
punishment will be examined. Finally, the controversial issue of capital punishment will be
briefly discussed.

Table of Contents
1.Utilitarianism
1.Utilitarian Justification
2.Objection and Response
2.Retributivism
1.Retributive Justification
2.Objection and Response
3.Compromise Theories
1.Hart’s Theory
2.Objection and Response
4.Amount of Punishment
1.Utilitarians on Amount
2.Retributivists on Amount
5.Capital Punishment
6.Conclusion
7.References and Further Reading

1. Utilitarianism

a. Utilitarian Justification
Utilitarianism is the moral theory that holds that the rightness or wrongness of an action is
determined by the balance of good over evil that is produced by that action. Philosophers
have argued over exactly how the resulting good and evil may be identified and to whom
the greatest good should belong. Jeremy Bentham identified good with pleasure and evil
with pain and held that the greatest pleasure should belong to the greatest number of
people. John Stuart Mill, perhaps the most notable utilitarian, identified good with
happiness and evil with unhappiness and also held that the greatest happiness should
belong to the greatest number. This is how utilitarianism is most often discussed in the
literature, so we will follow Mill in our discussion.

When attempting to determine whether a punishment is justifiable, utilitarians will attempt

to anticipate the likely consequences of carrying out the punishment. If punishing an
offender would most likely produce the greatest balance of happiness over unhappiness
compared with the other available options (not taking any action, publicly denouncing the
offender, etc.), then the punishment is justified. If another available option would produce a
greater balance of happiness over unhappiness, then that option should be chosen and
punishment is unjustified.

Clearly, crimes tend to produce unhappiness, so in seeking to promote a state of affairs in

which the balance of happiness over unhappiness is maximized, a utilitarian will be highly
concerned with reducing crime. Traditionally, utilitarians have focused on three ways in
which punishment can reduce crime. First, the threat of punishment can deter potential
offenders. If an individual is tempted to commit a certain crime, but he knows that it is
against the law and a punishment is attached to a conviction for breaking that law, then,
generally speaking, that potential offender will be less likely to commit the crime. Second,
punishment can incapacitate offenders. If an offender is confined for a certain period of
time, then that offender will be less able to harm others during that period of time. Third,
punishment can rehabilitate offenders. Rehabilitation involves making strides to improve
an offender’s character so that he will be less likely to re-offend.

Although utilitarians have traditionally focused on these three ways in which punishment
can reduce crime, there are other ways in which a punishment can affect the balance of
happiness over unhappiness. For example, whether or not a given offender is punished
will affect how the society views the governmental institution that is charged with
responding to violations of the law. The degree to which they believe this institution is
functioning justly will clearly affect their happiness. Utilitarians are committed to taking into
account every consequence of a given punishment insofar as it affects the balance of
happiness over unhappiness.

b. Objection and Response

Perhaps the most common objection to the utilitarian justification of punishment is that its
proponent is committed to punishing individuals in situations in which punishment would
clearly be morally wrong. H.J. McCloskey offers the following example:

Suppose a utilitarian were visiting an area in which there was racial strife,
and that, during his visit, a Negro rapes a white woman, and that race riots
occur as a result of the crime, white mobs, with the connivance of the police,
 bashing and killing Negroes, etc. Suppose too that our utilitarian is in the
area of the crime when it is committed such that his testimony would bring
about the conviction of a particular Negro. If he knows that a quick arrest will
stop the riots and lynchings, surely, as a utilitarian, he must conclude that he
has a duty to bear false witness in order to bring about the punishment of an
innocent person (127).

A utilitarian is committed to endorsing the act that would be most likely to produce the
greatest balance of happiness over unhappiness, and, in this situation, it appears that the
act that meets this criterion is bearing false witness against an innocent person. But, so
the argument goes, it cannot be morally permissible, let alone morally mandatory, to
perform an act that leads directly to the punishment of an innocent person. Therefore,
since the utilitarian is committed to performing this clearly wrong act, the utilitarian
justification must be incorrect.

The standard utilitarian response to this argument demands that we look more closely at
the example. Once we do this, it supposedly becomes clear that the utilitarian is not
committed to performing this clearly wrong act. In his reply to McCloskey’s argument,
T.L.S. Sprigge states that if faced with the decision presented in the example, a “sensible
utilitarian” will attach a great deal of weight to the near-certain fact that framing an innocent
man would produce a great deal of misery for that man and his family. This consideration
would receive such weight because “the prediction of misery… rests on well confirmed
generalizations” (72). Furthermore, the sensible utilitarian will not attach much weight to
the possibility that framing the man would stop the riots. This is because this prediction
“will be based on a hunch about the character of the riots” (72). Since well confirmed
generalizations are more reliable than hunches, happiness is most likely to be maximized
when individuals give the vast majority of the weight to such well confirmed generalizations
when making moral decisions. Therefore, since the relevant well confirmed generalization
tells us that at least a few people (the innocent man and his family) would be made
miserable by the false testimony, the utilitarian would give much weight to this
consideration and choose not to bear false witness against an innocent man.

This type of response can in turn be challenged in various ways, but perhaps the best way
to challenge it is to point out that even if it is true that the greatest balance of good over
evil would not be promoted by punishing an innocent person in this situation, that is not the
reason why punishing an innocent person would be wrong. It would be wrong because it
would be unjust. The innocent man did not rape the woman, so he does not deserve to be
punished for that crime. Because utilitarianism focuses solely on the balance of happiness
over unhappiness that is produced by various actions, it is unable to take into account
important factors such as justice and desert. If justice and desert cannot be incorporated
into the theory, then the punishment of innocents cannot be ruled out as unjust, so a
prohibition against it will have to be dependent upon the likelihood of various
consequences. This strikes many theorists as problematic.
2. Retributivism

a. Retributive Justification
Regarding retributive theories, C.L. Ten states that, “There is no complete agreement
about what sorts of theories are retributive except that all such theories try to establish an
essential link between punishment and moral wrongdoing” (38). He is surely right about
this, so, therefore, it is difficult to give a general account of retributive justification.
However, it is possible to state certain features that characterize retributive theories
generally. Concepts of desert and justice occupy a central place in most retributive
theories: in accordance with the demands of justice, wrongdoers are thought to deserve to
suffer, so punishment is justified on the grounds that it gives to wrongdoers what they
deserve. It is instructive to look at the form that a particular retributive theory can take, so
we will examine the views of Immanuel Kant.

Kant invokes what he refers to as the “principle of equality” in his discussion of

punishment. If this principle is obeyed, then “the pointer of the scale of justice is made to
incline no more to the one side than the other” (104). If a wrongful act is committed, then
the person who has committed it has upset the balance of the scale of justice. He has
inflicted suffering on another, and therefore rendered himself deserving of suffering. So in
order to balance the scale of justice, it is necessary to inflict the deserved suffering on him.
But it is not permissible to just inflict any type of suffering. Kant states that the act that the
person has performed “is to be regarded as perpetrated on himself” (104). This he refers
to as the “principle of retaliation”. Perhaps the most straightforward application of this
principle demands that murderers receive the penalty of death. So, for Kant, the
justification of punishment is derived from the principle of retaliation, which is grounded in
the principle of equality.

The concepts of desert and justice play a central role in Kant’s theory, and they are applied
in a way that rules out the possibility of justifying the punishment of innocents. Since an
innocent person does not deserve to be punished, a Kantian is not committed to punishing
an innocent person, and since it seems to some that utilitarians are committed to
punishing innocents (or participating in the punishment of innocents) in certain
circumstances, Kant’s theory may seem to be superior in this respect. Recall that the
failure to take desert and justice into consideration is thought by many to be a major
problem with utilitarian theory. However, while Kantian theory may seem superior because
it takes desert and justice into account, an influential criticism of the theory challenges the
idea that punishment can be justified on the grounds of justice and desert without requiring
that the balance of happiness over unhappiness be taken into account.

b. Objection and Response

Gertrude Ezorsky argues that we should test the Kantian position and other retributive
positions that resemble it “by imagining a world in which punishing criminals has no further
effects worth achieving” (xviii). In this world, punishment does not deter or rehabilitate. For
whatever reason, incapacitation is impossible. In addition, victims receive no satisfaction
from the punishment of those who have harmed them. In this world, a Kantian would be
committed to the position that punishments still ought to be inflicted upon wrongdoers.
Furthermore, the individuals that populated this world would be morally obligated to punish
wrongdoers. If they failed to punish wrongdoers, they would be failing to abide by the
dictates of justice. But surely it is quite odd to hold that these individuals would be morally
obligated to punish when doing so would not produce any positive effects for anyone.
According to Ezorsky, this terribly odd consequence suggests that the Kantian theory is
problematic.

Kant would not agree that this consequence of his theory is odd. According to Kant, “if
justice and righteousness perish, human life would no longer have any value in the world”
(104). So, even the inhabitants of our imaginary world are obliged to ensure that “every
one may realize the desert of his deeds” (106). If they do not live up to this obligation, then
they will be failing to abide by the dictates of justice, and their lives will be of lesser value.
Of course, critics of the Kantian theory are unlikely to be persuaded by this response.
Indeed, it is appropriate to be highly skeptical of a conception of justice that holds that
justice can be promoted without anyone’s welfare being promoted.

As stated earlier, many of the theories that are referred to as “retributive” vary significantly
from one another. However, as the Kantian theory possesses many central features that
other retributive theories possess, criticisms similar to Ezorsky’s have been leveled against
many of them. Predictably, the responses to these criticisms vary depending on the
particular theory.

3. Compromise Theories
Many theorists have attempted to take features of utilitarianism and retributivism and
combine them into a theory that retains the strengths of both while overcoming their
weaknesses. The impetus for attempting to develop this sort of theory is clear: the idea
that punishment should promote good consequences, such as the reduction of crime,
surely seems attractive. However, the idea that it would be justified to punish an innocent
in any circumstance where such punishment would be likely to promote the greatest
balance of happiness over unhappiness surely seems wrong. Likewise, the idea that
justice and the desert of the offender should play a central role in a justification of
punishment is attractive, while being committed to punishing an offender even when
nobody’s welfare would be promoted as a result seems to be problematic. So, each type of
theory seems to have positive and negative aspects. But how to combine these seemingly
opposed theories and produce a better one? Is a compromise between them really
possible? In an attempt to explore this possibility, we will examine the theory of H.L.A.
Hart.

a. Hart’s Theory
According to Hart, in order to clarify our thinking on the subject of punishment,
 What is needed is the realization that different principles… are relevant at
different points in any morally acceptable account of punishment. What we
should look for are answers to a number of different questions such as: What
justifies the general practice of punishment? To whom may punishment be
applied? (3)

The failure to separate these questions from one another and consider that they might be
answered by appealing to different principles has prevented many previous theorists from
generating an acceptable account of punishment. Hart states that the first question (“What
justifies the general practice of punishment?”) is a question of “General Justifying Aim” and
ought to be answered by citing utilitarian concerns. The second (“To whom may
punishment be applied?”) is a question of “Distribution” and ought to be answered by citing
retributive concerns. So, the general practice is to be justified by citing the social
consequences of punishment, the main social consequence being the reduction of crime,
but we ought not be permitted to punish whenever inflicting a punishment is likely to
reduce crime. In other words, we may not apply punishment indiscriminately. We may only
punish “an offender for an offense” (9). With few exceptions, the individual upon whom
punishment is inflicted must have committed an offense, and the punishment must be
attached to that offense.

Hart’s theory attempts to avoid what may have appeared to be an impasse blocking the
construction of an acceptable theory of punishment. Utilitarian concerns play a major role
in his theory: the practice of punishment must promote the reduction of crime, or else it is
not justifiable. But retributive concerns also play a major role: the range of acceptable
practices that can be engaged in by those concerned with reducing crime is to be
constrained by a retributive principle allowing only the punishment of an offender for an
offense. Hart’s theory, at the very least, represents a plausible attempt at a “compromise”
between those inclined towards utilitarianism and those inclined towards retributivism.

Hart does admit that on certain occasions the principle stating that we may only punish an
offender for an offense (referred to as the principle of “retribution in Distribution”) may be
overridden by utilitarian concerns. When the utilitarian case for punishing an innocent
person is particularly compelling, it may be good for us to do so, but “we should do so with
the sense of sacrificing an important principle” (12). Many people will agree with Hart that it
may be necessary to punish an innocent person in extreme cases, and it is thought to be
an advantage of his theory that it captures the sense that, in these cases, an important
principle is being overridden.

b. Objection and Response

This overriding process, however, cannot work in the opposite direction. In Hart’s theory,
some social good must be promoted or some social evil must be reduced in order for
punishment to be justified. Because of this, it is unjustifiable to punish a person who
seems to deserve punishment unless some utilitarian aim is being furthered. Imagine the
most despicable character you can think of, a mass-murderer perhaps. The justifiability of
punishing a person guilty of such crimes is beholden to the social consequences of the
punishment. That a depraved character would suffer for his wrongdoing is not enough. So,
for Hart, considerations of desert cannot override utilitarian considerations in this way.
Some theorists find this consequence of his theory unacceptable. Ten argues that, “it
would be unfair to punish an offender for a lesser offense and yet not punish another
offender for a more serious offense” (80). If we are behaving in accordance with Hart’s
theory, we may, on occasion, have to avoid punishing serious offenders while continuing to
punish less serious offenders for utilitarian reasons. Since doing so would be unfair, it
seems that Hart’s theory may be seriously flawed.

In order to assess Ten’s criticism, it is important to ask the following question: If we were to
avoid punishing the more serious offender, to whom would we be being unfair? In an effort
to answer this question, we must consider whether the offender who has committed the
lesser crime has grounds for complaint if the more serious offender is not punished. By
stipulation, the lesser offender committed the crime and cannot thereby claim a violation of
justice on those grounds. Is the justification of his punishment contingent upon the
punishment of others? Arguably not: The punishment of the lesser offender is justified
regardless of whoever else is punished. He may bemoan his bad luck and wish that his
punishment were not likely to further any utilitarian aims so that he may avoid it, but he
cannot rightly accuse society of a violation of justice for failing to punish others when he
does in fact deserve the punishment that is being inflicted upon him. The attractiveness of
Ten’s argument is derived from the fact that its conclusion fits with our intuitions regarding
the idea that some people just deserve to suffer no matter what. Perhaps we ought to
reexamine that intuition and consider that it may be rooted in an urge to revenge, not a
concern for justice.

4. Amount of Punishment
The belief that, in most cases, the amount of punishment should vary directly with the
seriousness of the offense is widely accepted. However, utilitarians and retributivists have
different ways of arriving at this general conclusion.

a. Utilitarians on Amount
Bentham, a utilitarian, states that, “The greater the mischief of the offence, the greater is
the expense, which it may be worth while to be at, in the way of punishment” (181). Crime
and punishment both tend to cause unhappiness. Recall that utilitarianism is solely
concerned with the balance of happiness over unhappiness produced by an action. When
attempting to determine the amount of punishment that ought to be permitted for a given
offense, it is necessary to weigh the unhappiness that would be caused by the offense
against the unhappiness caused by various punishments. The greater the unhappiness
caused by a given offense, the greater the amount of punishment that may be inflicted for
that offense in order to reduce its occurrence before the unhappiness caused by the
punishment outweighs the unhappiness caused by the offense (Ten, 143).
So, utilitarians would often be committed to abiding by the rule that the amount of
punishment should vary directly with the seriousness of the offense. However, it seems
that there are cases in which they would be committed to violating this rule. Critics argue
that utilitarians would sometimes be committed to inflicting a severe punishment for a
relatively minor offense. Ten asks us to imagine a society in which there are many petty
thefts and thieves are very difficult to catch. Since there are many thefts, the total amount
of unhappiness caused by them is great. Imagine that one thief is caught and the
authorities are deciding how severely to punish him. If these authorities were utilitarians,
they would be committed to giving him a very severe sentence, 10 years perhaps, if this
were the only way to deter a significant number of petty thieves. But surely making an
example of the one thief who was unlucky or unskilled enough to be caught is unjust.
Since utilitarians are sometimes committed to inflicting such harsh punishments for
relatively minor offenses, their approach must be inadequate (143-144).

b. Retributivists on Amount
Retributivists argue that more serious offenses should be punished more severely
because offenders who commit more serious crimes deserve harsher punishment than
those who commit less serious crimes. Given our previous discussion of retributivism, it
should not come as a surprise that the concept of desert plays a central role here.
According to many classic versions of retributivism, including Kant’s, the deserved
punishment is determined by invoking the lex talionis. The old adage, “An eye for an eye, a
tooth for a tooth,” is derived from the lex talionis, which “requires imposing a harm on a
criminal identical to the one he imposed on his victim” (Shafer-Landau, 773). Those who
argue that murderers ought to be put to death have often invoked this principle, but it is
rarely invoked when attempting to determine the proper punishment for other crimes. Its
lack of popularity can be explained by noting a couple of objections. First, it is difficult to
apply to many offenses, and it seems to be outright inapplicable to some. How should we
punish the counterfeiter, the hijacker, or the childless kidnapper? Applying the lex talionis
to these crimes is, at the very least, problematic. Second, there are many cases in which it
would require that we punish offenders by performing actions that ought not to be carried
out by any government (773). Surely we should not rape rapists! For these and other
reasons, except when the topic at hand is capital punishment, appeals to the lex talionis in
the contemporary literature are rare.

Many contemporary retributivists hold that the principle of proportionality should be used in
order to determine the amount of punishment to be meted out in particular cases. This
principle states that, “the amount of punishment should be proportionate to the moral
seriousness or moral gravity of offenses…” (Ten, 154). Different versions of the
proportionality principle call for different ways of establishing how severe a punishment
must be in order to meet the demands set by the principle. Must it merely be the case that
there be a direct relationship between the amount of punishment and the seriousness of
the offense, or must offenders suffer the same amount as their victim(s) in order for the
demands of the principle to be met? Retributivists are not in complete agreement on how
to answer this question.
While retributivists seem to have an easier time ensuring that there be a direct relationship
between the amount of punishment and the seriousness of the offense, their position is
subject to criticism. Because they are committed to inflicting the deserved punishment,
they must do so even when a lesser punishment would produce the same social effects.
Clearly, this criticism runs parallel to the objection to retributivism discussed in section 2: if
the retributivist is committed to inflicting the deserved punishment regardless of the social
effects, then it seems that he is committed to inflicting gratuitous pain on an offender. Of
course, some resist the idea that inflicting suffering in such a case would be gratuitous,
which is why this debate continues. In any case, the perceived shortcomings of both the
utilitarian and retributive approaches have led theorists to attempt to develop approaches
that combine elements of both. For reasons similar to those cited in support of the
aforementioned “compromise” theories, it seems that these approaches are the most
promising.

5. Capital Punishment
Capital punishment involves the deliberate killing of a supposed or actual offender for an
offense. Throughout history and across different societies, criminals have been executed
for a variety of offenses, but much of the literature is devoted to examining whether those
convicted of murder ought to be executed, and this discussion will be similarly focused.

A combination of utilitarian and retributive considerations are usually invoked in an effort to

justify the execution of murderers. The centerpiece of most arguments in favor of capital
punishment is retributive: Murderers deserve to be put to death. This is usually argued for
along Kantian lines: By deliberately causing an innocent person’s death, the murderer has
rendered himself deserving of death. Utilitarian considerations generally play a large role
as well. Proponents argue that the threat of capital punishment can deter potential
murderers. Since many human beings’ greatest fear is death, the intuitive plausibility of
this claim is clear. In addition, proponents point to the fact that capital punishment is the
ultimate incapacitation. Clearly, if a murderer is dead, then he can never harm anyone
again.

Opponents of capital punishment challenge proponents on each of these points. Albert

Camus denies that murder and capital punishment are equivalent to one another:

But what is capital punishment if not the most premeditated of murders, to

which no criminal act, no matter how calculated, can be compared? If there
were to be a real equivalence, the death penalty would have to be
pronounced upon a criminal who had forewarned his victim of the very
moment he would put him to a horrible death, and who, from that time on,
had kept him confined at his own discretion for a period of months. It is not in
private life that one meets such monsters (25).

This argument and others that resemble it are often put forth in an attempt to counter the
retributive argument. Also, any criminal justice system that executes convicted criminals
 runs the risk of executing some individuals who do not deserve to be executed: the
wrongfully convicted. Some argue that a fallible criminal justice system ought not to
impose a penalty that removes the possibility of mistakes being rectified. The utilitarian
arguments have also come under attack. Some argue that the proponents of capital
punishment have overstated its deterrent value, and it has been argued that it may even
incite some people to commit murder (Bedau, 198-200). Regarding incapacitation, it has
been argued that the danger involved in failing to execute murderers has been similarly
overstated (196-198).

6. Conclusion
These issues introducing punishment have received a great deal of attention in the
professional literature, and many philosophers continue to discuss them and offer various
answers to the questions that are raised. However, the issues raised here are not the only
ones. There are many, including the role of excuses and mitigating circumstances, the
usage of insanity as a defense, the imprisonment of offenders, and the cultural and
historical context of punishment.

7. References and Further Reading

 Beccaria, Cesare. On Crimes and Punishments. Trans. David Young. Indianapolis: Hackett
Publishing Company, 1986.
 Bedau, Hugo Adam. “Capital Punishment.” In Matters of Life and Death: New Introductory
Essays in Moral Philosophy. Ed. Tom Regan. New York: Random House, 1986. 175-212.
 Bedau, Hugo Adam, and Paul Cassell, eds. Debating the Death Penalty: Should America
Have Capital Punishment? The Experts on Both Sides Make Their Best Case. New York:
Oxford University Press, 2004.
 Bentham, Jeremy. The Principles of Morals and Legislation. New York: Hafner Publishing
Company, 1948.
 Camus, Albert. Reflections on the Guillotine. Trans. Richard Howard. Michigan City, IN:
Fridtjof-Karla Publications, 1959.
 Duff, R.A. “Penal Communications: Recent Work in the Philosophy of Punishment.” Crime
and Justice 20 (1996): 1-97.
 Duff, R.A., and David Garland, eds. A Reader on Punishment. New York: Oxford University
Press, 1994.
 Ezorsky, Gertrude. “The Ethics of Punishment.” In Philosophical Perspectives on
Punishment. Ed. Gertrude Ezorsky. Albany: State University of New York Press, 1972. xi-
xxvii.
 Foucault, Michel. Discipline and Punish: The Birth of the Prison. Trans. Alan Sheridan.
New York: Random House, 1977.
 Hart, H.L.A. “Prolegomenon to the Principles of Punishment.” In Punishment and
Responsibility: Essays in the Philosophy of Law. New York: Oxford University Press, 1968.
1-27.
 Kant, Immanuel. “Justice and Punishment.” Trans. W. Hastie. In Philosophical
Perspectives on Punishment. Ed. Gertrude Ezorsky. Albany: State University of New York
Press, 1972. 102-106.
 McCloskey, H.J. “A Non-Utilitarian Approach to Punishment.” In Philosophical Perspectives
on Punishment. Ed. Gertrude Ezorsky. Albany: State University of New York Press, 1972.
119-134.
 Mill, John Stuart. Utilitarianism. Indianapolis: Hackett Publishing Company, 1979. Shafer-
Landau, Russ. “The Failure of Retributivism.” In Philosophy of Law. Ed. Joel Feinberg and
Jules Coleman. Belmont, CA: Wadsworth/Thompson Learning, 2000. 769-779.
 Sprigge, T.L.S. “A Utilitarian Reply to Dr. McCloskey.” In Philosophical Perspectives on
Punishment. Ed. Gertrude Ezorsky. Albany: State University of New York Press, 1972. 66-
79.
 Ten, C.L. Crime, Guilt, and Punishment. Oxford: Clarendon Press, 1987.

Kevin Murtagh
Email: KevinJMurtagh@aol.com
John Jay College of Criminal Justice
U. S. A.
Affirmative Action
First published Fri Dec 28, 2001; substantive revision Fri Jun 21, 2024
“Affirmative action” means positive steps taken to increase the representation of women
and minorities in areas of employment, education, and culture from which they have been
historically excluded. When those steps involve preferential selection—selection on the
basis of race, gender, or ethnicity—affirmative action generates intense controversy.
The ebb and flow of public controversy over affirmative action can be pictured as three
spikes on a line, the first spike representing a period of passionate debate that began
around 1972 and tapered off after 1980, and the second indicating a resurgence of debate
in the 1990s leading up to Supreme Court’s decisions in 2003 and 2016 upholding certain
kinds of affirmative action in higher education. The third spike reflects the Supreme Court’s
decision in 2023 voiding race-conscious-programs at Harvard and the University of North
Carolina, potentially opening a new era of conflict.
The first spike encompassed controversy about gender and racial/ethnic preferences alike.
This is because in the beginning affirmative action was largely about the factory, the
firehouse, the office, and the classroom podium. The second and third spikes represent a
quarrel about race and ethnicity. This is because the debate at the turn of the twentieth-
first century focused on college admissions. At selective colleges, women need no boost;
African-Americans and Hispanics seemingly do.[1] The first part of this entry looks at
affirmative action’s origins in the employment context. The longer second part
concentrates on the college admissions controversy.
 1. In the Beginning
 2. The Controversy Engaged
 3. Rights and Consistency
 4. The Campus Wars
 5. Equality’s Rule
 6. Diversity’s Dominion
 7. Legitimacy
 8. Diversity’s Limit
 Cases
 Bibliography
 Academic Tools
 Other Internet Resources
 Related Entries

1. In the Beginning
In 1972, affirmative action became an inflammatory public issue. True enough, the Civil
Rights Act of 1964 made something called “affirmative action” a remedy federal courts
could impose on violators of the Act. Likewise, after 1965 federal contractors fell subject to
President Lyndon Johnson’s Executive Order 11246, requiring them to take “affirmative
action” to make sure they were not discriminating in their employment practices. But what
did this 1965 mandate amount to? The Executive Order assigned to the Secretary of Labor
the job of specifying rules of implementation. In the meantime, as the federal courts were
enforcing the Civil Rights Act against discriminating companies, unions, and other
institutions, the Department of Labor mounted ad hoc attacks on the construction industry
by cajoling, threatening, negotiating, and generally strong-arming reluctant construction
firms into a series of region-wide “plans” in which they committed themselves to numerical
hiring goals. Through these contractor commitments, the Department could indirectly
pressure recalcitrant labor unions, who supplied the employees at job sites, to
desegregate.
While the occasional court case and government initiative made the news and stirred
some controversy, affirmative action was pretty far down the list of public excitements until
the autumn of 1972, when the Secretary of Labor’s Revised Order No. 4, fully
implementing the Executive Order, landed on campus by way of directives from the
Department of Health, Education, and Welfare. The Revised Order’s predecessor, Order
No. 4, first promulgated in 1970, cast a wide net over American institutions, both public
and private. By extending to all contractors the basic apparatus of the construction industry
“plans,” the Order imposed a one-size-fits-all system of “underutilization analyses,” “goals,”
and “timetables” on hospitals, banks, trucking companies, steel mills, printers, airlines—
indeed, on all the scores of thousands of institutions, large and small, that did business
with the government, including a special set of institutions with a particularly voluble and
articulate constituency, namely, American universities.
At first, university administrators and faculty found the demands of Order No. 4 murky but
hardly a threat to the established order. The number of racial and ethnic minorities
receiving PhDs each year and thus eligible for faculty jobs was tiny. Any mandate to
increase their representation on campus would require more diligent searches by
universities, to be sure, but searches fated nevertheless largely to mirror past results. The
1972 Revised Order, on the other hand, effected a change that punctured any campus
complacency: it included women among the “protected classes” whose “underutilization”
demanded the setting of “goals” and “timetables” for “full utilization” (Graham 1990, 413).
Unlike African-Americans and Hispanics, women were getting PhDs in substantial and
growing numbers. If the affirmative action required of federal contractors was a recipe for
“proportional representation,” then Revised Order No. 4 was bound to leave a large
footprint on campus. Some among the professoriate exploded in a fury of opposition to the
new rules, while others responded with an equally vehement defense of them.[2]
As it happened, these events coincided with another development, namely the “public turn”
in philosophy. For several decades Anglo-American philosophy had treated moral and
political questions obliquely. On the prevailing view, philosophers were suited only to do
“conceptual analysis”—they could lay bare, for example, the conceptual architecture of the
idea of justice, but they were not competent to suggest political principles, constitutional
arrangements, or social policies that actually did justice. Philosophers might do “meta-
ethics” but not “normative ethics.” This view collapsed in the 1970s under the weight of two
counter-blows. First, John Rawls published in 1971 A Theory of Justice, an elaborate,
elegant, and inspiring defense of a normative theory of justice (Rawls 1971). Second, in
the same year Philosophy & Public Affairs, with Princeton University’s impeccable
pedigree, began life, a few months after Florida State’s Social Theory and Practice. These
journals, along with a re-tooled older periodical, Ethics, became self-conscious platforms
for socially and politically engaged philosophical writing, born out of the feeling that in time
of war (the Vietnam War) and social tumult (the Civil Rights Movement, Women’s
Liberation), philosophers ought to engage with real issues, not merely talk about words
and concepts. In 1973, Philosophy & Public Affairs published Thomas Nagel’s “Equal
Treatment and Compensatory Justice” (Nagel 1973) and Judith Jarvis Thomson’s
“Preferential Hiring” (Thomson 1973), and the philosophical literature on affirmative action
burgeoned forth.[3]
In contention was the nature of those “goals” and “timetables” imposed on every contractor
by Revised Order No. 4. Weren’t the “goals” tantamount to “quotas,” requiring institutions
to use racial or gender preferences in their selection processes? Some answered “no”
(Ezorsky 1977, 86). Properly understood, affirmative action did not require (or even permit)
the use of gender or racial preferences. Others said “yes” (Goldman 1976, 182–3).
Affirmative action, if it did not impose preferences outright, at least countenanced them.
Among the yea-sayers, opinion divided between those who said preferences were morally
permissible and those who said they were not. Within the “morally permissible” set,
different writers put forward different justifications.

2. The Controversy Engaged

The essays by Thomson and Nagel defended the use of preferences but on different
grounds. Thomson endorsed job preferences for women and minorities as a form of
redress for their past exclusion from the workplace. Preferential policies, in her view,
worked a kind of justice. Nagel, by contrast, argued that preferences might work a kind of
social good, and without doing violence to justice. Institutions could for one or another
compelling reason properly depart from standard meritocratic selection criteria because
the whole system of tying economic reward to earned credentials was itself indefensible.
Justice and desert lay at the heart of subsequent arguments. Several writers took to task
Thomson’s argument that preferential hiring justifiably makes up for past wrongs.
Preferential hiring seen as redress looks perverse, they contended, since it benefits
individuals (minorities and women possessing good educational credentials) least likely
harmed by past wrongs while it burdens individuals (younger white male applicants) least
likely to be responsible for past wrongs (Simon 1974, 315–19; Sher 1975, 162; Sher 1979,
81–82; and Goldman 1976, 190–1).[4] Instead of doing justice, contended the critics,
preferential treatment violated rights. What rights were at issue? The right of an applicant
“to equal consideration” (Thomson 1973, 377; Simon 1974, 312), the right of the maximally
competent to an open position (Goldman 1976, 191; Goldman 1979, 24–8), or the right of
everyone to equal opportunity (Gross 1977a, 382; Gross 1978, 97). Moreover, according
to the critics, preferential treatment confounded desert by severing reward from a
“person’s character, talents, choices and abilities” (Simon 1979, 96), by “subordinating
merit, conduct, and character to race” (Eastland and Bennett 1979, 144), and by
disconnecting outcomes from actual liability and damage (Gross 1978, 125–42).
Defenders of preferences were no less quick to enlist justice and desert in their cause.
Mary Anne Warren, for example, argued that in a context of entrenched gender
discrimination, gender preferences might improve the “overall fairness” of job selections.
Justice and individual desert need not be violated.
 If individual men’s careers are temporarily set back because of…[job
preferences given to women], the odds are good that these same men will have
benefited in the past and/or will benefit in the future—not necessarily in the job
competition, but in some ways—from sexist discrimination against women.
Conversely, if individual women receive apparently unearned bonuses [through
preferential selection], it is highly likely that these same women will have
suffered in the past and/or will suffer in the future from…sexist attitudes.
(Warren 1977, 256)

Likewise, James Rachels defended racial preferences as devices to neutralize unearned

advantages by whites. Given the pervasiveness of racial discrimination, it is likely, he
argued, that the superior credentials offered by white applicants do not reflect their greater
effort, desert, or even ability. Rather, the credentials reflect their mere luck at being born
white. “Some white…[applicants] have better qualifications…only because they have not
had to contend with the obstacles faced by their African-American competitors” (Rachels
1978, 162). Rachels was less confident than Warren that preferences worked uniformly
accurate offsets. Preferences might do injustice to some whites; yet their absence would
result in injustices to African-Americans who have been unfairly handicapped by their
lesser advantages.
Rachels’ diffidence was warranted in light of the counter-responses. If racial and gender
preferences for jobs were supposed to neutralize unfair competitive advantages, they
needed to be calibrated to fit the variety of backgrounds aspirants brought to any
competition for these goods. Simply giving blanket preferences to blacks or women
seemed much too ham-handed an approach if the point was to micro-distribute
opportunities fairly (Sher 1975, 165ff).

3. Rights and Consistency

To many of its critics, “reverse discrimination” was simply incoherent. When “the employers
and the schools favor women and blacks,” objected Lisa Newton, they commit the same
injustice perpetrated by Jim Crow discrimination. “Just as the previous discrimination did,
this reverse discrimination violates the public equality which defines citizenship” (Newton
1973, 310).[5]
William Bennett and Terry Eastland likewise saw racial preferences as in some sense
illogical:
To count by race, to use the means of numerical equality to achieve the end of
moral equality, is counterproductive, for to count by race is to deny the end by
virtue of the means. The means of race counting will not, cannot, issue in an
end where race does not matter (Eastland and Bennett 1979, 149).[6]

When Eastland and Bennett alluded to those who favored using race to get to a point
where race doesn’t count, they had in mind specifically the Supreme Court’s Justice
Blackmun who, in the famous 1978 Bakke case (discussed below), put his own view in just
those simple terms. For Blackmun, the legitimacy of racial preferences was to be
measured by how fast using them moves us toward a society where race doesn’t matter (a
view developed in subtle detail by Richard Wasserstrom in Wasserstrom 1976). While the
critics of preferences claimed to find the very idea of using race to end racism illogical and
incoherent, they also fell back on principle to block Blackmun’s instrumental defense
should it actually prove both reasonable and plausible. “The moral issue comes in classic
form,” wrote Carl Cohen. “Terribly important objectives…appear to require impermissible
means.” He went on to ask, “might we not wink at the Constitution this once” and allow
preferences to do their good work (Cohen 1995, 20)? Neither he nor other critics thought
so. Principle must hold firm. “[I]n the distribution of benefits under the laws all racial
classifications are invidious” (Cohen 1995, 52).
But what, exactly, is the principle—constitutional or moral—that always bars the use of
race as a means to “terribly important objectives”? Alan Goldman did more than anyone in
the early debate to formulate and ground a relevant principle. Using a contractualist
framework, he surmised that rational contractors would choose a rule of justice requiring
positions to be awarded by competence. On its face, this rule would seem to preclude
filling positions by reference to factors like race and gender that are unrelated to
competence. However, Goldman’s “rule” blocked preferences only under certain empirical
conditions. Goldman explained the derivation of the rule and its consequent limit this way:
The rule for hiring the most competent was justified as part of a right to equal
opportunity to succeed through socially productive effort, and on grounds of
increased welfare for all members of society. Since it is justified in relation to a
right to equal opportunity, and since the application of the rule may simply
compound injustices when opportunities are unequal elsewhere in the system,
the creation of more equal opportunities takes precedence when in conflict with
the rule for awarding positions. Thus short-run violations of the rule are justified
to create a more just distribution of benefits by applying the rule itself in future
years (Goldman 1979, 164–165).

In other words, if “terribly important objectives” having to do with equalizing opportunities in

a system rife with unjust inequalities could in fact be furthered by measured and targeted
“reverse discrimination,” justice wouldn’t stand in the way. Goldman’s principle did not
have the adamantine character Cohen and other critics sought in a bar to preferences.
Where can such an unyielding principle be found? In the next two sections I examine
different efforts to answer this question.

4. The Campus Wars

In the 1970s, while campuses were embroiled in debate about how to increase blacks,
Hispanics, and women on the faculty, universities were also putting into effect schemes to
increase minority presence within the student body. This proved difficult for institutions
whose admissions criteria demanded a high level of demonstrated academic performance
from applicants. The demand was reasonable: applicants chosen had to be prepared to
succeed in a rigorous academic setting. Yet the demand effectively excluded minority
students. Only a handful of high school graduates from under-represented and
disadvantaged minorities possessed test scores and grades good enough to make them
eligible for admission. Even so, some of these would have succeeded had they been
admitted. Nor was this exclusion likely to remedy itself without some sort of intervention,
and selective colleges and universities increasingly felt the public interest was ill-served by
virtually all-white campuses. What, exactly, was the public interest in question? As we shall
see, over the decades higher education gave different answers to this question. As it was,
very selective institutions faced a choice: retain their admissions criteria unchanged and
live with the upshot—hardly any students from underrepresented and disadvantaged
groups on campus—or reconceive their criteria to get a more substantial minority
representation. Most selected the second path.
The Medical School of the University of California at Davis exemplified a particularly
aggressive approach. It reserved sixteen of the one hundred slots in its entering classes
for minorities. In 1973 and again in 1974, Allan Bakke, a white applicant, was denied
admission although his test scores and grades were better than most or all of those
admitted through the special program. He sued. In 1977, his case, Regents of the
University of California v. Bakke, reached the Supreme Court. The Court rendered its
decision a year later (Bakke 1978).
An attentive reader of Title VI of the Civil Rights Act—“[N]o person…shall, on the ground of
race, color, or national origin, be excluded from participation in, be denied the benefits of,
or be subjected to discrimination under any program or activity receiving Federal financial
assistance”—might have thought this case was an easy call. So, too, thought four Justices
on the Court, who voted to order Bakke admitted to the Medical School. Led by Justice
Stevens, they saw the racially segregated, two-track scheme at the Medical School (a
recipient of federal funds) as a clear violation of the plain language of the Title.
Four other members of the Court, led by Justice Brennan, wanted very keenly to save the
Medical School program. To find a more attractive terrain for doing battle, they made an
end-run around Title VI, arguing that, whatever its language, it had no independent
meaning itself. It meant in regard to race only what the Constitution meant. Thus, instead
of having to parse the stingy and unyielding language of Title VI (“no person shall be
subjected to…on the ground of race”), the Brennan group could turn their creative
energies to interpreting the broad and vague language of the Fourteenth Amendment (“no
person shall be denied the equal protection of the laws”), which provided much more
wiggle-room for justifying racial preferences. The Brennan Justices persuaded one other
member, Justice Powell, to join them in their view of Title VI. But Powell didn’t agree with
their view of the Constitution. He argued that the Medical School’s policy was
unconstitutional and voted that Bakke must be admitted. His vote, added to the four votes
of the Stevens group, meant that Allan Bakke won his case and that Powell got to write the
opinion of the Court. The Brennan strategy didn’t reap the fruit it intended.
Against the leanings of the Brennan group, who would distinguish between “benign” and
“malign” uses of race and deal more leniently with the former, Powell insisted that the
Fourteenth Amendment’s promise of “equal protection of the law” must mean the same
thing for all, black and white alike. To paraphrase Powell:
The Constitution can tolerate no “two-class” theory of equal protection. There is
no principled basis for deciding between classes that deserve special judicial
attention and those that don’t. To think otherwise would involve the Court in
making all kinds of “political” decisions it is not competent to make. In
expounding the Constitution, the Court’s role is to discern “principles sufficiently
absolute to give them roots throughout the community and continuity over
significant periods of time, and to lift them above the pragmatic political
judgments of a particular time and place” (Bakke 1978, 295–300 [Powell
quoting Cox 1976, 114]).
What, then, was the practical meaning of a “sufficiently absolute” rendering of the principle
of equal protection? It was this: when the decisions of state agents “touch upon an
individual’s race or ethnic background, he is entitled to a judicial determination that the
burden he is asked to bear on that basis is precisely tailored to serve a compelling
governmental interest” (Bakke 1978, 300).
Powell, with this standard in hand, then turned to look at the four reasons the Medical
School offered for its special program: (i) to reduce “the historic deficit of traditionally
disfavored minorities in medical schools and the medical profession;” (ii) to counter “the
effects of societal discrimination;” (iii) to increase “the number of physicians who will
practice in communities currently underserved;” and (iv) to obtain “the educational benefits
that flow from an ethnically diverse student body” (Bakke 1978, 307). Did any or all of them
specify a compelling governmental interest? Did they necessitate use of racial
preferences?
As to the first reason, Powell dismissed it out of hand.
If [the School’s] purpose is to assure within its student body some specified
percentage of a particular group merely because of its race or ethnic origin,
such a preferential purpose must be rejected not as insubstantial but as facially
invalid. Preferring members of any one group for no reason other than race or
ethnic origin is discrimination for its own sake.

As to the second reason, Powell allowed it more force. A state has a legitimate interest in
ameliorating the effects of past discrimination. Even so, contended Powell, the Court,
has never approved a classification that aids persons perceived as members of
relatively victimized groups at the expense of other innocent individuals in the
absence of judicial, legislative, or administrative findings of constitutional or
statutory violations (Bakke 1978, 308).

And the Medical School does not purport to have made, and is in no position to
make, such findings. Its broad mission is education, not the formulation of any
legislative policy or the adjudication of particular claims of illegality.…[I]solated
segments of our vast governmental structures are not competent to make those
decisions, at least in the absence of legislative mandates and legislatively
determined criteria (Bakke 1978, 309).

As to the third reason, Powell found it, too, insufficient. The Medical School provided no
evidence that the best way it could contribute to increased medical services to
underserved communities was to employ a racially preferential admissions scheme.
Indeed, the Medical School provided no evidence that its scheme would result in any
benefits at all to such communities (Bakke 1978, 311).
This left the fourth reason. Here Powell found merit. A university’s interest in a diverse
student body is legitimated by the First Amendment’s implied protection of academic
freedom. This constitutional halo makes the interest “compelling.” However, the Medical
School’s use of a racial and ethnic classification scheme was not “precisely tailored” to
effect the School’s interest in diversity, argued Powell.
The diversity that furthers a compelling state interest encompasses a far
broader array of qualifications and characteristics of which racial or ethnic origin
 is but a single though important element. [The Medical School’s] special
admissions program, focused solely on ethnic diversity, would hinder rather
than further attainment of genuine diversity (Bakke1978, 316).

The diversity which provides an educational atmosphere “conducive to speculation,

experiment and creation” includes a nearly endless range of experiences, talents, and
attributes that students might bring to campus. In reducing diversity to racial and ethnic
quotas, the Medical School wholly misconceived this important educational interest.
In sum, although the last of the Medical School’s four reasons encompassed a “compelling
governmental interest,” the School’s special admissions program was not necessary to
effect the interest. The special admissions program was unconstitutional. So concluded
Justice Powell.

5. Equality’s Rule
This was a conclusion Justice Brennan tried vigorously to forestall. Brennan agreed with
Powell that “equal protection” must mean the same thing—that is, remain one rule—
whether applied to blacks or whites. But the same rule applied to different circumstances
need not yield the same results. Racial preferences created for different reasons and
producing different outcomes need not all be judged in the same harsh, virtually fatal,
manner. This point was the crux of Brennan’s defense of the Medical School’s policy.
Powell thought there was no principled way to distinguish “benign” from “malign”
discrimination, but Brennan insisted there was. He argued that if the Court looked carefully
at its past cases striking down Jim Crow laws, it would see the principle at work. What the
Court found wrong in Jim Crow was that it served no purpose except to mark out and
stigmatize one group of people as inferior. The “cardinal principle” operating in the Court’s
decisions condemned racial classifications “drawn on the presumption that one race is
inferior to another” or that “put the weight of government behind racial hatred and
separation” (Bakke,1978, 358 [Brennan, dissenting]). Brennan agreed with Powell that no
public racial classification motivated by racial animus, no classification whose purpose is to
stigmatize people with the “badge of inferiority,” could withstand judicial scrutiny. However,
the Medical School’s policy, even if ill-advised or mistaken, reflected a public purpose far
different from that found in Jim Crow. The policy ought not be treated as though it were cut
from the same cloth.
Brennan granted that if a state adopted a racial classification for the purpose of humiliating
whites, or stigmatizing Allan Bakke as inferior and confining him to second-class
citizenship, that classification would be as odious as Jim Crow. But the Medical School’s
policy had neither this purpose nor this effect. Allan Bakke may have been upset and
resentful at losing out under the special plan, but he wasn’t “in any sense stamped as an
inferior by the Medical School’s rejection of him.” Nor did his loss constitute a “pervasive
injury,” in the sense that wherever he went he would be treated as a “second-class citizen”
because of his color (Bakke 1978, 376 [Brennan, dissenting]).
In short, argued Brennan, the principle expressed in the Equal Protection Clause should
be viewed as an anti-caste principle, a principle that uniformly and consistently
rejects all public law whose purpose is to subject people to an inferior and degraded
station in life, whether they are black or white.[7] Of course, given the asymmetrical
position of whites and blacks in our country, we are not likely to encounter laws that try to
stigmatize whites as an inferior caste (much less succeed at it). But this merely shows that
a principle applied to different circumstances produces different results. Because the
Medical School’s program sought to undo the effects of a racial caste system long-
enduring in America, it represented a purpose of great social importance and should not
be found constitutionally infirm: so maintained Brennan (Bakke 1978, 363 [Brennan,
dissenting]).
Justice Powell never successfully engaged this way of reading “constitutional equality.” His
insistence on clear, plain, unitary, absolute principle did not cut against the Brennan view.
The issue between Powell and Brennan was not the consistency and stringency of the
principle but its content. If the Constitution says, “The state cannot deliberately burden
someone by race if its purpose is to create or maintain caste,” then constitutional law
doesn’t block any of the Medical School’s justifications.
If we turn away from exegesis of the Constitution, are we likely to find in political theory
itself any principle of equality implying that every use of racial preferences
in every circumstance works an intolerable injustice?
The prospects seem dim. Won’t very general notions of equality lend themselves both to
defending affirmative action (and the social inclusion it effects) as well as condemning it
(and the racial non–neutrality it involves)? The challenge here is well-illustrated by Carl
Cohen’s effort, in a debate with James Sterba, to extract a strict prohibition on racial
preferences from the Aristotelian principle that “equals should be treated equally” (Cohen
and Sterba 2003, 23). This principle, urges Cohen, “certainly entails at least this: It is
wrong, always and everywhere, to give special advantage to any group simply on the
basis of physical characteristics that have no relevance to the award given or the burden
imposed” (Cohen and Sterba 2003, 25). Whether anything interesting follows from this
proposition depends on how we construe “relevance.” Cohen admits that public policy may
rightly treat some people differently because of their physical characteristics. For example,
the state might offer special assistance to the old or disabled. Now, this example lends
itself to the thought that the relevance of physical differences is something independent of
social policy. Age and disability, it seems, are real features of persons and public policy
simply tracks them. However, the relevance that differences have is not something itself
given by nature; it is determined by public purposes. Age and disability are made relevant
in this manner—in the one case, by the social purpose of assuring that people do not have
to live in poverty when they can no longer work; in the other case, by the social purpose of
assuring that people are not foreclosed from developing and marketing their talents by
impediments in the (largely constructed) physical environment.
Purpose determines relevancy, and this is true whether or not the relevant differences are
physical. If the nation thinks it desirable to change white institutions so that they are less
uniformly white, that purpose links skin color to recruitment.
Because the Aristotelian principle by itself doesn’t rule out racial preferences (since blacks
and whites may be relevantly different with respect to certain legitimate public purposes), it
is not surprising that Cohen also invokes a substantive conception of equality: “All
members of humankind are equally ends in themselves, all have equal dignity—and
therefore all are entitled to equal respect from the community and its laws” (Cohen and
Sterba, 24). This principle, however, brings us back to the interpretive questions about
“equal protection of the laws” played out in the Powell-Brennan exchange in Bakke. What
do “equal dignity” and “equal respect” mean?
In a series of essays beginning in the 1970s, Ronald Dworkin marked out and developed a
view of equality that generalized elements of the Brennan argument in Bakke. He
emphasized a distinction between “two different sorts of rights…the right to equal
treatment…[and] the right to treatment as an equal.” The latter “is the right… to be treated
with the same respect and concern as anyone else.”
In some cases (voting, for example) equal treatment applies: government must treat every
eligible voter the same. In other contexts, however, government distinguishes among
groups and individuals in making policy for the common good (marking differences
between the young and old in the example noted above). In this classifying process, “[a]n
individual’s right to be treated as an equal means that his potential loss must be treated as
a matter of concern, but that loss may nevertheless be outweighed by the gain to the
community” (Dworkin 1977, 227). Treating the individual’s loss “as a matter of concern”
means that the ingredients which go into making public policy cannot include public
contempt or hatred toward the individual (and the group of which he/she is a member).
Every citizen has a constitutional right that he not suffer disadvantage, at least
in the competition for any public benefit, because the race or religion or region
or other artificial group to which he belongs is the subject of prejudice or
contempt (Dworkin 1985, 300–301).

The equal protection clause is violated, not whenever some group has lost an
important decision on the merits of the case or through politics, but when its
loss results from its special vulnerability to prejudice or hostility or stereotype
and its consequent diminished standing…in the political community. The clause
does not guarantee each citizen that he will benefit equally from every political
decision; it guarantees him only that he will be treated as an equal—with equal
concern and respect—in the political processes and deliberations that produce
those decisions (Dworkin 2000, 411).

It is true that the equal protection clause lays down a general principle of
political morality, and that its contemporary interpreters must remain faithful to
that general principle. If racial classifications were inherently morally wrong,
then they might well be deemed unconstitutional for that reason. But…racial
classifications are not inherently wrong, any more than other classifications
based on physical or genetically grounded properties (Dworkin 2000, 417;
footnote omitted).

Carl Cohen disagreed (Cohen 1995, 47). But he never specified a conception of equality in
which bearing unequal burdens on behalf of urgent social ends invariably amounts to an
assault on dignity if the burden happens to be assigned by race. He insisted (like Lisa
Newton above) that “[u]sing race to award benefits now does injustice in precisely the
same way injustice was done originally [to blacks], by giving moral weight to skin color
itself.” But comparing affirmative action to Jim Crow can’t be right, unless the purpose and
effects of a policy have no bearing on its moral standing.[8]
6. Diversity’s Dominion
How, if it held the Medical School’s policy unconstitutional, did Justice
Powell’s Bakke opinion become the basis upon which universities across the land enacted
—or maintained—racially preferential admissions policies?
If Powell had concluded with his assessment of the Medical School’s four
justifications, Bakke would have left university affirmative action in a precarious situation.
However, he didn’t stop there. In an earlier ruling on Bakke’s lawsuit, the California
Supreme Court had forbidden the Medical School to make any use of race or ethnicity in
its admissions decisions. Powell thought this went too far. Given higher education’s
protected interest in “diversity,” and given that a student’s race or ethnicity might add to
diversity just in the same way that her age, work experience, family background, special
talents, foreign language fluency, athletic prowess, military service, and unusual
accomplishments might, Powell vacated that portion of the California Supreme Court’s
order.
Then he added some dicta for guidance. If universities want to understand diversity and
the role that race and ethnicity might play in achieving it, they should look to Harvard,
proposed Powell, and he appended to his opinion a long statement of Harvard’s affirmative
action admissions program. In such a program, Powell contended, racial or ethnic
background might
be deemed a “plus” in a particular applicant’s file, yet it does not insulate the
individual from comparison with all other candidates for the available seats.…
This kind of program treats each applicant as an individual in the admissions
process. The applicant who loses out on the last available seat to another
candidate receiving a “plus” on the basis of ethnic background will not have
been foreclosed from all consideration for that seat simply because he was not
the right color or had the wrong surname. It would mean only that his combined
qualifications…did not outweigh those of the other applicant. His qualifications
would have been weighed fairly and competitively, and he would have had no
basis to complain of unequal treatment under the Fourteenth Amendment
(Bakke, at 318, 319).

In these off-hand comments, universities saw a green light for pushing ahead aggressively
with their affirmative action programs. Justice Powell’s basic holding could not have been
plainer: any system like the Medical School’s that assessed applications along two
different tracks defined by race or that used numerical racial quotas must fail constitutional
muster. Yet by the mid-1980s universities across the land had in place systems of
admissions and scholarship programs that exhibited one or both of these features. When
the University of Maryland’s Banneker scholarships—awarded only to African-American
students—were held in violation of the Constitution in 1994, the house of cards forming
university affirmative action began to fall (Podberesky v. Kirwan 1994). In 1996, the Court
of Appeals for the Fifth Circuit struck down the University of Texas Law School’s two-track
admissions program (Hopwood v. Texas 1996). In 1998, the Court of Appeals for the First
Circuit struck down a Boston plan assigning students to selective high schools by race
(Wessman v. Gittens 1998). In 2001, two more schools saw their admissions programs
invalidated by federal courts: the University of Georgia (Johnson v. Board of
Regents 2001) and the University of Michigan Law School (Grutter v. Bollinger 2001). In
many of these cases, educational institutions were using schemes that made race
something very much more than Justice Powell’s “plus” factor.[9] The Fifth Circuit Court’s
ruling in the University of Texas case (Hopwood v. Texas 1996) threw a cloud even over
Powell’s small window for affirmative action, boldly asserting that the Bakke holding was
now dead as law and that race could not be used at all in admissions.
Given Justice Powell’s singular opinion, supported by no one else on the Court, and given
the drift of Supreme Court decisions on racial preferences since 1978,
[10] the Hopwood court was not outlandish, if a bit presumptuous, in declaring Powell’s
holding in Bakke dead. As it happened, Powell’s opinion was far from dead. In the
University of Michigan Law School case, Grutter v. Bollinger, eventually decided by the
Supreme Court in June 2003, Justice Sandra Day O’Connor’s lead opinion declared:
“today we endorse Justice Powell’s view that student body diversity is a compelling state
interest that can justify the use of race in university admissions” (Grutter 2003, 330).
Diversity was alive after all. But how it worked its affirmative action elixir remained as
unclear in 2003 as it had been in 1978.
To see why, consider how in Grutter Justice O’Connor posed the issue:
The [Law School’s] policy aspires to “achieve that diversity which has the
potential to enrich everyone’s education and thus make a law class stronger
than the sum of its parts.”…The policy does not restrict the types of diversity
contributions eligible for substantial weight in the admissions process, but
instead recognizes “many possible bases for diversity admissions.”…The policy
does, however, reaffirm the Law School’s longstanding commitment to “one
particular type of diversity,” that is, “racial and ethnic diversity with special
reference to the inclusion of students from groups which have been historically
discriminated against” (Grutter 2003, 325).

Now, posing the issue this way and allowing the Law School to assert a special interest in
“one particular type of diversity” invites the conflation of general diversity—a diversity of
opinions, experiences, backgrounds, talents, aspirations, and perspectives—
with ethnic and racial diversity that Justice Powell appeared strongly to resist.
Nevertheless, Justice O’Connor’s went on:
We find that the Law School’s admissions program bears the hallmark of a
narrowly tailored plan. As Justice Powell made clear in Bakke, truly
individualized consideration demands that race be used in a flexible, non-
mechanical way. It follows from this mandate that universities cannot establish
quotas for members of certain racial groups (Grutter 2003, 334).

What vindicated the Law School in O’Connor’s eyes was its “highly individualized, holistic
review of each applicant’s file, giving serious consideration to all the ways an applicant
might contribute to a diverse educational environment” (Grutter 2003, at 339). This
“individualized consideration” is crucial; in Gratz v. Bollinger, decided the same day
as Grutter, Justice O’Connor switched sides to hold unconstitutional the undergraduate
admissions process at the University of Michigan. The undergraduate admissions office
operated differently than the Law School. It computed an index score for each applicant by
assigning numerical points for academic factors such as high school grades, admissions
test scores, quality of high school, strength of curriculum; and for nonacademic factors
such as being a resident of Michigan, a child of an alumnus, a recruited athlete, or a
member of “an underrepresented minority group.” An applicant falling in this last category
automatically received 20 points (Gratz 2003, 287). In O’Connor’s view, this “mechanical”
procedure meant that the undergraduate admissions office did not fully take account in
each application “of all factors that may contribute to student body diversity” (Gratz 2003,
288).
But O’Connor’s conclusion here leaves a question hanging. Why should the
undergraduate admissions office take account of all the factors that may contribute to
student body if it especially wants to select from certain parts of the diversity spectrum?
Why can’t it, like the law school, claim a special interest in “one particular type of
diversity”?
Moreover, why bar the undergraduate admissions office from using an effective tool to
promote its interest even if the tool is “mechanical”? In fact, the Law School’s “non-
mechanical” procedure differed from the undergraduate admissions policy only on its face,
not in its results. During admissions season, the Law School’s director of admissions
frequently consulted the “daily reports” that “kept track of the racial and ethnic
composition” of the incoming class. He did so to make sure a “critical mass” of minority
students was included (Grutter 2003, 326). In short, the Law School “managed” its
admissions process so that roughly 6 to 7 percent of each entering class was African-
American. The undergraduate admissions procedure, with its index scores, yielded a
similar outcome (Grutter 2003, 367–69 [Rehnquist, dissenting] and 374 [Kennedy,
dissenting]). Only surface appearance distinguished the two procedures. Justice Scalia
called the Law School’s “holistic” admissions process “a sham,” and not without some
reason (Grutter 2003, 375 [Scalia, dissenting]).
As it turned out, Grutter failed to close the book on university affirmative action. A new
legal challenge soon arose, this time against the University of Texas, which had revised its
own admissions program in 2004 to emulate the scheme validated in Grutter. The
case, Fisher v. Texas, wound its way through the courts for a decade, twice landing on the
steps of the Supreme Court before final disposition in 2016.
Justice Kennedy, writing for the Court, left the Grutter defense of racial preferences intact,
essentially retracing Justice O’Connor’s opinion. Universities bent upon pursuing the
“educational benefits that flow from student body diversity,” he wrote, are due a degree of
judicial deference (Fisher 2016, 2419).
The University explains that it strives to provide an ‘academic environment’ that
offers a ‘robust exchange of ideas, exposure to differing cultures, preparation
for the challenges of an increasingly diverse workforce, and acquisition of
competencies required of future leaders.’ … All of these objectives … mirror the
‘compelling interest’ this Court has approved in its prior cases (Fisher 2016,
2211).

Persuaded that race–neutral policies didn’t allow the University fully to succeed in deriving
the “educational benefits” of diversity, the Court majority found the University’s modest use
of race permissible.
But despite these legal victories, have universities actually made a case for diversity as the
justifying basis for race–conscious admissions? Caught napping in the mid–1990s when
the legal challenges began, higher education rushed to put meat on the Bakke bones and
turn Justice Powell’s off–hand remarks into a full–fledged defense. As the University of
Michigan cases approached a final test in 2003, the Supreme Court was bombarded with
scores of friend–of–the–court briefs from business groups, military officers, higher
education associations, coteries of scholars, and other interested parties lauding the
benefits of diversity. A similar outpouring preceded the decision in Fisher.
Consider some of the claims in these briefs. “[S]tudent body diversity is essential” if
universities are to provide students with “skills necessary for…success in an increasingly
globalized world” (Leading Public Research Universities 2015, 11–12; emphasis added).
Racial and ethnic diversity on campus are “vital” to securing a capable workforce; “it
is essential that [students] be educated in an environment where they are exposed to
diverse people, ideas, perspectives, and interactions” (65 Leading American Businesses
2003, 1, 2; emphasis added). Otherwise, the education of students is “degraded” (823
Social Scientists 2015, 5; emphasis added). “[S]tudents today must receive direct
experience with people of different backgrounds” (American Council on Education 2015, 6;
emphasis added). The “only means of obtaining properly qualified employees is through
diversity at institutions of higher education…[Appropriate] skills can only be developed
through exposure to widely diverse people, culture, ideas, and viewpoints” (Fortune 100
and Other Leading Businesses 2013, 3, 5; emphasis added). “Diversity is
an indispensable prerequisite to establishing the most productive problem–solving groups”
(Social and Organizational Psychologists 2015, 47; emphasis added).
According to the briefs the positive effects of diversity are bountiful. They include
“improvements in cognitive abilities, analytical problem solving skills and complex thinking
skills…[C]ross racial interactions are more strongly linked with cognitive growth than are
interactions with non–racial diversity” (American Educational Research Association 2012,
12). “[R]esearch shows that by increasing diversity, universities can help their graduates
enter society with better problem–solving capabilities than students who are not exposed
to diversity” (Social and Organizational Psychologists 2015, 37).
Now consider four points. First, the promiscuous use of “diversity” in the argument for
affirmative action opens the door to waffling and equivocation. Students learn through
“exposure to widely diverse people, culture, ideas, and viewpoints” announces the Fortune
100 brief. Of course. But the issue at hand is racial diversity. Wrapping the latter into the
former is not an aid to precision.[11] Second, can we imagine that the University of
Michigan or the University of Texas would abandon its affirmative action program if studies
showed no particular educational benefit to diversity? Suppose it turns out that students in
general don’t show additional “cognitive growth” from increased racial diversity. Suppose it
turns out students would best prepare for our “increasingly globalized world” by learning
Mandarin Chinese or Spanish.
Third, the straining by academics to show that cross-racial interaction
is essential, indispensable, vital, necessary, or imperative to a good education, if taken at
face value, leads to an unpleasant conclusion, namely that a lot of black college students
suffer a deficient education. The young woman who excels at Dunbar High School in
Washington, DC (enrollment 95 % black, 1 % white) and then gets her Bachelor of Science
degree in statistics magna cum laude at Spelman College (enrolling, in 2017, one white
among its 2097 students) is, according to Fortune 100 companies, not the sort of
employee they want. Graduates of Fisk University (0.7% white) or Tougaloo (0.6% white)
or Florida A & M (3.5% white) haven’t achieved enough cognitive growth, haven’t
sufficiently honed their problem–solving skills—is that what we are supposed to conclude?
The University of Texas (enrolling 25% Hispanics and 5% blacks) contends that without a
“critical mass” of minority students it can’t reap “the educational benefits of diversity.” Does
this mean that Prairie View University, two hours down the road from Austin and enrolling
7% Hispanics and 1.8% whites, has to do without those benefits? Are its students poorly
prepared for “our increasingly diverse workforce and society” (Fisher 2016, 2210, 2211)?
Defenders of affirmative action should think twice about claiming an education in a “non–
diverse” setting must be degraded.
Fourth, there is a noticeable lack of “fit” between university affirmative action practices and
the diversity rationale (see, for example, Anderson 2002, 27; Anderson 2004, 1217, 1221;
Anderson 2010, 142–143). Race is not treated just like any other special factor that might
warrant an admission “plus.” Recall that “special interest” asserted by the Michigan Law
School. Universities work diligently to maintain a relatively constant percentage of black
and Hispanic enrollees but give variable attention to other of the myriad qualities students
can bring to campus. As one proponent of affirmative action notes, “affirmative action
programs…do not look or operate as if they were testing for ‘diversity’.” He suggests that
many supporters of affirmative action nevertheless disregard this fact, believing “that there
is no harm in [a] miscast reliance on diversity because, with a wink and a nod, everyone
understands that diversity is really a proxy for integration” (Issacharoff 2002, 30, 34;
emphasis added; more generally see Jones 2005, Bell, 2003, Lippert-Rasmussen 2020,
124–143; Sabbagh 2007, 31–48).

7. Legitimacy
In the midst of her recitation in Grutter of the educational values of diversity, Justice
O’Connor introduced a new argument, one that didn’t locate the value of diversity in
students’ “cognitive growth” or in their future skills in negotiating a “diverse world,” but in its
contribution to the health of major American institutions.
Leading up to oral arguments in the case, among the amici briefs the Court received was
one by retired Army General Julius Becton and twenty-eight former Defense Department
officials and retired high-ranking military officers. It explained the crucial role racial
preferences play at the service academies and ROTC programs in maintaining an
integrated officer corps (Consolidated Brief 2003, 1–30).[12] This brief obviously had a
considerable impact on the Justices, several of whom during oral argument pressed the
lawyers representing the parties to address its claims (Grutter 2003 Oral Argument, 1–23).
The brief prompted Justice O’Connor to include its concerns in her opinion. She noted that
high-ranking retired officers and civilian leaders of the United States military
assert that, “based on [their] decades of experience,” a “highly qualified, racially
diverse officer corps…is essential to the military’s ability to fulfill its principal
mission to provide national security….” At present, “the military cannot achieve
an officer corps that is both highly qualified and racially diverse unless the
 service academies and ROTC used limited race-conscious recruiting and
admissions policies”…To fulfill its mission, the military “must be selective in
admissions for training and education for the officer corps, and it must train and
educate a highly qualified, racially diverse officer corps in a racially diverse
setting” (Grutter 2003, 331, citations omitted).

And then she quickly expanded the scope of her observation:

We agree that “it requires only a small step from this analysis to conclude that
our country’s other most selective institutions must remain both diverse and
selective”…The United States as amicus curiae, affirms that “ensuring that
public institutions are open and available to all segments of American society,
including people of all races and ethnicities, represents a paramount
government objective” (Grutter 2003, 331, citations omitted).

Moreover, universities, and in particular, law schools, represent the training

ground for a large number of our Nation’s leaders…Individuals with law degrees
occupy roughly half the state governorships, more than half the seats in the
United States Senate, and more than a third of the seats in the United States
House of Representatives…The pattern is even more striking when it comes to
selective law schools. A handful of these schools accounts for 25 of the United
States Senators, 74 United States Court of Appeals judges, and nearly 200 of
the more than 600 United States District Court judges…In order to cultivate a
set of leaders with legitimacy in the eyes of the citizenry, it is necessary that the
path to leadership be visibly open to talented and qualified individuals of every
race and ethnicity. All members of our heterogeneous society must have
confidence in the openness and integrity of the educational institutions that
provide that training (Grutter 2003, 332; emphasis added).

This new argument has little to do with the standard educational diversity claims. West
Point and Annapolis are not seeking new, enlarged, unusual perspectives. They are in the
business of molding their students to a uniformly traditional point of view. There is no
African American or Hispanic way to fight a war; there’s just an “Army way” and a “Navy
way” (Levinson, 584). The armed services seek out African American and Hispanic
students to integrate the officer corps. When officers roughly mirror the racial and ethnic
profile of the enlisted ranks, military leaders seem legitimate to those they lead; and the
military as a whole seems legitimate to the citizens it serves. The history of dysfunction
that spread throughout combat operations—especially ground operations—during the
Vietnam War testifies to the importance of “seeming legitimate.”
Consider another example, one that extends Justice O’Connor’s remarks about the
importance of law schools. The Supreme Court is composed of four women and five men.
Two of the Justices are black, one is Hispanic. This composition is no accident.
Administrations since the 1960s—Republican and Democrat alike—have been sensitive to
the gender and racial/ethnic make-up of the Court. President Lyndon Johnson seized the
opportunity in 1967 to appoint Thurgood Marshall. When Marshall retired, President
George H. W. Bush appointed another black Justice, Clarence Thomas. In 1981, President
Ronald Reagan appointed Sandra Day O’Connor, after pledging in his 1980 campaign to
appoint the first woman to the Court. She was followed by Ruth Bader Ginsburg, appointed
in 1993 by President Bill Clinton; by Sonia M. Sotomayor and Elena Kagan, appointed in
2009 and 2010 by President Barack Obama; by Amy Coney Barrett, appointed in 2020 by
President Donald Trump; and by Ketanji Brown Jackson, appointed in 2022 by President
Joe Biden.
These appointments were not made blindly—they were neither color-blind nor gender-
blind. They were made with an eye toward political gain, no doubt, but also with an eye
toward sustaining the legitimacy of the Court before the American public. Moreover, the
composition of the Court reflects the fact that Harvard and Yale, because of their
affirmative action policies, had been producing a significant number of black and Hispanic
law graduates in the 1970s, 1980s, and 1990s. The current Justices possess law degrees
from Harvard (4), Yale (4), and Notre Dame. Excluding current members of the Court,
Justices appointed since 1950 came from Harvard (6), Yale (4) Northwestern (2), Stanford
(2), Columbia, and other (3).
O’Connor’s legitimacy argument constituted a strong counter-point to the standard
diversity line. It directed attention away from the “cognitive growth” and “enhanced
problem-solving skills” in students’ heads and toward the necessity of integrating major
American institutions. In a similar vein Elizabeth Anderson has argued for integration as a
value the Court should recognize as a compelling interest:
Given the realities of race in the U.S., people of different race occupy different
walks of life. So in the U.S., democracy requires racial integration of the main
institutions of civil society, the places where discussion of public import among
citizens take place: public accommodations, workplaces, schools, and
neighborhoods. The same applies to society’s elites, those who play a pivotal
role in formulating and adopting policies of public import. Elites, to be legitimate,
must serve a representative function: they must be capable of and dedicated to
representing the concerns of people from all walks of life, so that the policies
they forge are responsive to these concerns. An elite drawn only from segments
of society that live in isolation from other segments will be ignorant of the
circumstances and concerns of those who occupy other walks of life (Anderson
2004, 22; emphasis added; see also Anderson 2002; for a discussion of
Anderson’s full integration argument, see Lippert-Rasmussen 2020, 144–158).

Unfortunately Justice O’Connor offered her legitimacy argument as though it were a

continuation of the standard diversity justification, inserting it unexpectedly and then
abandoning it just as quickly to resume tracing the byways of diversity. Then she
undermined the argument by suggesting that race-conscious admissions shouldn’t be
necessary after another 25 years. Such measures must be time limited (Grutter 2003,
343).

8. Diversity’s Limit
On June 29, 2023, the Supreme Court decided Students for Fair Admissions v. Harvard. (It
also decided a companion case, Students for Fair Admissions v. University of North
Carolina. I will dwell here only on the Harvard decision. The issues in the two cases are
similar, differing only in minor details.) Students for Fair Admissions (SFFA) contended that
Harvard’s affirmative action policy violated Title VI of the Civil Rights Act of 1964 (and by
extension the Fourteenth Amendment to the Constitution) by discriminating against Asian
American applicants. And it asked the Court to overturn Grutter.
The deciding opinion, written by Chief Justice Roberts, is something of a puzzle. It ruled
that Harvard’s policy transgressed the Constitution but it did not overrule Grutter and it did
not find that Harvard discriminated against Asian American applicants (although it’s
admissions policy had a “negative effect” on them).[13]
The Chief Justice employed the Grutter framework to make his case against Harvard. The
main elements of his decision were these: (1) Harvard’s purported aims—training future
leaders, producing new knowledge through the clash of different perspectives, and the like
—were “not sufficiently precise” to be measured by the Court (SFFA 2023, 22–23); (2) the
racial and ethnic categories used by Harvard were so “opaque” and incoherent that
Harvard couldn’t plausibly establish a tight connection between means (using racial
classifications) and ends (producing new leaders, new knowledge, and so on)
(SFFA 2023, 24–26); (3) Harvard’s use of race fostered stereotypes (something Justice
Powell in Bakke and Justice O’Connor in Grutter warned against) (SFFA 2023, 28–29);
and Harvard’s policy had no end point (SFFA 2023, 30).
The Chief Justice fully embraced Justice O’Connor’s proposition that race-conscious
affirmative action policies must have an end point. Otherwise, he never addressed her
reasoning in Grutter that the University of Michigan Law School’s “holistic” search for
diversity satisfied the strictures of the Constitution. Nor did he did he address her
legitimacy argument. The closest he came was in a footnote:
The United States as amicus curiae contends that race-based admissions
programs further compelling interests at our Nation’s military academies. No
military academy is a party to these cases, however, and none of the courts
below addressed the propriety of race-based systems in that context. This
opinion also does not address that issue, in light of the potentially distinct
interests that military academies may present (SFFA 2023, 22, ftnt 4).

But it was Justice O’Connor who in 2003 broached these interests and gave them weight.
More importantly, she extended her observations about the military academies to colleges
and universities in general: by propelling blacks and Hispanics into elite levels of
leadership, affirmative action was helping legitimate major American institutions.
The Chief Justice’s footnote was a red flag. Within months SFFA had brought suit against
the United States Naval Academy at Annapolis and the United States Military Academy at
West Point.[14]
In the aftermath of the Court’s decision, proponents of affirmative action did not rise up
loudly to defend “diversity in the head” (students’ cognitive growth and analytical skill
development) as a still viable justification of racial and ethnic preferences in admissions.
They turned their sights to other targets. “Legacy” admissions—favorable admission
decisions extended to sons and daughters of an institution’s alumni—immediately jumped
to the top of the list. A commentator in the Wall Street Journal expressed a very common
view:
Fairness requires equal opportunity for all and special privileges for none. Yet
today admission to elite colleges and universities is rife with preferences that
have nothing to do with race and give additional advantages to the
advantaged…Such policies [in particular legacy preferences] are flatly
inconsistent with the nation’s promise of equal opportunity, and institutions like
 Harvard—whose endowment at the end of fiscal 2022 amounted to $51 billion—
can afford to change them (Galston 2023).

So obviously unjust do legacies seem that after the Supreme Court’s SFFA decisions, two
different bills were introduced in Congress to ban them. States are getting into the act, too,
passing or contemplating passing legislation of their own.[15]
Likewise, standardized tests used in college and professional school admissions—
especially the SAT, ACT, and LSAT—are attracting renewed scrutiny. They long have been
charged as biased against black and Hispanic students, who score less well on them than
whites and Asian Americans. In recent years many selective colleges and universities
made the SAT or ACT optional, especially so during the pandemic. But a shift back may be
underway: MIT, Yale, Brown, and Dartmouth recently reinstated the SAT.[16]
Indeed, the whole idea of “merit” is up for grabs now, drawing on, among other sources,
the debate spawned by Michael Sandel’s (2022) thoroughgoing critique of the meritocratic
ideal.
Race-conscious affirmative action is not broadly popular (Gramlich 2023; Borter 2023).
Nine states have banned it by law. California began the trend with Proposition 209, passed
by popular vote in 1996. In 2020, fifty-seven percent of California voters defeated an effort
to roll it back. No national groundswell of protest against the SFFA decisions has emerged
paralleling the reaction to the Dobbs 2022 decision overturning Roe v. Wade.
As Chief Justice Roberts observed in his decision, the issue isn’t diversity or no diversity
(read: racial and ethnic representation), it’s how much (SFFA 2023, 23). And by what
means. Three years after the Grutter decision in 2003, voters in Michigan amended its
constitution to forbid racial and ethnic preferences in admissions. How has the Law School
fared since? Its class of 2026 is 9% black, 10% Hispanic, 4% Native American, 17% Asian
American, and 62% white (Morioka 2023). Is that enough? Too little?
The moral arguments for and against the use of racial and ethnic preferences have not
changed much since the 1970s. They ring variations on a few common themes. For those
readers who want to understand these arguments in depth, the essential resource is
Kasper Lippert-Rasmussen’s Making Sense of Affirmative Action (2020). He shows that
the most common arguments that proponents and opponents deploy are much more
complex than they imagined. But he deftly guides the reader through the complexities; all
he requires is close attention.

Bibliography
Abortion

This article gives an overview of the moral and legal aspects of abortion and evaluates the
most important arguments. The central moral aspect concerns whether there is any
morally relevant point during the biological process of the development of the fetus from its
beginning as a unicellular zygote to birth itself that may justify not having an abortion after
that point. Leading candidates for the morally relevant point are: the onset of movement,
consciousness, the ability to feel pain, and viability. The central legal aspect of the abortion
conflict is whether fetuses have a basic legal right to live, or, at least, a claim to live. The
most important argument with regard to this conflict is the potentiality argument, which
turns on whether the fetus is potentially a human person and thus should be protected.
The question of personhood depends on both empirical findings and moral claims.
The article ends with an evaluation of a pragmatic account. According to this account, one
has to examine the different kinds of reasons for abortion in a particular case to decide
about the reasonableness of the justification given. Take the example of a young, raped
woman. The account suggests that it would seem cruel and callous to force her to give
birth to “her” child. So, if this pragmatic account is correct, some abortions may be morally
justifiable whereas other abortions may be morally reprehensible.

Table of Contents
1.Preliminary Distinctions
1.Three Views on Abortion
2.The Standard Argument
3.The Modified Standard Argument
2.Personhood
3.Moral Aspects of the Abortion Conflict
1.Moral Rights
2.At Birth
3.Viability
4.First Movement
5.Consciousness and the Ability to Feel Pain
6.Unicellular Zygote
7.Thomson and the Argument of The Sickly Violinist
4.Legal Aspects of the Abortion Conflict
1.The Account of Quasi-Rights
2.The Argument of Potentiality
5.A Pragmatic Account
1.First Order Reasons
1.Rape
2.Endangerment of the Woman’s Life
3.Serious Mentally or Physically Disabled Fetuses
2.Second Order Reasons
1.A Journey to Europe
2.Financial and Social Reasons
 3.First Order Reasons vs. Second Order Reasons
6.Public Policy and Abortion
7.Clinical Ethics Consultation and Abortion
8.References and Further Reading

1. Preliminary Distinctions
One of the most important issues in biomedical ethics is the controversy surrounding
abortion. This controversy has a long history and is still heavily discussed among
researchers and the public—both in terms of morality and in terms of legality. The following
basic questions may characterize the subject in more detail: Is abortion morally justifiable?
Does the fetus (embryo, conceptus, and zygote) have any moral and/or legal rights? Is the
fetus a human person and, thus, should be protected? What are the criteria for being a
person? Is there any morally relevant break along the biological process of development
from the unicellular zygote to birth? This list of questions is not meant to be exhaustive, but
it describes the issues of the following analysis.

a. Three Views on Abortion

There are three main views: first, the extreme conservative view (held by the Catholic
Church); second, the extreme liberal view (held by Singer); and third, moderate views
which lie between both extremes. Some opponents (anti-abortionists, pro-life activists)
holding the extreme view, argue that human personhood begins from the unicellular zygote
and thus – according to the religious stance – one should not have an abortion by virtue of
the imago dei of the human being (for example, Schwarz 1990). To have an abortion would
be, by definition, homicide. The extreme liberal view is held by proponents (abortionists).
They claim that human personhood begins immediately after birth or a bit later (Singer).
Thus, they consider the relevant date is at birth or a short time later (say, one month). The
proponents of the moderate views argue that there is a morally relevant break in the
biological process of development – from the unicellular zygote to birth – which determines
the justifiability and non-justifiability of having an abortion. According to them, there is a
gradual process from being a fetus to being an infant where the fetus is not a human being
but a human offspring with a different moral status.
The advantage of the extreme conservative view is the fact that it defines human
personhood from the beginning of life (the unicellular zygote); there is no slippery slope.
However, it seems implausible to say that the zygote is a human person. The advantage of
the extreme liberal view is that its main claim is supported by a common philosophical
usage of the notion “personhood” and thus seems more sound than the extreme
conservative view because the offspring is far more developed; as the unicellular zygote.
This view also faces severe problems; for example, it is not at all clear where the morally
relevant difference is between the fetus five minutes before birth and a just born offspring.
Some moderate views have commonsense plausibility especially when it is argued that
there are significant differences between the developmental stages. The fact that they also
claim for a break in the biological process, which is morally relevant, seems to be a
relapse into old and unjustified habits. As Gillespie stresses in his article “Abortion and
Human Rights” (1984, 94-102) there is no morally relevant break in the biological process
of development. But, in fact, there are differences, which make a comparative basis
possible without having to solve the problem of drawing a line. How should one decide?

b. The Standard Argument

The standard argument is the following practical syllogism:

1.The killing of human beings is prohibited.

2.A fetus is a human being.
3.The killing of fetuses is prohibited.
Hence, abortion is not allowed since homicide is prohibited. It seems obvious to question
the result of the practical syllogism since one is able to argue against both premises. First,
there are possible situations where the first premise could be questioned by noting, for
example that killing in self-defense is not prohibited. Second, the second premise could
also be questioned since it is not at all clear whether fetuses are human beings in the
sense of being persons, although they are of course human beings in the sense of being
members of the species of homo sapiens. Consecutively, one would deny that fetuses are
persons but admit that a young two year old child may be a person. Although, in the end, it
may be difficult to claim that every human being is a person. For example, people with
severe mental handicaps or disorder seem not to have personhood. That is, if personhood
is defined with regard to specific criteria like the capacity to reason, or to have
consciousness, self-consciousness, or rationality, some people might be excluded. But, in
fact, this does not mean that people with severe mental handicaps who lack personhood
can be killed. Even when rights are tied to the notion of personhood, it is clearly prohibited
to kill disabled people. Norbert Hoerster, a well-known German philosopher, claims that
fetuses with severe handicaps can be – like all other fetuses – aborted, as born human
beings with severe handicaps they have to be protected and respected like all other
human beings, too (1995, 159).

c. The Modified Standard Argument

However, it seems appropriate to modify the standard argument and to use a more
sophisticated version. Replace the notion “human being” with “human life form.” The new
practical syllogism is:

1.The killing of human life forms is prohibited.

2.A fetus is a human life form.
3.The killing of fetuses is prohibited.
The objection against the first premise of the standard argument still holds for the new
more sophisticated version. But, the second modified premise is much stronger than the
previous one because one has to determine what a human life form really is. Is a fetus a
human life form? But, even if the fetus is a human life form, it does not necessarily follow
that it should be protected by that fact, simpliciter. The fetus may be a human life form but
it hardly seems to be a person (in the ordinary sense of the notion) and thus has no
corresponding basic right to live. However, as already stated, this kind of talk seems to go
astray because the criteria for personhood may be suitable for just-borns but not
appropriate for fetuses, embryos, or unicellular zygotes, like some biological (human
being), psychological (self-consciousness), rational (ability to reasoning), social (sympathy/
love), or legal (being a human life form with rights) criteria may indicate (for example, Jane
English 1984). Jane English persuasively argues in “Abortion and the Concept of a
Person” that even if the fetus is a person, abortion may be justifiable in many cases, and if
the fetus is no person, the killing of fetuses may be wrong in many cases.

2. Personhood
What does it mean to claim that a human life form is a person? This is an important issue
since the ascription of rights is at stake. I previously stated that it is unsound to say that a
fetus is a person or has personhood since it lacks, at least, rationality and self-
consciousness. It follows that not every human being is also a person according to the
legal sense, and, thus, also lacks moral rights (extreme case). The fetus is by virtue of his
genetic code a human life form but this does not mean that this would be sufficient to grant
it legal and moral rights. Nothing follows from being a human life form by virtue of one’s
genes, especially not that one is able to derive legal or moral rights from this very fact (for
example, speciesism). Is a human person exclusively defined by her membership of the
species Homo sapiens sapiens and thus should be protected? To accept this line of
argumentation would entail the commitment of the existence of normative empirical
features. It seems premature to derive the prohibition to kill a life form from the bare fact of
its genetic feature – including the human life form – unless one argues that human beings
do have the basic interest of protecting their offspring. Is a human life form a moral entity?
This seems to be a good approach. The argument runs as follows: It seems plausible to
claim that human beings create values and, if they have the basic interest of protecting
their offspring, human beings may establish a certain morality by which they can argue, for
example, for the prohibition of abortions. The moral judgment can be enforced through
legal norms (see below).
To be more precise about the assumption of the existence or non-existence of normative,
empirical features: Critics of the view to tie the right to live and the biological category of
being a human being claim that the protagonists effect the is-ought fallacy. Why is it
unsound to take the bare fact of being a member of the biological species Homo
sapiens as a solid basis for granting the right to live? The linkage seems only justified
when there are sound factual reasons. If there are none, the whole line of reasoning would
“hang in the air” so that one could also easily argue for the right to live for cats and dogs.
Only factual relevant features may be important for the linkage. What could these relevant
features look like?
Jane English presents in her article “Abortion and the Concept of a Person” several
features of personhood which characterize the human person. Her notion of personhood
can be grouped into five sectors (English 1984, pp. 152): (i) the biological sector (being a
human being, having extremities, eating and sleeping); (ii) the psychological sector
(perception, emotions, wishes and interests, ability to communicate, ability to make use of
tools, self-consciousness); (iii) the rational sector (reasoning, ability to make
generalizations, to make plans, learning from experience); (iv) the social sector (to belong
to different groups, other people, sympathy and love); and (v) the legal sector (to be a
legal addressee, ability to make contracts, to be a citizen). According to English, it is not
necessary for a human life form to comply with all five sectors and different aspects to
count as a person. A fetus lies right in the penumbra where the concept of personhood is
hard to apply. There is no core of necessary and sufficient features that could be ascribed
to a human life form in order to be sure that these features constitute a person (English
1984, 153).

Mary Anne Warren claims that a human life form should qualify as a person when, at least,
some of the following aspects (especially i-iii) are at stake: (i) consciousness and the
ability to feel pain; (ii) reasoning; (iii) a self-motivated activity; (iv) ability to communicate;
and (v) the existence of a self-concept (for example, individual, racial) and self-
consciousness (Warren 1984, 110-113). Warren argues that the fetus is no person since it
lacks the criteria of personhood and, thus, an abortion is justified.

The aim is not to give an airtight definition of the concept of personhood. The main
question is whether a fetus could qualify as a person. The following can be stated: The
fetus is a human offspring but is not a legal, social, and rational person in the ordinary
sense of the notions. Some aspects of the psychological sector for example, the ability to
feel and perceive can be ascribed to the fetus but not to the embryo, conceptus, or the
(unicellular) zygote. It seems implausible to say that a fetus (or embryo, conceptus,
zygote) is a person, unless one additionally claims that the genetic code of the fetus is a
sufficient condition. However, this does not mean, in the end, that one could always justify
an abortion. It only shows that the fetus could hardly be seen as a human person.

It is hard to keep the legal and moral aspects of the conflict of abortion apart. There are
overlaps which are due to the nature of things since legal considerations are based on the
ethical realm. This can also be seen according to the notion person. What a person is is
not a legal question but a question which is to be decided within a specific ethics. If one
characterizes the notion of a person along some criteria, then the question of which criteria
are suitable or not will be discussed with regard to a specific moral approach (for
example, Kantianism, utilitarianism, virtue ethics). The relevant criteria, in turn, may come
from different areas like the psychological, rational, or social sphere. If the criteria are
settled, this influences the legal sector because the ascription of legal rights – especially
the right to live in the abortion debate – is tied to persons and respectively to the concept
of personhood.

3. Moral Aspects of the Abortion Conflict

The main question with regard to the moral sphere concerns identification of the right
developmental point of the fetus (or the embryo, conceptus, zygote) to decide which break
may morally justify an abortion or not (proponents of the moderate view and the extreme
liberal view claim that there is such a break). The main arguments in the debate will be
evaluated in the following. Before we analyze the arguments, it is necessary to say
something about moral rights.
a. Moral Rights
Some authors claim that the talk of moral rights and moral obligations is an old never-
ending tale. There are no “moral rights” or “moral obligations” per se; at least, in the sense
that there are also moral rights and moral obligations apart from legal rights and legal
obligations. There is no higher ethical authority which may enforce a specific moral
demand. Rights and obligations rest on law. According to ethics, one should better say
“moral agreements” (for example, Gauthier). The proponents claim that moral agreements
do have a similar status to legal rights and legal obligations but stress that no person has
an enforceable demand to have her moral rights prevail over others. The suitability is the
essential aspect of the metaphysics of rights and obligations. Only the formal constraint
establishes rights and obligations within a given society (for example, Hobbes); the
informal constraint within a given society – though it may be stronger – is not able to do so.
Without a court of first instance there are no rights and obligations. Only by using the legal
system is one able to establish specific moral rights and specific moral obligations. Those
authors claim that there are no absolute moral rights and moral obligations which are
universally valid; moral agreements are always subjective and relative. Hence, there are
also no (absolute) moral rights which the fetus (embryo, conceptus, or zygote) may call for.
The only solution may be that the survival of the fetus rests on the will of the human beings
in a given moral society. According to their view, it is only plausible to argue that an
abortion is morally reprehensible if the people in a given society do have a common
interest not to abort and make a moral agreement which is enforced by law.

b. At Birth
Proponents of the liberal view contend that the morally significant break in the biological
development of the fetus is at birth. This means that it is morally permitted to have an
abortion before birth and morally prohibited to kill the offspring after birth. The objection
against this view is simple because there seems to be no morally relevant difference
between a short time (say five minutes) before birth and after it. Factually, the only
biological difference is the physical separation of the fetus from the mother. However it
seems unsound to interpret this as the morally significant difference; the bare evidence
with regard to the visibility of the offspring and the physical separation (that is, the offspring
is no longer dependent on the woman’s body) seems insufficient.

c. Viability
Proponents of the moderate view often claim that the viability criterion is a hot candidate
for a morally significant break because the dependence of the nonviable fetus on the
pregnant woman gives her the right to make a decision about having an abortion. The
aspect of dependence is insufficient in order to determine the viability as a possible break.
Take the following counter-example: A son and his aged mother who is nonviable without
the intensive care of her son; the son has no right to let his mother die by virtue of her
given dependence. However, one may object that there is a difference between “needing
someone to care for you” and “needing to live off a particular person’s body.” Furthermore,
one may stress that the nonviable and the viable fetus both are potential human adults.
But as we will see below the argument of potentiality is flawed since it is unclear how
actual rights could be derived from the bare potentiality of having such rights at a later
time. Hence, both types of fetuses cannot make claim for a right. There is also another
objection that cannot be rebutted: the viability of the fetus regarding the particular level of
medical technology. On the one hand, there is a temporal relativity according to medical
technology. The understanding of what constitutes the viability of the fetus has developed
over time according to the technical level of embryology in the last centuries and decades.
Today, artificial viability allows physicians to rescue many premature infants who would
have previously died. On the other hand, there exists a local relativity according to the
availability of medical supplies in and within countries which determines whether the life of
a premature infant will be saved. The medical supply may vary greatly. Consequently, it
seems inappropriate to claim that viability as such should be regarded as a significant
break by being a general moral justification against abortions.

d. First Movement
The first movement of the fetus is sometimes regarded as a significant break because
proponents stress its deeper meaning which usually rests on religious or non-religious
considerations. Formerly the Catholic Church maintained that the first movement of the
fetus shows that it is the breathing of life into the human body (animation) which separates
the human fetus from animals. This line of thinking is out-of-date and the Catholic Church
no longer uses it. Another point is that the first movement of the fetus that women
experience is irrelevant since the real first movement of the fetus is much earlier.
Ultrasonic testing shows that the real first movement of the fetus is somewhere between
the 6th and 9th week. But even if one considers the real first movement problems may
arise. The physical ability to move is morally irrelevant. One counter-example: What about
an adult human being who is quadriplegic and is unable to move? It seems out of the
question to kill such people and to justify the killing by claiming that people who are
disabled and simply lack the ability to move are, therewith, at other people’s disposal.

e. Consciousness and the Ability to Feel Pain

In general, proponents of moderate views believe that consciousness and the ability to feel
pain will develop after about six months. However the first brain activities are discernable
after the seventh week so that it is possible to conclude that the fetus may feel pain after
this date. In this respect, the ability to suffer is decisive for acknowledging a morally
significant break. One may object to this claim, that the proponents of this view redefine
the empirical feature of “the ability to suffer” as a normative feature (is-ought fallacy). It is
logically unsound to conclude from the bare fact that the fetus feels pain that it is morally
reprehensible or morally prohibited per se to abort the fetus.

f. Unicellular Zygote
Proponents of the extreme conservative view claim that the morally significant break in the
biological development of the fetus is given with the unicellular human zygote. They argue
that the unicellular zygote is a human person, and thus, it is prohibited to have an abortion
because one kills a human being (for example, Schwarz).

The extreme conservative proponents argue that biological development from the fetus to
a human being is an incremental process which leaves no room for a morally significant
break (liberals deny this line of thinking). If there is no morally significant break, then the
fetus has the same high status of a newborn, or the newborn has the same low status of
the fetus.

To many opponents of the “extreme” conservative position, it seems questionable to claim

that a unicellular zygote is a person. At best, one may maintain that the zygote will
potentially develop into a human being. Except the potentiality argument is flawed since it
is impossible to derive current rights from the potential ability of having rights at a later
time. Opponents (for example, Gert) also object to any attempt to base conclusions on
religious considerations that they believe cannot stand up to rational criticism. For these
reasons, they argue that the conservative view should be rejected.

g. Thomson and the Argument of The Sickly Violinist

Judith Jarvis Thomson presents an interesting case in her landmark article “A Defense of
Abortion” (1971) in order to show that, even if the fetus has a right to live, one is still able
to justify an abortion for reasons of a woman’s right to live/integrity/privacy. Thomson’s
famous example is that of the sickly violinist: You awake one morning to find that you have
been kidnapped by a society of music lovers in order to help a violinist who is unable to
live on his own by virtue of his ill-health. He has been attached to your kidneys because
you alone have the only blood type to keep him alive. You are faced with a moral dilemma
because the violinist has a right to live by being a member of the human race; there seems
to be no possibility to unplug him without violating this right and thus killing him. However,
if you leave him attached to you, you are unable to move for months, although you did not
give him the right to use your body in such a way (Thomson 1984, 174-175).

First, Thomson claims that the right to live does not include the right to be given the means
necessary for survival. If the right to live entails the right to those means, one is not
justified in preventing the violinist from the on-going use of one’s kidneys. The right to the
on-going use of the kidneys necessarily implies that the violinist’s right to his means for
survival always trumps the right to another person’s body. Thomson refuses this and
claims that “the fact that for continued life that violinist needs the continued use of your
kidneys does not establish that he has a right to be given the continued use of your
kidneys” (Thomson 1984, 179). She argues that everybody has a right of how his own
body is used. That is, the violinist has no right to use another person’s body without her
permission. Therefore, one is morally justified in not giving the violinist the use of one’s
own kidneys.

Second, Thomson contends that the right to live does not include the right not to be killed.
If the violinist has the right not to be killed, then another person is not justified in removing
the plug from her kidneys although the violinist has no right to their use. According to
Thomson, the violinist has no right to another person’s body and hence one cannot be
unjust in unplugging him: “You surely are not being unjust to him, for you gave him no right
to use your kidneys, and no one else can have given him any such right” (Thomson 1984,
180). If one is not unjust in unplugging oneself from him, and he has no right to the use of
another person’s body, then it cannot be wrong, although the result of the action is that the
violinist will be killed.

4. Legal Aspects of the Abortion Conflict

What is the legal status of the fetus (embryo, conceptus, and zygote)? Before the question
is answered, one should pay some attention to the issue of the genesis of a legal system.
Which ontological status do legal rights have? Where do they come from? Usually we
accept the idea that legal rights do not “fall from the blue sky” but are made by human
beings. Other conceptions which had been provided in the history of human kind are:

1.rights rest on God’s will;

2.rights rest on the strongest person; or
3.rights rest on a specific human feature like a person’s wisdom or age.
However, let us take the following description for granted: There is a legal community in
which the members are legal entities with (legal) claims and legal addressees with (legal)
obligations. If someone refuses the addressee’s legal obligation within such a system, the
legal entity has the right to call the legal instance in order to let his right be enforced. The
main question is whether the fetus (or the embryo, conceptus, zygote) is a legal person
with a basic right to live or not and, furthermore, whether there will be a conflict of legal
norms, that is a conflict between the fetus’ right to live and the right of self-determination of
the pregnant woman (principle of autonomy). Is the fetus a legal entity or not?

a. The Account of Quasi-Rights

It was previously stated that the fetus as such is no person and that it seems unsound to
claim that fetuses are persons in the ordinary sense of the notion. If rights are tied to the
notion of personhood, then it seems appropriate to say that fetuses do not have any legal
rights. One can object that animals of higher consciousness (or even plants, see
Korsgaard 1996, 156) have some “rights” or quasi-rights because it is prohibited to kill
them without good reason (killing great apes and dolphins for fun is prohibited in most
countries). Their “right” not to be killed is based on the people’s will and their basic interest
not to kill higher developed animals for fun. But, it would be wrong to assume that those
animals are legal entities with “full” rights, or that they have only “half” rights. Thus, it
seems reasonable to say that animals have “quasi-rights.” There is a parallel between the
so-called right of the fetus and the quasi-rights of some animals: both are not persons in
the normal sense of the notion but it would cause us great discomfort to offer them no
protection and to deliver them to the vagaries of the people. According to this line of
argument, it seems sound to claim that fetuses also have quasi-rights. It does not follow
that the quasi-rights of the fetuses and the quasi-rights of the animals are identical; people
would normally stress that the quasi-rights of fetuses are of more importance than that of
animals.
However, there are some basic rights of the pregnant woman, for example, the right of
self-determination, the right of privacy, the right of physical integrity, and the right to live.
On the other hand, there is the existential quasi-right of the fetus, that is, the quasi-right to
live. If the presumption is right that legal rights are tied to the notion of personhood and
that there is a difference between rights and quasi-rights, then it seems right that the fetus
has no legal right but “just” a quasi-right to live. If this is the case, what about the relation
between the existential quasi-right of the fetus and the basic legal rights of the pregnant
woman? The answer seems obvious: quasi-rights cannot trump full legal rights. The fetus
has a different legal status that is based on a different moral status (see above). On this
view there is no legal conflict of rights.

b. The Argument of Potentiality

Another important point in the debate about the ascription of legal rights to the fetus is the
topic of potential rights. Joel Feinberg discusses this point in his famous article
“Potentiality, Development, and Rights” (1984, 145-151) and claims that the thesis that
actual rights can be derived from the potential ability of having such rights is logically
flawed because one is only able to derive potential rights from a potential ability of having
rights. Feinberg maintains that there may be cases where it is illegal or wrong to have an
abortion even when the fetus does not have any rights or is not yet a moral person. To
illustrate his main argument – that rights do not rest on the potential ability of having them
– Feinberg considers Stanley Benn’s argument which I slightly modified:

If person X is President of the USA and thus is Commander in Chief of the

army, then person X had the potential ability to become the President of the
USA and Commander in Chief of the army in the years before his rule.

But, it does not follow that:

The person X has the authority to command the army as potential President
of the USA.

Thus, it seems incorrect to derive actual rights from the bare potential ability to have legal
rights at a later time. It should be added that Benn – despite his criticism on the argument
of potential rights – also claims that there are valid considerations which do not refer to the
talk of rights and may provide plausible reasons against infanticide and late abortions even
when fetuses and newborns are lawless beings with no personhood.
5. A Pragmatic Account
There is always a chance that women get pregnant when they have sex with their
(heterosexual) partners. There is not a 100% certainty of not getting pregnant under
“normal circumstances”; there is always a very small chance even by using contraception
to get pregnant. However, what does the sphere of decisions look like? A pregnancy is
either deliberate or not. If the woman gets deliberately pregnant, then both partners
(respectively the pregnant woman) may decide to have a baby or to have an abortion. In
the case of having an abortion there may be good reasons for having an abortion with
regard to serious health problems, for example, a (seriously) disabled fetus or the
endangerment of the woman’s life. Less good reasons seem to be: vacation, career
prospects, or financial and social grievances. If the pregnancy is not deliberate, it is either
self-caused in the sense that the partners knew about the consequences of sexual
intercourses and the contraception malfunctioned or it is not self-caused in the sense of
being forced to have sex (rape). In both cases the fetus may be aborted or not. The
interesting question concerns the reasons given for the justification of having an abortion.

There are at least two different kinds of reasons or justifications: The first group will be
called “first order reasons”; the second “second order reasons.” First order reasons are
reasons of justifications which may plausibly justify an abortion, for example, (i) rape, (ii)
endangerment of the woman’s life, and (iii) a serious mentally or physically disabled
fetus. Second order reasons are reasons of justifications which are, in comparison to first
order reasons, less suitable in providing a strong justification for abortion, for example, (i) a
journey, (ii) career prospects, (iii) by virtue of financial or social grievances.

a. First Order Reasons

i. Rape
It would be cruel and callous to force the pregnant woman who had been raped to give
birth to a child. Judith Jarvis Thomson maintains in her article “A Defense of Abortion” that
the right to live does not include the right to make use of a foreign body even if this means
having the fetus aborted (Thomson 1984, pp. 174 and pp. 177). Both the fetus and the
raped woman are “innocent,” but this does not change “the fact” that the fetus has any
rights. It seems obvious in this case that the raped woman has a right to abort. Forcing her
not to abort is to remind her of the rape day-by-day which would be a serious mental strain
and should not be enforced by law or morally condemned.

However, this assumption would be premature from John Noonan’s viewpoint according to
his article “An Almost Absolute Value in History” (Noonan 1970, 51-59). He claims that
the fetus as human [is] a neighbor; his life [has] parity with one’s own […] [which] could be
put in humanistic as well as theological terms: do not injure your fellow man without
reasons. In these terms, once the humanity of the fetus is perceived, abortion is never
right except in self-defense. When life must be taken to save life, reason alone cannot say
that a mother must prefer a child’s life to her own. With this exception, now of great rarity,
abortion violates the rational humanist tenet of the equality of human lives.

Hence, the woman has no right to abort the fetus even if she had been raped and got
pregnant against her will. This is the consequence of Noonan’s claim since he only permits
having an abortion in self-defense while Thomson argues that women, in general, have a
right to abort the fetus when the fetus is conceived as an intruder (for example, due to
rape). But, it remains unclear what Noonan means by “self-defense.” At the end of his
article he states that “self-sacrifice carried to the point of death seemed in extreme
situations not without meaning. In the less extreme cases, preference for one’s own
interests to the life of another seemed to express cruelty or selfishness irreconcilable with
the demands of love” (Noonan 1970). On this view, even in the standard case of self-
defense — for example, either the woman’s life or the life of the fetus — the pregnant
woman’s death would not be inappropriate and in less extreme cases the raped woman
would express cruelty or selfishness when she aborts the fetus — a judgment not all
people would agree with.

ii. Endangerment of the Woman’s Life

Furthermore, there is no good reason to proceed with a pregnancy when the woman’s life
is in serious danger. Potential life should not be more valued then actual life. Of course, it
is desirable to do everything possible to rescue both but it should be clear that the
woman’s life “counts more” in this situation. To force her at the risk of her life means to
force her to give up her right of self-defense and her right to live. There seems to be no
good reason to suspend her basic right of self-defense.
iii. Serious Mentally or Physically Disabled Fetuses
It is hard to say when exactly a fetus is seriously mentally or physically disabled because
this hot issue raises the vital question of whether the future life of the disabled fetus is
regarded as worth living (problem of relativity). Hence, there are simple cases and, of
course, borderline cases which lie in the penumbra and are hard to evaluate. Among the
simple cases take the following example: Imagine a human torso lacking arms and legs
that will never develop mental abilities like self-consciousness, the ability to communicate,
or the ability to reason. It seems quite obvious to some people that such a life is not worth
living. But what about the high number of borderline cases? Either parents are not entitled
to have a healthy and strong offspring, nor are the offspring entitled to become healthy and
strong. Society should not force people to give birth to seriously disabled fetuses or
morally worse to force mothers who are willing to give birth to a disabled fetus to have an
abortion (for example, Nazi Germany). It seems clear that a rather small handicap of the
fetus is not a good reason to abort it.

Often radical groups of disabled persons claim that, if other people hold the view that it is
all right to abort fetuses with (serious) genetic handicaps, the same people therewith deny
the basic right to live of disabled adults with serious handicaps (see Singer debate). This
objection is unreasonable since fetuses in contrast to adult human beings have no basic
interest in continuing to live their lives. Disabled fetuses may be aborted like other fetuses,
disabled (adult) human persons have to be respected like other people.

b. Second Order Reasons

i. A Journey to Europe
With regard to the reasons of justification according to the second group, there is a specific
view which is based on the argument that it is the decision of the woman to have an
abortion or not.

There is a related view that rests on the assumption of the pregnant woman who claims
that the fetus is a part of her body like a limb so that she has the right to do what ever she
wants to do with the fetus. The argument is wrong. The fetus is certainly not a simple part
of the pregnant woman but, rather, a dependent organism that relies on the woman.
The following example, the journey to Europe from North America, is based on the feminist
argument but it is somewhat different in stressing another point in the line of
argumentation: A young woman is pregnant in the seventh month and decides to make a
journey to Europe for a sight-seeing tour. Her pregnancy is an obstacle to this and she
decides to have an abortion. She justifies her decision by claiming that it will be possible
for her to get pregnant whenever she wants but she is only able to make the journey now
by virtue of her present career prospects. What can be said of her decision? Most authors
may feel a deep discomfort not to morally condemn the action of the woman or not to
reproach her for her decision for different reasons. But, there seems only two possible
answers which may count as a valid basis for morally blaming the woman for her decision:
First, if the young woman lives in a moral community where all members hold the view that
it is immoral to have an abortion with regard to the reason given, then her action may be
morally reprehensible. Furthermore, if the (moral) agreement is enforced by law, the
woman also violated the particular law for which she has to take charge of. Second, one
could also blame her for not showing compassion for her potential child. People may think
that she is a callous person since she prefers to make the journey to Europe instead of
giving birth to her almost born child (seventh month). If the appeal to her mercy fails, one
will certainly be touched by her “strange” and “inappropriate” action. However, the
community would likely put some informal pressure on the pregnant woman to influence
her decision not to have an abortion. But some people may still contend that this social
pressure will not change anything about the fact that the fetus has no basic right to live
while claiming that the woman’s decision is elusive.

ii. Financial and Social Reasons

A woman got pregnant (not deliberately) and wants to have an abortion by virtue of her
bad financial and social background because she fears that she will be unable to offer the
child an appropriate life perspective. In this case, the community should do everything
possible to assist the woman if she wants to give birth to her child. Or, some may argue,
that society should offer to take care of her child in special homes with other children or to
look for other families who are willing to house another child. According to this line of
thinking, people may claim that the financial or social background should not be decisive
for having an abortion if there is a true chance for help.

c. First Order Reasons vs. Second Order Reasons

There is a difference between the first order reasons and the second order reasons. We
already saw that the first order reasons are able to justify an abortion while the second
order reasons are less able to do so. That is because people think that the second order
reasons are weaker than the reasons of the first group. It seems that the human ability to
show compassion for the fetus is responsible for our willingness to limit the woman’s basic
right of autonomy where her reasons are too elusive. However, one may state that there
are no strong compulsive reasons which could morally condemn the whole practice of
abortion. Some people may not unconvincingly argue that moral agreements and legal
rights are due to human beings so that reasons for or against abortion are always
subjective and relative. According to this view, one is only able to contend the “trueness” or
“wrongness” of a particular action in a limited way. Of course, there are other people who
argue for the opposite (for example, Kantians, Catholic Church). One reason why people
have strong feelings about the conflict of abortion is that human beings do have strong
intuitive feelings, for example, to feel compassion for fetuses as helpless and most
vulnerable human entities. But moral intuitionism falls short by being a valid and objective
basis for moral rights.
In the end, it is a question of a particular moral approach whether one regards an abortion
as morally justifiable or not. But not every approach is justified. There is no anything goes.

6. Public Policy and Abortion

One of the most difficult issues is how to make a sound policy that meets the needs of
most people in a given society without focusing on the extreme conservative view, or the
extreme liberal view, or the many moderate views on the conflict of abortion. The point is
simple, one cannot wait until the philosophical debate is settled, for maybe there is no one
solution available. But, in fact, people in a society must know what the policy is; that is,
they have to know when and under what circumstances abortion is permitted or altogether
prohibited. What are the reasons for a given policy? Do they rest on religious beliefs or do
they depend on cultural claims? Whose religious beliefs and whose cultural claims? Those
beliefs and claims of most people or of the dominant group in a given society ? What about
the problem of minority rights? Should they be respected or be refused? These are hard
questions; no one is able to yet give a definite response.

But, of course, the problem of abortion has to be “solved,” at least, with regard to practical
matters. This means that a good policy does not rest on extreme views but tries to cover
as many points of views, although being aware of the fact that one is not able to please
every person in society. This would be an impossible task. It seems that one should adopt
a moderate view rather than the proposed extreme views. This is not because the
moderate view is “correct” but because one needs a broad consensus for a sound policy.
The hardliners in the public debate on the conflict of abortion, be they proponents or
opponents, may not be aware of the fact that neither view is sustainable for most people.

A sound way for governments with regard to a reasonable policy could be the acceptance
of a more or less neutral stance that may function as a proper guide for law. But, in fact,
the decisive claim of a “neutral stance” is, in turn, questionable. All ethical theories try to
present a proper account of a so-called neutral stance but there is hardly any theory that
could claim to be sustainable with regard to other approaches. However, the key seems to
be, again, to accept a middle way to cover most points of views. In the end, a formation of
a policy seeks a sound compromise people could live with. But this is not the end of the
story. One should always try to find better ways to cope with hard ethical problems. The
conflict of abortion is of that kind and there is no evidence to assume otherwise.

7. Clinical Ethics Consultation and Abortion

The vital issue of how one chooses whether or not to have an abortion is of utmost
importance since people, in particular women, want to have a proper “guideline” that can
support them in their process of ethical decision-making. According to pregnant women,
the most crucial point seems not to be whether abortion is morally legitimate or not but,
rather, how one should deliberate in the particular case. In fact, observations regularly
show that women will nearly have the same number of abortions in contexts in which it is
legal or not.
 Gert is right in claiming that “the law can allow behavior that some people regard as
morally unacceptable, such as early abortion, and it can prohibit behavior that some
people regard as morally acceptable, such as late abortion. No one thinks that what the
law decides about abortion settles the moral issue” (Gert 2004, 138). But what follows from
that? What aspects should one consider and how should one decide in a particular case?

It would be best to consult a neutral person who has special knowledge and experiences
in medicine and medical ethics (for example, clinical ethics consultation). Most people are
usually not faced with hard conflicts of abortion in their daily lives and get simply swamped
by it; they are unable to determine and evaluate all moral aspects of the given case and to
foresee the relevant consequences of the possible actions (for example, especially with
regard to very young women who get pregnant by mistake). They need professional help
without being dominated by the person in order to clarify their own (ethical) stance.

However, the conflict of abortion as such may not be solvable, in the end, but the
experienced professional is able to provide persons with feasible solutions for the
particular case.

8. References and Further Reading

 Boonin, David (2002), A Defense of Abortion Cambridge: Cambridge University Press.
 Boylan, Michael (2002), “The Abortion Debate in the 21st Century” in Medical Ethics, ed.
Michael Boylan. Upper Saddle River, NJ: Prentice Hall.
 Chadwick, Ruth, Kuhse, Helga, Landman, Willem et al. (2007), The Bioethics Reader.
Editor’s Choice Oxford: Blackwell Publishers.
 English, Jane (1984), “Abortion and the Concept of a Person,” in: The Problem of
Abortion, 151-161.
 Feinberg, Joel (1984), “Potentiality, Development, and Right,” in: The Problem of
Abortion, 145-150.
 Feinberg, Joel (1984), The Problem of Abortion, Belmont: Wadsworth.
 Gauthier, David (1986), Morals by Agreement, Oxford: Oxford University Press.
 Gert, Bernard (2004), Common Morality. Deciding What to Do, Oxford: Oxford University
Press.
 Gillespie, Norman (1984), “Abortion and Human Rights,” in: The Problem of Abortion, 94-
102.
 Gordon, John-S. (2005), “Die moralischen und rechtlichen Dimensionen der
Abtreibungsproblematik,” in: Conjectura, 43-62.
 Hoerster, Norbert (1995), Abtreibung im säkularen Staat, Frankfurt am Main: Suhrkamp.
 Hobbes, Thomas (1996), Leviathan, Ed. Richard Tuck Cambridge: Cambridge University
Press.
 Korsgaard, Christine (1996), The Sources of Normativity, Cambridge: Cambridge
University Press.
 Noonan, John T. (1970), “An Almost Absolute Value in History,” in: The Morality of
Abortion: Legal and Historical Perspectives, Cambridge: Harvard University Press, 51-59.
 Noonan, John T. (1970), The Morality of Abortion: Legal and Historical
Perspectives, Cambridge: Harvard University Press.
 Schwarz, Stephen (1990), Moral Questions of Abortion, Chicago: Loyola University Press.
 Singer, Peter (1993), Practical Ethics, Cambridge: Cambridge University Press.
 Sumner, Wayne (1980), Abortion and Moral Theory, Princeton: Princeton University Press.
 Thomson, Judith J. (1984), “A Defense of Abortion,” in: The Problem of Abortion, 173-188.
 Tooley, Michael (1983), Abortion and Infanticide, Oxford: Oxford University Press.
 Warren, Mary A. (1984), “On the Moral and Legal Status of Abortion,” in: The Problem of
Abortion, 102-119.
 Warren, Mary A. (1997), “Abortion,” in: A Companion to Ethics, Oxford: Blackwell
Publishers, 303-314.

John-Stewart Gordon
Email: john-stewart.gordon@rub.de
Ruhr-University Bochum
Germany

The Philosophy of Climate Science

Climate change is one of the defining challenges of the 21st century. But what is climate change, how do we know
about it, and how should we react to it? This article summarizes the main conceptual issues and questions in the
foundations of climate science, as well as of the parts of decision theory and economics that have been brought to
bear on issues of climate in the wake of public discussions about an appropriate reaction to climate change.

We begin with a discussion of how to define climate. Even though “climate” and “climate change” have become
ubiquitous terms, both in the popular media and in academic discourse, the correct definitions of both notions are
hotly debated topics. We review different approaches and discuss their pros and cons. Climate models play an
important role in many parts of climate science. We introduce different kinds of climate models and discuss their
uses in detection and attribution, roughly the tasks of establishing that the climate of the Earth has changed and of
identifying specific factors that cause these changes. The use of models in the study of climate change raises the
question of how well-confirmed these models are and of what their predictive capabilities are. All this is subject to
considerable debate, and we discuss a number of different positions. A recurring theme in discussions about
climate models is uncertainty. But what is uncertainty and what kinds of uncertainties are there? We discuss
different attempts to classify uncertainty and to pinpoint their sources. After these science-oriented topics, we turn
to decision theory. Climate change raises difficult questions such as: What is the appropriate reaction to climate
change? How much should we mitigate? To what extent should we adapt? What form should adaptation take? We
discuss the framing of climate decision problems and then offer an examination of alternative decision rules in the
context of climate decisions.

Table of Contents
1.Introduction
2.Defining Climate and Climate Change
3.Climate Models
4.Detection and Attribution of Climate Change
5.Confirmation and Predictive Power
6.Understanding and Quantifying Uncertainty
7.Conceptualising Decisions Under Uncertainty
 8.Managing Uncertainty
9.Conclusion
10.Glossary
11.References and Further Reading

1. Introduction
Climate science is an umbrella term referring to scientific disciplines studying aspects of the Earth’s climate. It
includes, among others, parts of atmospheric science, oceanography, and glaciology. In the wake of public
discussions about an appropriate reaction to climate change, parts of decision theory and economics have also
been brought to bear on issues of climate. Contributions from these disciplines that can be considered part of the
application of climate science fall under the scope of this article. At the heart of the philosophy of climate science
lies a reflection on the methodology used to reach various conclusions about how the climate may evolve and what
we should do about it. The philosophy of climate science is a new sub-discipline of the philosophy of science that
began to crystalize at the turn of the 21 st century when philosophers of science started having a closer look at
methods used in climate modelling. It comprises a reflection on almost all aspects of climate science, including
observation and data, methods of detection and attribution, model ensembles, and decision-making under
uncertainty. Since the devil is always in the detail, the philosophy of climate science operates in close contact with
science itself and pays careful attention to the scientific details. For this reason, there is no clear separation
between climate science and the philosophy thereof, and conferences in the field are often attended by both
scientists and philosophers.
This article summarizes the main problems and questions in the foundations of climate science. Section 2 presents
the problem of defining climate. Section 3 introduces climate models. Section 4 discusses the problem of detecting
and attributing climate change. Section 5 examines the confirmation of climate models and the limits of
predictability. Section 6 reviews classifications of uncertainty and the use of model ensembles. Section 7 turns to
decision theory and discusses the framing of climate decision problems. Section 8 introduces alternative decision
rules. Section 9 offers a few conclusions.

Two qualifications are in order. First, we review issues and questions that arise in connection with climate science
from a philosophy of science perspective, and with special focus on epistemological and decision-theoretic
problems. Needless to say, this is not the only perspective. Much can be said about climate science from other
points of view, most notably science studies, sociology of science, political theory, and ethics. For want of space, we
cannot review contributions from these fields.

Second, to guard against possible misunderstandings, it ought to be pointed out that engaging in a critical
philosophical reflection on the aims and methods of climate science is in no way tantamount to adopting a position
known as climate scepticism. Climate sceptics are a heterogeneous group of people who do not accept the results of
‘mainstream’ climate science, encompassing a broad spectrum from those who flat out deny the basic physics of
the greenhouse effect (and the influence of human activities on the world’s climate) to a small minority who
actively engage in scientific research and debate and reach conclusions at the lowest end of climate impacts.
Critical philosophy of science is not the handmaiden of climate scepticism; nor are philosophers ipso facto climate
sceptics. So, it should be stressed here that we do not endorse climate scepticism. We aim to understand how
climate science works, reflect on its methods, and understand the questions that it raises.

2. Defining Climate and Climate Change

Climate talk is ubiquitous in the popular media as well as in academic discourse, and climate change has become a
familiar topic. This veils the fact that climate is a complex concept and that the correct definitions
of climate and climate change are a matter of controversy. To gain an understanding of the notion of climate, it is
important to distinguish it from weather. Intuitively speaking, the weather at a particular place and a particular
time is the state of the atmosphere at that place and at the given time. For instance, the weather in central London
at 2 pm on 1 January 2015 can be characterised by saying that the temperature is 12 degrees centigrade, the
humidity is 65%, and so forth. By contrast, climate is an aggregate of weather conditions: it is a distribution of
particular variables (called the climate variables) arising for a particular configuration of the climate system.
The question is how to make this basic idea precise, and this is where different approaches diverge. 21st-century
approaches to defining climate can be divided into two groups: those that define climate as a distribution over time,
and those that define climate as an ensemble distribution. The climate variables in both approaches include those
that describe the state of the atmosphere and the ocean, and sometimes also variables describing the state of
glaciers and ice sheets [IPCC 2013].
Distribution over time. The state of the Earth depends on external conditions of the system such as the amount of
energy received from the sun and volcanic activity. Assume that there is a period of time over which the external
conditions are relatively stable in that they exhibit small fluctuations around a constant mean value c. One can
then define the climate over this time period as the distribution of the climate variables over that period under
constant external conditions c [for example, Lorenz 1995]. Climate change then amounts to successive time periods
being characterised by different distributions. However, in reality the external conditions are not constant and
even when there are just slight fluctuations around c, the resulting distributions may be very different. Hence this
definition is unsatisfactory [Werndl 2015].
This problem can be avoided by defining climate as the empirically observed distribution over a specific period of
time, where external conditions are allowed to vary. Again, climate change amounts to different distributions for
successive time periods. This definition is popular because it is easy to estimate from the observations, for
example, from the statistics taken over thirty years that are published by the World Meteorological Organisation
[Hulme et al. 2009]. A major problem of this definition can be illustrated by the example in which, in the middle of
a period of time, the Earth is hit by a meteorite and becomes a much colder place. Clearly, the climate before and
after the hit of the meteor differ. Yet this definition has no resources to recognize this because all it says is that
climate is a distribution arising over a specific time period.

To circumvent this problem, Werndl [2015] introduces the idea of regimes of varying external conditions and
suggests defining climate as the distribution over time of the climate variables arising under a specific regime of
varying external conditions. The challenge for this account is to spell out what exactly is meant by a regime of
varying external conditions.

Ensemble Distribution. An ensemble of climate systems (not to be confused with a model ensemble) is an infinite
collection of virtual copies of the climate system. Consider the sub-ensemble of these that satisfy the condition that
the present values of the climate variables lie in a specific interval around the values measured in the actual climate
system (that is, the values compatible with the measurement accuracy). Now assume again that there is period of
time over which the external conditions are relatively stable in that they exhibit small fluctuations around a
constant mean value c. Then climate at future time t is defined as the distribution of values of the climate variables
that arises when all systems in the ensemble evolve from now to t under constant external conditions c [for
example, Lorenz 1995]. In other words, the climate in the future is the distribution of the climate variables over all
possible climates that are consistent with current observations under the assumption of constant external
conditions c.
As we have seen previously, in reality, external conditions are not constant and even small fluctuations around a
mean value can lead to different distributions [Werndl 2015]. This worry can be addressed by tracing the
development of the initial condition ensemble under actual external conditions. The climate at future time t then is
the distribution of the climate variables that arises when the initial conditions ensemble is evolved forward for the
actual path taken by the external conditions [for example, Daron and Stainforth 2013].
This definition faces a number of conceptual challenges. First, it makes the world’s climate dependent on our
knowledge (via measurement accuracy), but this is counterintuitive because we think of climate as something
objective that is independent of our knowledge. Second, the above definition is a definition of future climate, and it
is difficult to see how the present and past climate should be defined. Yet without a notion of the present and past
climate one cannot define climate change. A third problem is that ensemble distributions (and thus climate) do not
relate in a straightforward way to the past time series of observations of the actual Earth and this would imply that
the climate cannot be estimated from them [compare, Werndl 2015].
These considerations show that defining climate is nontrivial and there is no generally accepted or uncontroversial
definition of climate.

3. Climate Models
A climate model is a representation of particular aspects of the climate system. One of the simplest climate models
is an energy-balance model, which treats the Earth as a flat surface with one layer of atmosphere above it. It is
based on the simple principle that in equilibrium the incoming and outgoing radiation must be equal (see Dessler
[2011], Chapters 3-6, for a discussion of such models). This model can be refined by dividing the Earth into zones,
allowing energy transfer between zones, or describing a vertical profile of the atmospheric characteristics. Despite
their simplicity, these models provide a good qualitative understanding of the greenhouse effect.

Modern climate science aims to construct models that integrate as much as possible of the known science (for an
introduction to climate modelling see [McGuffie and Henderson-Sellers 2005]). Typically, this is done by dividing
the Earth (both the atmosphere and ocean) into grid cells. In 2020, global climate models have a horizontal grid
scale of around 150 km. Climatic processes can then be conceptualised as flows of physical quantities such as heat
or vapour from one cell to another. These flows are mathematically described by equations. These equations form
the ‘dynamical core’ of a global circulation model (GCM). The equations typically are intractable with analytical
methods, and powerful supercomputers are used to solve them. For this reason, they are often referred to as
‘simulation models’. To solve equations numerically, time is discretised. Current state-of-the-art simulations use
time steps of approximately 30 minutes, taking weeks or months in real time on supercomputers to simulate a
century of climate evolution.

In order to compute a single hypothetical evolution of the climate system (a ‘model run’), we also require an initial
condition and boundary conditions. The former is a mathematical description of the state of the climate system
(projected into the model’s own domain) at the beginning of the period being simulated. The latter are values for
any variables which affect the system, but which are not directly output by the calculations. These include, for
instance, the concentration of greenhouse gases, the amount of aerosols in the atmosphere at a given time, and the
amount of solar radiation received by the Earth. Since these are drivers of climatic change, they are often referred
to as external forcings or external conditions.
Where processes occur on a smaller scale than the grid, they may be included via parameterisation, whereby the net
effect of the process is separately calculated as a function of the grid variables. For instance, cloud formation is a
physical process that cannot be directly simulated because typical clouds are much smaller than the grid. So, the
net effect of clouds is usually parameterised (as a function of temperature, humidity, and so forth) in each grid cell
and fed back into the calculation. Sub-grid processes are one of the main sources of uncertainty in climate models.
There are now dozens of global climate models under continuous development by national modelling centres like
NASA, the UK Met Office, and the Beijing Climate Center, as well as by smaller institutions. An exact count is
difficult because many modelling centres maintain multiple versions based on the same foundation. As an
indication, in 2020 there were 89 model-versions submitted to CMIP6 (Coupled Model Intercomparison Project
phase 6), from 35 modelling groups, though not all of these should be thought of as being “independent” models
since assumptions and algorithms are often shared between institutions. In order to be able to compare outputs of
these different models, the Coupled Model Intercomparison Project (CMIP) defines a suite of standard
experiments to be run for each climate model. One standard experiment is to run each model using the historical
forcings experienced during the twentieth century so that the output can be directly compared against real climate
system data.

Climate models are used in many places in climate science, and their use gives rise to important questions. These
questions are discussed in the next three sections.

4. Detection and Attribution of Climate Change

Every empirical study of climate has to begin by observing the climate. Meteorological observatories measure a
number of variables such as air temperature near the surface of the Earth using thermometers. But more or less
systematic observations are available since about 1750, and hence to reconstruct the climate before then scientists
have to rely on proxy data: data for climate variables that are derived from observing other natural phenomena such
as tree rings, ice cores, and ocean sediments.
The use of proxy data raises a number of methodological problems centred around the statistical processing of
such data, which are often sparse, highly uncertain, and several inferential steps away from the climate variable of
interest. These issues were at the heart of what has become known as the Hockey Stick controversy, which broke at
the turn of the century in connection with a proxy-based reconstruction of the Northern Hemisphere temperature
record [Mann, Bradley and Hughes, 1998]. The sceptics pursued two lines of argument. They cast doubt on the
reliability of the available data, and they argued that the methods used to process the data are such that they would
produce a hockey-stick-shaped curve from almost any data [for example, McIntyre and McKitrick 2003]. The
papers published by the sceptics raised important issues and stimulated further research, but they were found to
contain serious flaws undermining their conclusions. There are now more than two dozen reconstructions of this
temperature record using various statistical methods and proxy data sources. Although there is indeed a wide
range of plausible past temperatures, due to the constraints of the data and methods, these studies do robustly
support the consensus that, over the past 1400 years, temperatures during the late 20th century are likely to have
been the warmest [Frank et al. 2010].
Do rising temperatures indicate that there is climate change, and if so, can the change be attributed to human
action? These two problems are known as the problems of detection and attribution. The Intergovernmental Panel
on Climate Change (IPCC) defines these as follows:
Detection of change is defined as the process of demonstrating that climate or a system affected by climate has
changed in some defined statistical sense without providing a reason for that change. An identified change is
detected in observations if its likelihood of occurrence by chance due to internal variability alone is determined to
be small […]. Attribution is defined as ‘the process of evaluating the relative contributions of multiple causal
factors to a change or event with an assignment of statistical confidence.’ [IPCC 2013]

These definitions raise a host of issues. The root cause of the difficulties is the clause that climate change has been
detected only if an observed change in the climate is unlikely to be due to internal variability. Internal variability is
the phenomenon that the values of climate variables such as temperature and precipitation would change over
time due to the internal dynamics of the climate system even in the absence of a change in external conditions,
because of fluctuations in the frequency of storms, ocean currents, and so on.
Taken at face value, this definition of detection has the consequence that there cannot be internal climate change.
The ice ages, for instance, would not count as climate change if they occurred because of internal variability. This is
not only at odds with basic intuitions about climate and with the most common definitions of climate as a finite
distribution over a relatively short time period (where internal climate change is possible); it also leads to
difficulties with attribution: if detected climate change is ipso facto change not due to internal variability, then from
the very beginning it is excluded that particular factors (namely, internal climate dynamics) can lead to a change in
the climate, which seems to be an unfortunate conclusion.
For the case of the ice ages, many researchers would stress that internal variability is different from natural
variability. Since, say, orbital changes explain the ice ages, and orbital changes are natural but external, this is a
case of external climate change. While this move solves some of the problems, there remains the problem that
there is no generally accepted way to separate internal and external factors, and the same factor is sometimes
classified as internal and sometimes as external. For instance, glaciation processes are sometimes treated as
internal factors and sometimes as prescribed external factors. Likewise, sometimes the biosphere is treated as an
external factor, but sometimes it is also internally modelled and treated as an internal factor. One could even go so
far to ask whether human activity is an external forcing on the climate system or an internally-generated Earth
system process. Research studies usually treat human activity as an external forcing, but it could consistently be
argued that human activities are an internal dynamical process. The appropriate definition simply depends on the
research question of interest. For a discussion of these issues, see Katzav and Parker [2018].

The effects of internal variability are present on all timescales, from the sub-daily fluctuations experienced as
weather to the long-term changes due to cycles of glaciation. Since internal variability stems from processes in a
highly complex nonlinear system, it is also unlikely that the statistical properties of internal variability are constant
over time, which further compounds methodological difficulties. State-of-the-art climate models run with constant
forcing show significant disagreements both on the magnitude of internal variability and the timescale of
variations. (On http://www.climate-lab-book.ac.uk/2013/variable-variability/#more-1321 the reader finds a plot
showing the internal variability of all CMIP5 models. The plot indicates that models exhibit significantly different
internal variability, leaving considerable uncertainty.) The model must be deemed to simulate pre-industrial
climate (including variability) sufficiently well before it can be used for such detection and attribution studies, but
we do not have thousands of years of detailed observations upon which to base that judgement. Estimates of
internal variability in the climate system are produced from climate models themselves [Hegerl et al. 2010],
leading to potential circularity. This underscores the difficulties in making attribution statements based on the
above definition, which recognises an observed change as climate change only if is unlikely to be due to internal
variability.
Since the IPCC’s definitions are widely used by climate scientists, the discussion about detection and attribution in
the remainder of this section is based on these definitions. Detection relies on statistical tests, and detection
studies are often phrased in terms of the likelihood of a particular event or sequence of events happening in the
absence of climate change. In practice, the challenge is to define an appropriate null hypothesis (the expected
behaviour of the system in the absence of changing external influences), against which the observed outcomes can
be tested. Because the climate system is a dynamical system with processes and feedbacks operating on all scales,
this is a non-trivial exercise. An indication of the importance of the null hypothesis is given by the results of Cohn
and Lins [2005], who compare the same data against alternate null hypotheses, with results differing by 25 orders
of magnitude of significance! This does not in itself show that either null is more appropriate, but it demonstrates
the sensitivity of the result to the null hypothesis chosen. This, in turn, underscores the importance of the choice of
null hypothesis and the difficulty of making any such choice if the underlying processes are poorly understood.

In practice, the best available null hypothesis is often the best available model of the behaviour of the climate
system, including internal variability, which for most climate variables usually means a state of the art GCM. This
model is then used to perform long control runs with constant forcings in order to quantify the internal variability of
the model (see discussion above). Climate change is then said to have been detected when the observed values fall
outside a predefined range of the internal variability of the model. The difficulty with this method is that there is
no single “best” model to choose: many such models exist, they are similarly well developed, but, as noted above,
they have appreciably different patterns of internal variability.
The differences between different models are relatively unimportant for the clearest detection results such as
recent increases in global mean temperature. Here, as stressed by Parker [2010], detection is robust across
different models (for a discussion of robustness see Section 6), and, moreover, there is a variety of different pieces
of evidence all pointing to the conclusion that the global mean temperature has increased beyond that which can
be attributed to internal variability. However, the issues of which null hypothesis to use and how to quantify
internal variability, can be important for the detection of subtler local climate change.

If climate change has been detected, then the question of attribution arises. This might be an attribution of any
particular change (either a direct climatic change such as increased global mean temperature, or an impact such as
the area burnt by forest fires) to any identified cause (such as increased CO 2 in the atmosphere, volcanic eruptions,
or human population density). Where an impact is considered, a two-step or multi-step approach may be
appropriate. An example of this, taken from the IPCC Good Practice Guidance paper [Hegerl et al. 2010], is the
attribution of coral reef calcification impacts to rising CO 2 levels, in which an intermediate stage is used by first
attributing changes in the carbonate ion concentration to rising CO 2 levels, then attributing calcification processes
to changes in the carbonate ion concentration. This also illustrates the need for a clear understanding of the
physical mechanisms involved, in order to perform a reliable multi-step attribution in the presence of many
potential confounding factors.
In the interpretation of attribution results, in particular those framed as a question of whether human activity has
influenced a particular climatic change or event, there is a tendency to focus on whether the confidence interval of
the estimated anthropogenic effect crosses zero. The absence of such a crossing indicates that change is likely to be
due to human factors. This results in conservative attribution statements, but it reflects the focus of the present
debate where, in the eyes of the public and media, “attribution” is often understood as confidence in ruling out
non-human factors, rather than as giving a best estimate or relative contributions of different factors.

Statistical analysis quantifies the strength of the relationship, given the simplifying assumptions of the attribution
framework, but the level of confidence in the simplifying assumptions must be assessed outside that framework.
This level of confidence is standardised by the IPCC into discrete (though subjective) categories (“very high”,
“high”, and so forth.), which aim to take account of the process knowledge, data limitations, adequacy of models
used, and the presence of potential confounding factors. The conclusion that is reached will then have a form
similar to the IPCC’s headline attribution statement:

It is extremely likely [³95% probability] that more than half of the observed increase in global average surface
temperature from 1951 to 2010 was caused by the anthropogenic increase in greenhouse gas concentrations and
other anthropogenic forcings together. [IPCC 2013; Summary for Policymakers, section D.3].
One attribution method is optimal fingerprinting. The method seeks to define a spatio-temporal pattern of change
(fingerprint) associated with each potential driver (such as the effect of greenhouse gases or of changes in solar
radiation), normalised relative to the internal variability, and then perform a statistical regression of observed data
with respect to linear combinations of these patterns. The residual variability after observations have been
attributed to each factor should then be consistent with the internal variability; if not, this suggests that an
important source of variability remains unaccounted for. Parker [2010] notes that fingerprint studies rely on
several assumptions. Chief among them is linearity, that is, that the response of the climate system when several
forcing factors are present is equal to a linear combination of the effects of the forcings. Because the climate system
is nonlinear, this is clearly a source of methodological difficulty, although for global-scale responses (in contrast to
regional-scale responses) additivity has been shown to be a good approximation.
Levels of confidence in these attribution statements are primarily dependent on physical understanding of the
processes involved. Where there is a clear, simple, well-understood mechanism, there should be greater
confidence in the statistical result; where the mechanisms are loose, multi-factored or multi-step, or where a
complex model is used as an intermediary, confidence is correspondingly lower. The Guidance Paper cautions
that,

Where models are used in attribution, a model’s ability to properly represent the relevant causal link should be
assessed. This should include an assessment of model biases and the model’s ability to capture the relevant
processes and scales of interest. [Hegerl 2010, 5]
As Parker [2010] argues, there is also higher confidence in attribution results when the results are robust and there
is a variety of evidence. For instance, the finding that late twentieth-century temperature increase was mainly
caused by greenhouse gas forcing is found to be robust given a wide range of different models, different analysis
techniques, and different forcings; and there is a variety of evidence all of which supports this claim. Thus our
confidence that greenhouse gases explain global warming is high. (For further useful extended discussion of
detection and attribution methods in climate science, see pages 872-878 of IPCC [2013] and in the Good Practice
Guidance paper by Hegerl et al. [2010], and for a discussion of how such hypotheses are tested see Katzav [2013].)

In addition to the large-scale attribution of climate change, attribution of the degree to which individual weather
events have become either more likely or more extreme as a result of increasing atmospheric greenhouse gas
concentrations is now common. It has a particular public interest as it is perceived as a way both to communicate
that climate impacts are happening already, perhaps quantifying risk numerically to price insurance, and offering a
motivation for climate mitigation. There is therefore also an incentive to conduct these studies quickly, to inform
timely news articles, and some groups have formed to respond quickly to reports of extreme weather and conduct
attribution studies immediately. This relies on the availability of data, may suffer from unclear definitions of
exactly what category of event is being analysed, and is open to criticism for publicity prior to peer review. There
are also statistical implications of choosing to analyse only those events which have happened and not those that
did not happen. For a discussion of event attribution see Lloyd and Oreskes [2019] and Lusk [2017].

5. Confirmation and Predictive Power

Two questions arise in connection with models: how are models confirmed and what is their predictive
power? Confirmation concerns the question of whether, and to what degree, a specific model is supported by the
data. Lloyd [2009] argues that many climate models are confirmed by past data. Parker [2009] objects to this
claim. She argues that the idea that climate models per se are confirmed cannot be seriously entertained because all
climate models are known to be wrong and empirically inadequate. Parker urges a shift in thinking from
confirmation to adequacy for purpose: models can only be found to be adequate for specific purposes, but they
cannot be confirmed wholesale. For example, one might claim that a particular climate model adequately predicts
the global temperature increase that will occur by 2100 (when run from particular initial conditions and relative to
a particular emission scenario). Yet, at the same time, one might hold that the predictions of global mean
precipitation by 2100 by the same model cannot be trusted.
Katzav [2014] cautions that adequacy for purpose assessments are of limited use. He claims that these assessments
are typically unachievable because it is far from clear which of the model’s observable implications can possibly be
used to show that the model is adequate for the purpose. Instead, he argues that climate models can at best be
confirmed as providing a range of possible futures. Katzav is right to stress that adequacy for purpose assessments
are more difficult than appears at first sight. But the methodology of adequacy for purpose cannot be dismissed
wholesale; in fact, it is used successfully across the sciences (for example, when ideal gas models are confirmed to
be useful for particular purposes). Whether or not adequacy for purpose assessment is possible depends on the
case at hand.

If one finds that one model predicts specific variables well and another model doesn’t, then one would like to know
the reasons why the first model is successful and the second not. Lenhard and Winsberg [2010] argue that this is
often very difficult, if not impossible: For complex climate models a strong version of confirmation holism makes it
impossible to tell where the failures and successes of climate models lie. In particular, they claim that it is
impossible to assess the merits and problems of sub-models and the parts of models. There is a question, though,
whether this confirmation holism affects all models and whether it is here to stay. Complex models have different
modules for the atmosphere, the ocean, and ice. These modules can be run individually and also together. The aim
of the many new Model Intercomparison Projects (MIPs) is, by comparing individual and combined runs, to obtain
an understanding of the performance and physical merits of separate modules, which it is hoped will identify areas
for improvement and eventually result in better performance of the entire model.
Another problem concerns the use of data in the construction of models. The values of model parameters are often
estimated using observations, a process known as calibration. For example, the magnitude of the aerosol forcing is
sometimes estimated from data. When data have been used for calibration, the question arises whether the same
data can be used again to confirm the model. If data are used for confirmation that have not already been used for
calibration, they are use-novel. If data are used for both calibration and confirmation, this is referred to as double-
counting.
Scientists and philosophers alike have argued that double-counting is illegitimate and that data have to be use-
novel to be confirmatory [Lloyd 2010; Shackley et al. 1998; Worrall 2010]. Steele and Werndl [2013] oppose this
conclusion and argue that on Bayesian and relative-likelihood accounts of confirmation double-counting is
legitimate. Furthermore, Steele and Werndl [2015] argue that model selection theory presents a more nuanced
picture of the use of data than the commonly endorsed positions. Frisch [2015] cautions that Bayesian as well as
other inductive logics can be applied in better and worse ways to real problems such as climate prediction. Nothing
in the logic prevents facts from being misinterpreted and their confirmatory power exaggerated (as in ‘the problem
of old evidence’ which Frisch [2015] discusses). This is certainly a point worth emphasising. Indeed, Steele and
Werndl [2013] stress that the same data cannot inform a prior probability for a hypothesis and also further
(dis)confirm the hypothesis. But they do not address all the potential pitfalls in applying Bayesian or other logics to
the climate and other settings. Their argument must be understood as a limited one: there is no univocal logical
prohibition against the same data serving for calibration and confirmation. As far as non-Bayesian methods of
model selection goes, there are two cases. First, there are methods such as cross-validation where the data are
required to be use-novel. For cross-validation, the data are split up into two groups: the first group is used for
calibration and the second for confirmation. Second, there are the methods such as the Akaike Information
Criterion for which the data need not be use-novel, although information criteria methods are hard to apply in
practice to climate models because the number of degrees of freedom is poorly defined.

This brings us to the second issue: prediction. In the climate context this is typically framed as the issue
of projection. ‘Projection’ is a technical term in the climate modelling literature and refers to a prediction that is
conditional on a particular forcing scenario and a particular initial conditions ensemble. The forcing scenario is
specified either by the amount of greenhouse gas emissions and aerosols added to the atmosphere or directly by
their atmospheric concentrations, and these in turn depend on future socioeconomic and technological
developments.
Much research these days is undertaken with the aim of generating projections about the actual future evolution of
the Earth system under a particular emission scenario, upon which policies are made and real-life decisions are
taken. In these cases, it is necessary to quantify and understand how good those projections are likely to be. It is
doubtful that this question can be answered along traditional lines. One such line would be to refer to the
confirmation of a model against historical data (Chapter 9 of IPCC [2013] discusses model evaluation in detail) and
argue that the ability of a model to successfully reproduce historical data should give us confidence that it will
perform well in the future too. It is unclear at best whether this is a viable answer. The problem is that climate
projections for high forcing scenarios take the system well outside any previously experienced state, and at
least prima facie there is no reason to assume that success in low forcing contexts is a guide to success in high-
forcing contexts; for example, a model calibrated on data from a world with the Arctic Sea covered in ice might no
longer perform well when the sea ice is completely melted and the relevant dynamical processes are quite different.
For this reason, calibration to past data has at most limited relevance for the assessment of a model’s predictive
success [Oreskes et al. 1994; Stainforth et al. 2007a, 2007b, Steele and Werndl 2013].
This brings into focus the fact that there is no general answer to the question of the trustworthiness of model
outputs. There is widespread consensus that predictions are better for longer time averages, larger spatial
averages, low specificity and better physical understanding; and, all other things being equal, shorter lead times
(nearer prediction horizons) are easier to predict than longer ones. Global mean temperature trends are
considered trustworthy, and it is generally accepted that the observed upward trend will continue [Oreskes 2007],
although the basis of this confidence is usually a physical understanding of the greenhouse effect with which the
models are consistent, rather than a direct reliance on the output of models themselves. A 2013 IPCC report [IPCC
2013, Summary for Policymakers, section D.1] professes that modelled surface temperature patterns and trends
are trustworthy on the global and continental scale, but, even in making this statement, assigns a probability of at
least 66% (‘likely’) to the range within with 90% of model outcomes fall. In plainer terms, this is an expert-assigned
probability of at least tens of percent that the models are substantially wrong even about global mean temperature.
There still are interesting questions about the epistemic grounds on which such assertions are made (and we
return to them in the next section). A harder problem, however, concerns the use of models as providers of detailed
information about the future local climate. The United Kingdom Climate Impacts Programme produces projections
that aim to make high-resolution probabilistic projections of the local climate up to the end of the century, and
similar projects are run in many other countries [Thompson et al. 2016]. The Programme’s set of projections
known as UKCP09 [Sexton et al. 2012, Sexton and Murphy 2012] produces projections of the climate up to 2100
based on HadCM3, a global climate model developed at the UK Met Office Hadley Centre. Probabilities are given
for events on a 25km grid for finely defined specific events such as changes in the temperature of the warmest day
in summer, the precipitation of the wettest day in winter, or the change in summer-mean cloud amount, with
projections blocked into overlapping thirty-year segments which extend to 2100. It is projected, for instance, that
under a medium emission scenario the probability for a 20-30% reduction in summer mean precipitation in
central London in 2080 is 0.5. There is a question of whether these projections are trustworthy and policy relevant.
Frigg et al. urge caution on grounds that many of the UKCP09’s foundational assumptions seem to be questionable
[2013, 2015] and that structural model error may have significant repercussions on small scales [2014]. Winsberg
[2018] and Winsberg and Goodwin [2016] criticise these cautionary arguments as overstating the limitations of
such projections. In 2019, the Programme launched a new set of projections, known as UKCP18
(https://www.metoffice.gov.uk/research/collaboration/ukcp). It is an open question whether these projections are
open to the same objections, and, if so, how severe the limitations are.

6. Understanding and Quantifying Uncertainty

Uncertainty features prominently in discussions about climate models, and yet is a concept that is poorly
understood and that raises many difficult questions. In most general terms, uncertainty is a lack of knowledge. The
first challenge is to circumscribe more precisely what is meant by ‘uncertainty’ and what the sources of uncertainty
are. A number of proposals have been made, but the discussion is still in a ‘pre-paradigmatic’ phase. Smith and
Stern [2011] identify four relevant varieties of uncertainty: imprecision, ambiguity, intractability and
indeterminacy. Spiegelhalter and Riesch [2011] consider a five-level structure with three within-model levels-
event, parameter and model uncertainty-and two extra-model levels concerning acknowledged and unknown
inadequacies in the modelling process. Wilby and Dessai [2010] discuss the issue with reference to what they call
the cascade of uncertainty, studying how uncertainties magnify as one goes from assumptions about future global
emissions of greenhouse gases to the implications of these for local adaption. Petersen [2012, Chapters 3 and 6]
introduces a so-called uncertainty matrix listing the sources of uncertainty in the vertical and the sorts of
uncertainty in the horizontal direction. Lahsen [2005] looks at the issue from a science studies point of view and
discusses the distribution of uncertainty as a function of the distance from the site of knowledge production. And
these are but a few of the many proposals.

The next problem is the one of measuring and quantifying uncertainty in climate predictions. Among the
approaches that have been devised in response to this challenge, ensemble methods occupy centre stage. Current
estimates of climate sensitivity and increase in global mean temperature under various emission scenarios, for
instance, include information derived from ensembles containing multiple climate models. Multi-model ensembles
are sets of several different models which differ in mathematical structure and physical content. Such an ensemble
is used to investigate how predictions of relevant climate variables vary (or do not vary) according to model
structure and assumptions. A special kind of multi-model ensemble is known as a “perturbed parameter
ensemble”. It contains models with the same mathematical structure in which particular parameters assume
different values, thereby effectively conducting a sensitivity analysis on a single model by systematically varying
some of the parameters and observing the effect on the outcomes. Early analyses such as the climateprediction.net
simulations and the UKCP09 results rely on perturbed parameter ensembles only, due to resource limitations;
international projects such as the Coupled Model Intercomparison Projects (CMIP) and the work that goes into the
IPCC assessments are based on multi-model ensembles containing different model structures. The reason to use
ensembles is the acknowledged uncertainties in individual models, which concerns both the model structure and
the values of parameters in the model. It is a common assumption that ensembles help understand the effects of
these uncertainties either by producing and identifying “robust” predictions, or by providing estimates of this
uncertainty about future climate change. (Parker [2013] provides an excellent discussion of ensemble methods and
the problems that attach to them.)

A model-result is robust if all or most models in the ensemble show the same result; for general discussion of
robustness analysis see Weisberg [2006]. If, for instance, all models in an ensemble show more than 4º increase in
global mean temperature by the end of the century when run under a specific emission scenario, this result is
robust across the specified ensemble. Does robustness justify increased confidence? Lloyd [2010, 2015] argues that
robustness arguments are powerful in connection with climate models and lend credibility at least to core claims
such as the claim that there was global warming in the 20th Century. Parker [2011], by contrast, reaches a more
sober conclusion: ‘When today’s climate models agree that an interesting hypothesis about future climate change is
true, it cannot be inferred […] that the hypothesis is likely to be true or that scientists’ confidence in the hypothesis
should be significantly increased or that a claim to have evidence for the hypothesis is now more secure’ [ ibid. 579].
One of the main problems is that if today’s models share the same technological constraints posed by today’s
computer architecture and understanding of the climate system, then they inevitably share some common errors.
Indeed, such common errors have been widely acknowledged (see, for instance, Knutti et al. [2010]) and studies
have demonstrated and discussed the lack of model independence [Bishop and Abramowitz 2013; Jun et al.
2008a; 2008b]. But if models are not independent, then there is a question about how much epistemic weight
agreement between them carries.
When ensembles do not yield robust predictions, then the spread of results within the ensemble is sometimes used
to estimate quantitatively the uncertainty of the outcome. There are two main approaches to this. The first
approach aims to translate the histogram of model results directly into a probability distribution: in effect, the
guiding principle is that the probability of an outcome is proportional to the fraction of models in the ensemble
which produce that result. The thinking behind this method seems to be to invoke some sort of frequentist
approach to probabilities. The appeal to frequentism presupposes that models can be treated as exchangeable
sources of information (in the sense that there is no reason to trust one ensemble member any more than any
other). However, as we have previously seen, the assumption that models are independent has been questioned.
There is a further problem: MMEs are ‘ensembles of opportunity’, grouping together existing models. Even the
best ensembles such as CMIP6 are not designed to systematically explore all possibilities. It is therefore not clear
why the frequency of ensemble projections should double as a guide to probability. The IPCC acknowledges this
limitation (see discussion in Chapter 12 of IPCC [2013]) and thus downgrade the assessed likelihood of ensemble-
derived ranges, deeming it only “likely” (³66%) that the real-world global mean temperature will fall within the
90% model range (for a discussion of this case see Thompson et al [2016]).

A more modest approach regards ensemble outputs as a guide to possibility rather than probability. In this view,
the spread of an ensemble presents the range of outcomes that cannot be ruled out. The bounds of this set of
results-often referred to as a ‘non-discountable envelope’-provide a lower bound of the uncertainty [Stainforth et
al. 2007b]. In this spirit Katzav [2014] argues that a focus on prediction is misguided and that models ought to be
used to show that particular scenarios are real possibilities.

While undoubtedly less committal than the probability approach, also non-discountable envelopes raise questions.
The first is the relation between non-discountability and possibility. Non-discountable results are ones that cannot
be ruled out. How is this judgment reached? Do results which cannot be ruled out indicate possibilities? If not,
what is their relevance for estimating lower bounds? And, could the model, if pushed more deliberately towards
“interesting” behaviours, actually make that envelope wider? Furthermore, it is important to keep in mind that the
envelope just represents some possibilities. Hence it does not indicate the complete range of possibilities, making
particular types of formalised decision-making procedures impossible. For a further discussion of these issues see
Betz [2009, 2010].
Finally, a number of authors emphasise the limitations of model-based methods (such as ensemble methods) and
submit that any realistic assessment of uncertainties will also have to rely on other factors, most notably expert
judgement. Petersen [2012, Chapter 4] outlines the approach of the Netherlands Environmental Assessment
Agency (PBL), which sees expert judgment and problem framings as essential components of uncertainty
assessment. Aspinall [2010] suggests using methods of structured expert elicitation.

In light of the issues raised above, how should uncertainty in climate science be communicated to decision-
makers? The most prominent framework for communicating uncertainty is the IPCC’s, which is used throughout
the Fifth Assessment Report (AR5), is explicated in the ‘Guidance Note for Lead Authors of the IPCC Fifth
Assessment Report on Consistent Treatment of Uncertainties’ and further explicated in [Mastrandrea et al. 2011].
The framework appeals to two measures for communicating uncertainty. The first, a qualitative ‘confidence’ scale,
depends on both the type of evidence and the degree of agreement amongst experts. The second measure is a
quantitative scale for representing statistical likelihoods (or more accurately, fuzzy likelihood intervals) for
relevant climate/economic variables. The following statement exemplifies the use of these two measures for
communicating uncertainty in AR5: ‘The global mean surface temperature change for the period 2016–2035
relative to 1986–2005 is similar for the four RCPs and will likely be in the range 0.3°C to 0.7°C (medium confidence).
[IPCC 2013] A discussion of this framework can be found in Adler and Hirsch Hadorn [2014], Budescu et al.
[2014], Mach et al. [2017], and Wüthrich [2017].
At this point, it should also be noted that the role of ethical and social values in relation to uncertainties in climate
science is controversially debated. Winsberg [2012] appeals to complex simulation modelling to argue that it is
infeasible for climate scientists to produce results that are not influenced by their ethical and social values. More
specifically, he argues that assignments of probabilities to hypotheses about future climate change are influenced
by ethical and social values because of the way these values come into play in the building and evaluating of climate
models. Parker [2014] contends that pragmatic factors rather than social or ethical values often play a role in
resolving these modelling choices. She further objects that Winsberg’s focus on precise probabilistic uncertainty
estimates is misguided; coarser estimates like those used by the IPCC better reflect the extent of uncertainty and
are less influenced by values. She concludes that Winsberg has exaggerated the influence of ethical and social
values here but suggests that a more traditional challenge to the value-free ideal of science fits the climate case.
Namely, one could argue that estimates of uncertainty are themselves always somewhat uncertain, and that the
decision to offer a particular estimate of uncertainty thus might appropriately involve value judgments [compare,
Douglas 2009].

7. Conceptualising Decisions Under Uncertainty

What is the appropriate reaction to climate change? How much should we mitigate? To what extent should we
adapt? And what form should adaptation take? Should we build larger water reserves? Should we adapt houses,
and our social infrastructure more generally, to a higher frequency of extreme weather events like droughts, heavy
rainfalls, floods, and heatwaves, as well as the increased incidence of extremely high sea levels or the more
frequent occurrence of particularly hot days are extreme weather events? The decisions that we make in response
to these questions have consequences affecting both individuals and groups at different places and times.
Moreover, the circumstances of many of these decisions involve uncertainty and disagreement that is sometimes
both severe and wide-ranging, concerning not only the state of the climate (as discussed above) and the broader
social consequences of any action or inaction on our part, but also the range of actions available to us and what
significance we should attach to their possible consequences. These considerations make climate decision-making
both important and hard. The stakes are high, and so too are the difficulties for standard decision theory—plenty of
reason for philosophical engagement with this particular application of decision theory.

Let us begin by looking at the actors in the climate domain and the kinds of decision problems that concern them.
When introducing decision theory, it is common to distinguish three main domains: individual decision theory
(which concerns the decision problem of a single agent who may be uncertain of her environment), game theory
(which focuses on cases of strategic interaction amongst rational agents), and social choice theory (which concerns
procedures by which a number of agents may ‘think’ and act collectively). All three realms are relevant to the
climate-change predicament, whether the concern is adapting to climate change or mitigating climate change or
both.

Determining the appropriate agential perspective and type of engagement between agents is important, because
otherwise decision-modelling efforts may be in vain. For instance, it may be futile to focus on the plight of
individual citizens when the power to affect change really lies with states. It may likewise be misguided to analyse
the prospects for a collective action on climate policy, if the supposed members of the group do not see themselves
as contributing to a shared decision that is good for the group as a whole. It would also be misleading to exclude
from an individual agent’s decision model the impact of others who perceive that they are acting in a strategic
environment. This is not, however, to recommend a narrow view of the role of decision models-that they must
always represent the decisions of agents as they see them, and can never be aspirational; the point is rather that we
should not employ decision models with particular agential framings in a naïve way.

Getting the agential perspective right is just the first step in framing a decision problem so that it presents
convincing reasons for action. There remains the task of representing the details of the decision problem from the
appropriate epistemic and evaluative perspective. Our focus is individual decision theory, for reasons of space, and
because most decision settings ultimately involve the decision of an individual, whether this be a single person or a
group acting as an individual.
The standard model of (individual) decision-making under uncertainty used by decision theorists derives from the
classic work of von Neumann and Morgenstern [1944] and Leonard Savage [1954]. It treats actions as functions
from possible states of the world to consequences, these being the complete outcomes of performing the action in
question in that state of the world. All uncertainty is taken to be uncertainty about the state of the world and is
quantified by a single probability function over the possible states, where the probabilities in question measure
either objective risk or the decision maker’s degrees of belief (or a combination of the two). The relative value of
consequences is represented by an interval-scaled utility function over these consequences. Decision-makers are
advised to choose the action with maximum expected utility (EU); where the EU for an action is the sum of the
probability-weighted utility of the possible consequences of the action.
It is our contention that this model is inadequate for many climate-oriented decisions, because it fails to properly
represent the multidimensional nature and severity of the uncertainty that decision-makers face. To begin with,
not all the uncertainty that climate decision-makers face is empirical uncertainty about the actual state of the
world (state uncertainty). There may be further empirical uncertainty about what options are available to them and
what are the consequences of exercising each option for each respective state (option uncertainty). In what follows
we use the term ‘empirical uncertainty’ to cover both state uncertainty and option uncertainty. Furthermore,
decision-makers face a non-empirical kind of uncertainty-ethical uncertainty-about what values to assign to
possible consequences.

Let us now turn to empirical uncertainty. As noted above, standard decision theory holds that all empirical
uncertainty can be represented by a probability function over the possible states of the world. There are two issues
here. The first is that confining all empirical uncertainty to the state space is rather unnatural for complex decision
problems such as those associated with climate change. In fact, decision models are less convoluted if we allow the
uncertainty about states to depend on the actions that might be taken (compare, Richard Jeffrey’s [1965] expected
utility theory), and if we also permit further uncertainty about what consequence will arise under each state, given
the action taken (an aspect of option uncertainty). For instance, consider a crude version of the mitigation decision
problem faced by the global planner: it may be useful to depict the decision problem with a state-space partition in
terms of possible increases in average global temperature over a given time period. In this case, our beliefs about
the states (how likely they each are) would be conditional on the mitigation option taken. Moreover, for each
respective mitigation option, the consequence arising in each of the states depends on further uncertain features of
the world, for instance the extent to which, on average, regional conditions would be favourable to food production
and whether social institutions would facilitate resilience in food production.

The second issue is that using a precise probability function to represent uncertainty about states (and
consequences) can misrepresent the severity of this uncertainty. For instance, even if one assumes that the
position of the scientific community may be reasonably well represented by a precise probability distribution over
the state space, conditional on the mitigation option, precise probabilities over the possible food productions and
other economic consequences, given this option and average global temperature rise, are less plausible. Note that
the global social planner’s mitigation decision problem is typically analysed in terms of a so-called Integrated
Assessment Model (IAM), which does indeed involve dependencies between mitigation strategies and both climate
and economic variables. There is some disparity in the representation of empirical uncertainty: Nordhaus’s [2008]
reliance on ‘best estimates’ for parameters like climate sensitivity can be compared with Stern’s [2007] use of
‘confidence intervals’. But these are relatively minor differences. Critics argue that all extant IAMs inadequately
represent the uncertainty surrounding projections of future wealth under the status quo and alternative mitigation
strategies [see Weitzman 2009, Frisch 2013, Stern 2013]. In particular, both Nordhaus [2008] and Stern [2007]
controversially assume increasing wealth over time (or positive consumption growth rate) even for the status quo
where nothing is done to mitigate climate change.
Popular among philosophers is the use of sets of probability functions to represent severe uncertainty surrounding
decision states/consequences, whether the uncertainty is due to evidential limitations or due to evidential/expert
disagreement. This is a minimal generalisation of the standard decision model, in the sense that probability
measures still feature: roughly, the more severe the uncertainty, the more probability measures over the space of
possibilities needed to conjointly represent the epistemic situation (see, for instance, Walley [1991]). For maximal
uncertainty all possibilities are on a par-they are effectively assigned probability [0, 1]. Indeed it is a strength of
the imprecise probability representation that it generalises the two extreme cases, that is, the precise probabilistic as
well as the possibilistic frameworks. (See Halpern [2003] for a thorough treatment of frameworks, both qualitative
and quantitative, for representing uncertainty.) In some contexts, it may be suitable to weight the possible
probability distributions in terms of plausibility (as required for some of the decision rules discussed below). The
weighting approach may in fact match the IPCC’s representation of the uncertainty surrounding decision-relevant
climate and economic variables. Indeed, an important question is whether and how the IPCC’s representation of
uncertainty can be translated into an imprecise probabilistic framework, as discussed here and in the next section.
An alternative to the aforementioned proposal is that the IPCC’s confidence and likelihood measures for relevant
variables should be combined to form an unweighted imprecise set of probability distributions, or even a precise
probability distribution, suitable for input into an appropriate decision model.
Decision makers face uncertainty not only about what will or could happen, but also about what value to attach to
these possibilities-in other words, they face ethical uncertainty. Such value or ethical uncertainty can have a
number of different sources. The most important ones arise in connection with judgments about how to distribute
the costs and benefits of mitigation and adaptation amongst different regions and countries, about how to take
account of persons whose existence depends on what actions are chosen now, and about the degree to which future
wellbeing should be discounted. (For discussion and debate about the ethical significance of various climate
outcomes, particularly at the level of global rather than regional or national justice, see the articles in Gardiner et
al.’s [2010] edited collection, Climate Ethics.) Of these, the latter has been the subject of the most debate, because of
the extent to which (the global planner’s) decisions about how drastically to cut carbon emissions are sensitive to
the discount rate used in evaluating the possible outcomes of doing so (as highlighted in Broome [2008]).
Discounting thus provides a good illustration of the importance of ethical uncertainty.
In many economic models, a discount rate is applied to a measure of total wellbeing at different points in time (the
‘pure rate of time preference’), with a positive rate implying that future wellbeing carries less weight in the
evaluations of options than present wellbeing. Note that the overall ‘social discount rate’ in economic models is the
sum of the pure rate of time preference and a second term pertaining to the discounting of goods or consumption
rather than wellbeing per se. See Broome [1992] and Parfit [1984] for helpful discussions of the reasons for
discounting goods that do not imply discounting wellbeing. (The consumption growth rate is an important
component of this second discounting term that is subject to empirical uncertainty, as discussed above; see Greaves
[2017] for an examination of all the assumptions underlying the ‘social discount rate’ and its role in the standard
economic method for evaluating policy options.) Many philosophers regard any pure discounting of future
wellbeing as completely unjustified from an objective point of view. This is not to deny that temporal location may
nonetheless correlate with features of the distribution of wellbeing that are in fact ethically significant. If people
will be better off in the future, for instance, it is reasonable to be less concerned about their interests than those of
the present generation, much as one might prioritise the less well-off within a single generation. But the mere fact
of a benefit occurring at a particular time cannot be relevant to its value, at least from an impartial perspective.
Economists do nonetheless often discount wellbeing in their policy-oriented models, although they disagree
considerably about what pure rate of time preference should be used. One view, exemplified by the Stern Review
and representing the impartial perspective described above, is that only a very small rate (in the order of 0.5%) is
justified, and this on the grounds of the small probability of the extinction of the human population. Other
economists, however, regard a partial rather than an impartial point of view more appropriate in their models. A
view along these lines, exemplified by Nordhaus [2007] and Arrow [1995a], is that the pure rate of time preference
should be determined by the preferences of current people. But typical derivations of average pure time
discounting from observed market behaviour are much higher than those used by Stern (around 3% by Nordhaus’s
estimate). Although the use of this data has been criticised for providing an inadequate measure of people’s
reasoned preferences (see, for example, Sen [1982], Drèze and Stern [1990], Broome [1992]), the point remains
that any plausible method for determining the current generation’s attitude to the wellbeing of future generations
is likely to yield a rate higher than that advocated by the Stern Review. To the extent that this debate about the
ethical basis for discounting remains unresolved, there will be ethical uncertainty about the discount rate in
climate policy decisions. This ethical uncertainty may be represented analogously to empirical uncertainty-by
replacing the standard precise utility function with a set of possible utility functions.

8. Managing Uncertainty
How should a decision-maker choose amongst the courses of action available to her when she must make the
choice under conditions of severe uncertainty? The problem that climate decision-makers face is that, in these
situations, the precise utility and probability values required by standard EU theory may not be readily available.

There are, broadly speaking, three possible responses to this problem.

(1) The decision-maker can simply bite the bullet and try to settle on precise probability and utility judgements for
the relevant contingencies. Orthodox decision theorists argue that rationality requires that decisions be made as if
they maximise the decision maker’s subjective expectation of benefit relative to her precise degrees of belief and
values. Broome [2012, 129] gives an unflinching defence of this approach: “The lack of firm probabilities is not a
reason to give up expected value theory […] Stick with expected value theory, since it is very well-founded, and do
your best with probabilities and values.” This approach may seem rather bold, not least in the context of
environmental decision making. Weitzman [2009], for instance, argues that whether or not one assigns non-
negligible probability to catastrophic climate consequences radically changes the assessment of mitigation options.
Moreover, in many circumstances there remains the question of how to follow Broome’s advice: How should the
decision-maker settle, in a non-arbitrary way, on a precise opinion on decision-relevant issues in the face of an
effectively ‘divided mind’? There are two interrelated strategies: she can deliberate further and/or aggregate
conflicting views. The former aims for convergence in opinion, while the latter aims for an acceptable compromise
in the face of persisting conflict. (For a discussion of deliberation see Fishkin and Luskin [2005]; for more on
aggregation see, for instance, Genest and Zidek [1986], Mongin [1995], Sen [1970], List and Puppe [2009]. There
is a comparatively small formal literature on deliberation, a seminal contribution being Lehrer and Wagner’s
[1981] model for updating probabilistic beliefs.)

(2) The decision-maker can try to delay making a decision, or at least postpone parts of it, in the hope that her
uncertainty will become manageable as more information becomes available, or as disagreements resolve
themselves through a change in attitudes. The basic motive for delaying a decision is to maintain flexibility at zero
cost (see Koopmans [1962], Kreps and Porteus [1978], Arrow [1995b]). Suppose that we must decide between
building a cheap but low sea wall or a high, but expensive, one, and that the relative desirability of these two
courses of action depends on unknown factors, such as the extent to which sea levels will rise. In this case it would
be sensible to consider building a low wall first but leave open the possibility of raising it in the future. If this can
be done at no additional cost, then it is clearly the best option. In many adaptation scenarios, the analogue of the
‘low sea wall’ may in fact be social-institutional measures that enable a delayed response to climate change,
whatever the details of this change turn out to be. In many cases, however, the prospect of cost-free postponement
of a decision (or part thereof) is simply a mirage, since delay often decreases rather than increases opportunities
due to changes in the background environment. This is often true for climate-change adaptation decisions, not to
mention mitigation decisions.
(3) The decision-maker can employ a different decision rule to that prescribed by EU theory; one that is much less
demanding in terms of the information it requires. A great many different proposals for such rules exist in the
literature, involving more or less radical departures from the orthodox theory and varying in the informational
demands they make. It should be noted from the outset that there is one widely-agreed rationality constraint on
these non-standard decision rules: ‘(EU)-dominated options’ are not admissible choices, that is, if an option has
lower expected utility than another option according to all permissible pairs of probability and utility functions,
then the former dominated option is not an admissible choice. This is a relatively minimal constraint, but it may
well yield a unique choice of action in some decision scenarios. In such cases, the severe uncertainty is not in fact
decision relevant. For example, it may be the case that, from the global planner’s perspective, a given mitigation
option is better than continuing with business as usual, whatever the uncertain details of the climate system. This
is even more plausible to the extent that the mitigation option counts as a ‘win-win’ strategy [Maslin and Austin
2012], that is, to the extent that it has other positive impacts, say, on air quality or energy security, regardless of
mitigation results. In many more fine-grained or otherwise difficult decision contexts, however, the non-EU-
dominance constraint may exclude only a few of the available options as choice-worthy.

A consideration that is often appealed to in order to further discriminate between options is caution. Indeed, this is
an important facet of the popular but ill-defined Precautionary Principle. (The Precautionary Principle is referred
to in the IPCC [2014b] ARC-5 WGII report. See, for instance, Gardiner [2006] and Steele [2006] for discussion of
what the Precautionary Principle does/could stand for.) Cautious decision rules give more weight to the ‘down-
side’ risks; the possible negative implications of a choice of action. The Maxmin-EU rule, for instance, recommends
picking the action with greatest minimum expected utility (see Gilboa and Schmeidler [1989], Walley [1991]). The
rule is simple to use, but arguably much too cautious, paying no attention at all to the full spread of possible
expected utilities. The α-Maxmin rule, in contrast, recommends taking the action with the greatest α-weighted sum
of the minimum and maximum expected utilities associated with it. The relative weights for the minimum and
maximum expected utilities can be thought of as reflecting either the decision maker’s pessimism in the face of
uncertainty or else their degree of caution (see Binmore [2009]). (For a comprehensive survey of non-standard
decision theories for handling severe uncertainty in the economics literature, see Gilboa and Marinacci [2012]. For
applications to climate policy see Heal and Millner [2014])
A more informationally-demanding set of rules are those that draw on considerations
of confidence and/or reliability. The thought here is that an agent is more or less confident about the various
probability and utility functions that characterise her uncertainty. For instance, when the estimates derive from
different models or experts, the decision maker may regard some models as better corroborated by available
evidence than others or else some experts as more reliable than others in their judgments. In these cases, it is
reasonable, ceteris paribus, to favour actions of which you are more confident that they will have beneficial
consequences. One (rather sophisticated) way of doing this is to weight each of the expected utilities associated
with an action in accordance with how confident you are about the judgements supporting them and then choose
the action with the maximum confidence-weighted expected utility (see Klibanoff et al. [2005]). This rule is not
very different from maximising expected utility and indeed one could regard confidence weighting as an
aggregation technique rather than an alternative decision rule. But considerations of confidence may be appealed
to even when precise confidence weights cannot be provided. Gärdenfors and Sahlin [1982/ 1988], for instance,
suggest simply excluding from consideration any estimates that fall below a reliability threshold and then picking
cautiously from the remainder. Similarly, Hill [2013] uses an ordinal measure of confidence that allows for stake-
sensitive thresholds of reliability that can then be combined with varying levels of caution. This rule has the
advantage of allowing decision-makers to draw on the confidence grading of scientific claims adopted by the IPCC
(see Bradley et al [2017]).
One might finally distinguish decision rules that are cautious in a slightly different way-that compare options in
terms of ‘robustness’ to uncertainty, relative to a problem-specific satisfactory level of expected utility. Better
options are those that are more assured of having an expected utility that is good enough or regret-free, in the face
of uncertainty. The ‘information-gap theory’ developed by Ben-Haim [2001] provides one formalisation of this
basic idea that has proved popular in environmental management theory. Another prominent approach to robust
decision-making is that developed by Lempert, Popper and Bankes [2003]. These two frameworks are compared in
Hall et al. [2012]. Recall that the uncertainty in question may be multi-faceted, concerning probabilities of
states/outcomes, or values of final outcomes. Most decision rules that appeal to robustness assume that a best
estimate for the relevant variables is available, and then consider deviations away from this estimate. A robust
option is one that has a satisfactory expected utility relative to a class of estimates that deviate from the best one to
some degree; the wider the class in question, the more robust the option. Much depends on what expected utility
level is deemed satisfactory. For mitigation decision making, one salient satisfactory level of expected utility is that
associated with a 50% chance of average global temperature rise of 2 degrees Celsius or less. Note that one may
otherwise interpret any such mitigation temperature target in a different way, namely as a constraint on what
counts as a feasible option. In other words, mitigation options that do not meet the target are simply prohibited
options, not suitable for consideration. For adaptation decisions, the satisfactory level would depend on local
context, but roughly speaking, robust options are those that yield reasonable outcomes for all the inopportune
climate scenarios that have non-negligible probability given some range of uncertainty. These are plausibly
adaptation options that focus on resilience to any and all of the aforesaid climate scenarios, perhaps via the
development of social institutions that can coordinate responses to variability and change. (Robust decision-
making is endorsed, for instance, by Dessai et al. [2009] and Wilby and Dessai [2010], who indeed associate this
kind of decision rule with resilience strategies. See also Linkov and others [2014] for discussion of resilience
strategies vis-à-vis risk management.)

9. Conclusion
This article reviewed, from a philosophy of science perspective, issues and questions that arise in connection with
climate science. Most of these issues are the subject matter of ongoing research, and they indeed deserve further
attention. Rather than repeating these points, we would like to mention a topic that has not received the attention
that it deserves: the epistemic significance of consensus in the acceptance of results. As the controversy over the
Cook et al. [2013] paper shows, many people do seem to think that the level of expert consensus is an important
reason to believe in climate change given that they themselves are not expert; and conversely, attacking the
consensus and sowing doubt is a classic tactic of the other side. The role of consensus in the context of climate
change deserves more attention than it has received hitherto, but for some discussions about consensus see
(Inmaculada de Melo-Martín, Kristen Intemann, 2014).

10. Glossary
Attribution (of climate change): The process of evaluating the relative contributions of multiple causal factors to a
change or event with an assignment of statistical confidence.
Boundary conditions: Values for any variable which affect the system but which are not directly output by the
calculations.
 Calibration: The process of estimating values of model parameters which are most consistent with observations.
Climate model: A representation of certain aspects of the climate system.
Detection (of climate change): The process of demonstrating that climate or a system affected by climate has
changed in some defined statistical sense without providing a reason for that change.
Double counting: The use of data for both calibration and confirmation.
Expected utility (for an action): The sum of the probability-weighted utility of the possible consequences of the
action.
External conditions (of the climate system): Conditions that influence the state of the Earth such as the amount of
energy received from the sun.
Initial conditions: A mathematical descriptions of the state of the climate system at the beginning of the period being
simulated.
Internal variability: The phenomenon that climate variables such as temperature and precipitation would change
over time due to the internal dynamics of the climate system even in the absence of changing external conditions.
Null hypothesis: The expected behaviour of the climate system in the absence of changing external influences.
Projection: The prediction of a climate model that is conditional on a certain forcing scenario.
Proxy data: The data for climate variables that derived from observing natural phenomena such as tree rings, ice
cores and ocean sediments.
Robustness (of a result): A result is robust if separate (ideally independent) models or lines of evidence lead to the
same conclusion.
Use novel data: Data that are used for confirmation and have not been used for calibration.

11. References and Further Reading

 Adler C. E. and G. Hirsch Hadorn. (2014). The IPCC and treatment of uncertainties: topics and sources of dissensus. Wiley
Interdisciplinary Reviews: Climate Change 5.5, 663-676.
 Arrow K. J. (1995b). A Note on Freedom and Flexibility. Choice, Welfare and Development. (eds. K. Basu, P. Pattanaik, and K.
Suzumura), 7-15. Oxford: Oxford University Press.
 Arrow K. J. (1995a). ‘Discounting Climate Change: Planning for an Uncertain Future. Lecture given at Institut d’Économie
Industrielle, Université des Sciences Sociales, Toulouse.’ <http://idei.fr/doc/conf/annual/paper_1995.pdf>
 Aspinall W. (2010). A route to more tractable expert advice. Nature 463, 294-295.
 Ben-Haim Y. (2001). Information-Gap Theory: Decisions Under Severe Uncertainty, 330 pp. London: Academic Press.
 Betz G. (2009). What range of future scenarios should climate policy be based on? Modal falsificationism and its
limitations. Philosophia Naturalis 46, 133-158.
 Betz G. (2010). What’s the worst case?. Analyse und Kritik 32, 87-106.
 Binmore K. (2009). Rational Decisions, 216 pp. Princeton, NJ: Princeton University Press.
 Bishop C. H. and G. Abramowitz. (2013). Climate model dependence and the replicate Earth paradigm. Climate Dynamics 41,
885-900.
 Bradley, R, Helgeson, C. and B. Hill (2017). Climate Change Assessments: Confidence, Probability and Decision, Philosophy of
Science 84(3): 500-522.
 Bradley, R, Helgeson, C. and B. Hill (2018). Combining Probability with Qualitative Degree-of-Certainty Assessment. Climatic
Change 149 (3-4): 517-525,
 Broome J. (2012). Climate Matters: Ethics in a Warming World, 192 pp. New York: Norton.
 Broome J. (1992). Counting the Cost of Global Warming, 147 pp. Cambridge: The White Horse Press.
 Broome J. (2008). The Ethics of Climate Change. Scientific American 298, 96-102.
 Budescu, D. V., H. Por, S. B. Broomell and M. Smithson. (2014). The interpretation of IPCC probabilistic statements around the
world. Nature Climate Change 4, 508-512.
 Cohn T. A. and H. F. Lins. (2005). Nature’s style: naturally trendy. Geophysical Research Letters 32, L23402.
 Cook J. et al. (2013). Quantifying the consensus on the anthropogenic global warming in the scientific literature. Environmental
Research Letters 8, 1-7.
 Daron J. D. and D. Stainforth. (2013). On predicting climate under climate change. Environmental Research Letters 8, 1-8.
 de Melo-Martín I., and K. Intemann (2014). Who’s afraid of dissent? Addressing concerns about undermining scientific
consensus in public policy developments. Perspectives on Science 22.4, 593-615.
 Dessai S. et al. (2009). Do We Need Better Predictions to Adapt to a Changing Climate? Eos 90.13, 111-112.
 Dessler A. (2011). Introduction to Modern Climate Change. Cambridge: Cambridge University Press.
 Drèze J., and Stern, N. (1990). Policy reform, shadow prices, and market prices. Journal of Public Economics 42.1, 1-45.
 Douglas H. (2009). Science, Policy, and the Value-Free Ideal. Pittsburgh: Pittsburgh University Press.
 Fishkin J. S., and R. C. Luskin. (2005). Experimenting with a Democratic Ideal: Deliberative Polling and Public Opinion. Acta
Politica 40, 284-298.
 Frank D., J. Esper, E. Zorita and R. Wilson. (2010). A noodle, hockey stick, and spaghetti plate: A perspective on high-resolution
paleoclimatology. Wiley Interdisciplinary Reviews: Climate Change 1.4, 507-516.
 Frigg R. P., D. A. Stainforth and L. A. Smith. (2013). The Myopia of Imperfect Climate Models: The Case of UKCP09. Philosophy
of Science 80.5, 886-897.
 Frigg R. P., D. A. Stainforth and L. A. Smith. (2015). An Assessment of the Foundational Assumptions in High-Resolution
Climate Projections: The Case of UKCP09 2015, draft under review.
 Frigg R. P., S. Bradley, H. Du and L. A. Smith. (2014a). Laplace’s Demon and the Adventures of His Apprentices. Philosophy of
Science 81.1, 31-59.
 Frisch M. (2013). Modeling Climate Policies: A Critical Look at Integrated Assessment Models. Philosophy and Technology 26,
117-137.
 Frisch, M. (2015). Tuning climate models, predictivism, and the problem of old evidence. European Journal for Philosophy of
Science 5.2, 171-190.
 Gärdenfors P. and N.-E. Sahlin. [1982] (1988). Unreliable probabilities, risk taking, and decision making. Decision, Probability and
Utility, (eds. P. Gärdenfors and N.-E. Sahlin), 313-334. Cambridge: Cambridge University Press.
 Gardiner S. (2006). A Core Precautionary Principle. The Journal of Political Philosophy 14.1, 33-60.
 Gardiner S., S. Caney, D. Jamieson, H. Shue (2010). Climate Ethics: Essential Readings. Oxford: Oxford University Press
 Genest C. and J. V. Zidek. (1986). Combining Probability Distributions: A Critique and Annotated Bibliography. Statistical
Science 1.1, 113-135.
 Gilboa I. and M. Marinacci. (2012). Ambiguity and the Bayesian Paradigm. Advances in Economics and Econometrics: Theory and
Applications, Tenth World Congress of the Econometric Society (eds. D. Acemoglu, M. Arellano and E. Dekel), 179-242 Cambridge:
Cambridge University Press.
 Gilboa I. and D. Schmeidler. (1989). Maxmin expected utility with non-unique prior. Journal of Mathematical Economics 18, 141-
153.
 Greaves, H. (2017). Discounting for public policy: A survey. Economics and Philosophy 33.3, 391-439.
 Hall J. W., Lempert, R. J., Keller, K., Hackbarth, A., Mijere, C., McInerney, D. J. (2012). Robust Climate Policies Under
Uncertainty: A Comparison of Robust Decision-Making and Info-Gap Methods. Risk Analysis 32.10, 1657-1672.
 Halpern J. Y. (2003). Reasoning About Uncertainty, 483 pp. Cambridge, MA: MIT Press.
 Heal. G. and A. Millner (2014) Uncertainty and Decision Making in Climate Change Economics. Review of Environmental
Economics and Policy 8:120-137.
 Hegerl G. C., O. Hoegh-Guldberg, G. Casassa, M. P. Hoerling, R. S. Kovats, C. Parmesan, D. W. Pierce, P. A. Stott. (2010).
Good Practice Guidance Paper on Detection and Attribution Related to Anthropogenic Climate Change. Meeting Report of the
Intergovernmental Panel on Climate Change Expert Meeting on Detection and Attribution of Anthropogenic Climate Change (eds. T. F.
Stocker, C. B. Field, D. Qin, V. Barros, G.-K. Plattner, M. Tignor, P. M. Midgley and K. L. Ebi. Bern). Switzerland: IPCC Working
Group I Technical Support Unit, University of Bern.
 Held I. M. (2005). The Gap between Simulation and Understanding in Climate Modeling. Bulletin of the American Meteorological
Society 80, 1609-1614.
 Hill B. (2013). Confidence and Decision. Games and Economic Behavior 82, 675-692.
 Hulme M., S. Dessai, I. Lorenzoni and D. Nelson. (2009). Unstable Climates: exploring the statistical and social constructions of
climate. Geoforum 40, 197-206.
 IPCC. (2013). Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge and New York: Cambridge University Press.
 IPCC. (2014). Climate Change 2014: Impacts, Adaptation, and Vulnerability. Contribution of Working Group II to the Fifth Assessment
Report of the Intergovernmental Panel on Climate Change. Cambridge and New York: Cambridge University Press.
 Jeffrey R. (1965). The Logic of Decision, 231 pp. Chicago: University of Chicago Press.
 Jun M., R. Knutti, D. W Nychka. (2008). Local eigenvalue analysis of CMIP3 climate model errors. Tellus A 60.5, 992-1000.
 Katzav J. (2013). Severe testing of climate change hypotheses. Studies in History and Philosophy of Philosophy of Modern
Physics 44.4, 433-441.
 Katzav J. (2014). The epistemology of climate models and some of its implications for climate science and the philosophy of
science. Studies in History and Philosophy of Modern Physics 46, 228-238.
 Katzav, J. & W. S. Parker (2018). Issues in the Theoretical Foundations of Climate Science. Studies in History and Philosophy of
Modern Physics 63, 141-149.
 Klibanoff P., M. Marinacci and S. Mukerji. (2005). A smooth model of decision making under ambiguity. Econometrica 73, 1849-
1892.
 Klintman M. (2019). Knowledge Resistance: How We Avoid Insight From Others. Manchester: Manchester University Press.
 Knutti R., R. Furrer, C. Tebaldi, J. Cermak, and G. A. Meehl. (2010). Challenges in Combining Projections from Multiple Climate
Models. Journal of Climate 23.10, 2739-2758.
 Koopmans T. C. (1962). On flexibility of future preference. Cowles Foundation for Research in Economics, Yale University, Cowles
Foundation Discussion Papers 150.
 Kreps D. M. and E. L. Porteus. (1978). Temporal resolution of uncertainty and dynamic choice theory. Econometrica 46.1, 185-
200.
 Lahsen M. (2005). Seductive Simulations? Uncertainty Distribution Around Climate Models. Social Studies of Science 35.6, 895-
922.
 Lehrer K. and Wagner, C. (1981). Rational Consensus in Science and Society, 165 pp. Dordrecht: Reidel.
 Lempert R. J., Popper, S. W., Bankes, S. C. (2003). Shaping the Next One Hundred Years: New Methods for Quantitative Long-Term
Policy Analysis, 208 pp. Santa Monica, CA: RAND Corporation, MR-1626-RPC.
 Lenhard J. and E. Winsberg. (2010). Holism, entrenchment, and the future of climate model pluralism. Studies in History and
Philosophy of Modern Physics 41, 253-262.
 Linkov I. et al. (2014). Changing the resilience program. Nature Climate Change 4, 407-409.
 List C. and C. Puppe. (2009). Judgment aggregation: a survey. Oxford Handbook of Rational and Social Choice (eds. P. Anand, C.
Puppe and P. Pattanaik). Oxford: Oxford University Press.
 Lorenz E. (1995). Climate is what you expect. Prepared for publication by NCAR. Unpublished, 1-33.
 Lloyd E. A. (2010). Confirmation and robustness of climate models. Philosophy of Science 77, 971-984.
 Lloyd E. A. (2015). Model robustness as a confirmatory virtue: The case of climate science. Studies in History and Philosophy of
Science 49, 58-68.
 Lloyd E. A. (2009). Varieties of Support and Confirmation of Climate Models. Proceedings of the Aristotelian Society Supplementary
Volume LXXXIII, 217-236.
 Lloyd, E., N. Oreskes (2019). Climate Change Attribution: When Does it Make Sense to Add Methods? Epistemology &
Philosophy of Science 56.1, 185-201.
 Lusk, G. (2017). The Social Utility of Event Attribution: Liability, Adaptation, and Justice-Based Loss and Damage. Climatic
Change 143, 201–12.
 Mach, K. J., M. D. Mastrandrea, P. T. Freeman, and C. B. Field (2017). Unleashing Expert Judgment in Assessment. Global
Environmental Change 44, 1–14.
 Mann M. E., R. S. Bradley and M.K. Hughes (1998). Global-scale temperature patterns and climate forcing over the past six
centuries. Nature 392, 779-787.
 Maslin M. and P. Austin. (2012). Climate models at their limit?. Nature 486, 183-184.
 Mastrandrea M. D., K. J. Mach, G.-K. Plattner, O. Edenhofer, T. F. Stocker, C. B. Field, K. L. Ebi, and P. R. Matschoss. (2011).
The IPCC AR5 guidance note on consistent treatment of uncertainties: a common approach across the working groups. Climatic
Change 108, 675-691.
 McGuffie K. and A. Henderson-Sellers. (2005). A Climate Modelling Primer, 217 pp. New Jersey: Wiley.
 McIntyre S. and R. McKitrick. (2003). Corrections to the Mann et. al. (1998) proxy data base and northern hemispheric average
temperature series. Energy & Environment 14.6, 751-771.
 Mongin P. (1995). Consistent Bayesian Aggregation. Journal of Economic Theory 66.2, 313-51.
 Nordhaus W. D. (2007). A Review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature 45.3,
686-702.
 Nordhaus W. C. (2008). A Question of Balance, 366 pp. New Haven, CT: Yale University Press.
 Oreskes N. and E. M. Conway. (2012). Merchants of Doubt: How a Handful of Scientists Obscured the Truth on Issues from Tobacco
Smoke to Global Warming, 355 pp. New York: Bloomsbury Press.
 Oreskes N. (2007) The Scientific Consensus on Climate Change: How Do We Know We’re Not Wrong? Climate Change: What It
Means for Us, Our Children, and Our Grandchildren (eds. J. F. C. DiMento and P. Doughman), 65-99. Boston: MIT Press.
 Oreskes N., K. Shrader-Frechette and K. Belitz. (1994). Verification, validation, and confirmation of numerical models in the
Earth Science. Science New Series 263.5147, 641-646.
 Parfit D. (1984). Reasons and Persons, 560 pp. Oxford: Clarendon Press.
 Parker W. S. (2009). Confirmation and Adequacy for Purpose in Climate Modelling. Aristotelian Society Supplementary Volume 83.1
233-249.
 Parker W. S. (2010). Comparative Process Tracing and Climate Change Fingerprints. Philosophy of Science 77, 1083-1095.
 Parker W. S. (2011). When Climate Models Agree: The Significance of Robust Model Predictions. Philosophy of Science 78.4,
579-600.
 Parker W. S. (2013). Ensemble modeling, uncertainty and robust predictions. Wiley Interdisciplinary Reviews: Climate Change 4.3,
213-223.
 Parker W. S. (2014). Values and Uncertainties in Climate Prediction, Revisited. Studies in History and Philosophy of Science Part
A 46, 24-30.
 Petersen A. C. (2012). Simulating Nature: A Philosophical Study of Computer-Simulation Uncertainties and Their Role in Climate Science
and Policy Advice, 210 pp. Boca Raton, Florida: CRC Press.
 Resnik M. (1987). Choices: an introduction to decision theory, 221 pp. Minneapolis: University of Minnesota Press.
 Savage L. J. (1954). The Foundations of Statistics, 310 pp. New York: John Wiley & Sons.
 Sen A. (1982). Approaches to the choice of discount rate for social benefit–cost analysis. Discounting for Time and Risk in Energy
Policy (ed. R. C. Lind), 325-353. Washington, DC: Resources for the Future.
 Sen A. (1970). Collective Choice and Social Welfar. San Francisco: Holden-Day Inc.
 Sexton D. M. H., J. M. Murphy, M. Collins and M. J. Webb. (2012). Multivariate Probabilistic Projections Using Imperfect Climate
Models. Part I: Outline of Methodology. Climate Dynamics 38, 2513-2542.
 Sexton D. M. H., and J. M. Murphy. (2012). Multivariate Probabilistic Projections Using Imperfect Climate Models. Part II:
Robustness of Methodological Choices and Consequences for Climate Sensitivity. Climate Dynamics 38, 2543-2558.
 Shackley S., P. Young, S. Parkinson and B. Wynne. (1998). Uncertainty, Complexity and Concepts of Good Science in Climate
Change Modelling: Are GCMs the Best Tools? Climatic Change 38, 159-205.
 Smith L. A. and N. Stern. (2011). Uncertainty in science and its role in climate policy. Phil. Trans. R. Soc. A 369.1956, 4818-4841.
 Spiegelhalter D. J. and H. Riesch. (2011). Don’t know, can’t know: embracing deeper uncertainties when analysing risks. Phil.
Trans. R. Soc. A 369, 4730-4750.
 Stainforth D. A., M. R. Allen, E. R. Tredger and L. A. Smith. (2007a). Confidence, Uncertainty and Decision-support Relevance
in Climate Predictions. Philosophical Transactions of the Royal Society A 365, 2145-2161.
 Stainforth D. A., T. E. Downing, R. Washington, A. Lopez and M. New. (2007b). Issues in the Interpretation of Climate Model
Ensembles to Inform Decisions. Philosophical Transactions of the Royal Society A 365, 2163-2177.
 Steele K. (2006). The precautionary principle: a new approach to public decision-making?. Law Probability and Risk 5, 19-31.
 Steele K. and C. Werndl. (2013). Climate Models, Confirmation and Calibration. The British Journal for the Philosophy of
Science 64, 609-635.
 Steele K. and C. Werndl. forthcoming (2015). The Need for a More Nuanced Picture on Use-Novelty and Double-
Counting. Philosophy of Science.
 Stern N. (2007). The Economics of Climate Change: The Stern Review, 692 pp. Cambridge: Cambridge University Press.
 Stern, N. (2013). The Structure of Economic Modeling of the Potential Impacts of Climate Change: Grafting Gross
Underestimation of Risk onto Already Narrow Scientific Models. Journal of Economic Literature 51.3, 838-859.
 Thompson, Erica, Roman Frigg and Casey Helgeson. Expert Judgment for Climate Change Adaptation, Philosophy of Science
83(5), 2016, 1110-1121,
 von Neumann, J. and Morgenstern, O. (1944). Theory of Games and Economic Behaviour, 739 pp. Princeton: Princeton University
Press.
 Walley P. (1991). Statistical Reasoning with Imprecise Probabilities, 706 pp. New York: Chapman and Hall.
 Weitzman M. L. (2009). On Modeling and Interpreting the Economics of Catastrophic Climate Change. The Review of Economics
and Statistics 91.1, 1-19.
 Werndl C. (2015). On defining climate and climate change. The British Journal for the Philosophy of Science,
doi:10.1093/bjps/axu48.
 Wilby R. L. and S. Dessai. (2010). Robust adaptation to climate change. Weather 65.7, 180-185.
 Weisberg Michael. (2006). Robustness Analysis. Philosophy of Science 73, 730-742.
 Winsberg E. (2012). Values and Uncertainties in the Predictions of Global Climate Models. Kennedy Institute of Ethics Journal 22,
111-127.
 Winsberg, E. 2018. Philosophy and Climate Science. Cambridge: Cambridge University Press.
 Winsberg, E and W. M. Goodwin (2016). The Adventures of Climate Science in the Sweet Land of Idle Arguments. Studies in
History and Philosophy of Modern Physics 54, 9-17.
 Worrall J. (2010). Error, Tests, and Theory Confirmation. Error and Inference: Recent Exchanges on Experimental Reasoning,
Reliability, and the Objectivity and Rationality of Science (eds. D. G. Mayo and A. Spanos), 125-154. Cambridge: Cambridge
University Press.
 Wüthrich, N. (2017). Conceptualizing Uncertainty: An Assessment of the Uncertainty Framework of the Intergovernmental Panel
on Climate Change. In EPSA15 Selected Papers, 95-107. Cham: Springer.
Richard Bradley London School of Economics and Political Science UK

Roman Frigg London School of Economics and Political Science UK

Katie Steele Australian National University Australia

Erica Thompson London School of Economics and Political Science UK

Charlotte Werndl University of Salzburg Austria and

London School of Economics and Political Science
UK