Experimental philosophical bioethics, advance directives,
and the true self in dementia
Brian D. Earp,1,2* Ivar R. Hannikainen,3*
Samuel Dale,4 & Stephen R. Latham1
1. Yale University
2. University of Oxford
3. University of Granada
4. University of Toronto
* Co-first authors
There has been a resurgence of interest in how we should think about the validity of
advance directives following significant personal change (see, e.g., Walsh, 2020, and
associated commentaries). An advance directive (AD) is a formal record of a person’s
preferences regarding end-of-life care: for example, whether they would like to be
treated or have treatment withheld under certain conditions. Typically, an AD is not
legally binding, in that a proxy decision-maker is not obligated to follow it, but it does
relieve physicians of liability if they choose to follow the recorded preferences in an AD.
In this chapter, however, we will not be interested in the legal status of advance
directives. Instead, we will be interested in their ethical status. When deciding about a
person’s end-of-life care, we ask, how much weight should be given to the preferences
expressed in an AD as compared to other considerations—for example, the apparent
preferences of the person upon losing capacity, or what seems to be in their current
best interests. We will focus on cases in which a person undergoes significant changes,
whether in their apparent preferences or in their very self, between Time 1 and Time 2
as described next.
Consider someone at Time 1 who is diagnosed with a neurodegenerative
disease such as Alzheimer’s that is expected to gradually undermine their autonomy
and thus, on some views, their decision-making competence. At Time 2, the person is
judged to be no longer capable of making ethically valid decisions regarding their own
care, in the sense of having sole or final say about certain weighty matters (they should
of course contribute to any care-related decisions to the best of their ability, for
example, by expressing their wishes) (Boumans et al., 2022; Jaworska, 1999; Reiheld,
This is a pre-print of a forthcoming article. Please cite as:
Earp, B. D., Hannikainen, I. R., Dale, S., & Latham, S. R. (forthcoming). Experimental philosophical
bioethics, advance directives, and the true self in dementia. In A. De Block & K. Hens (eds.),
Experimental Philosophy of Medicine. London: Bloomsbury.
2022; Veit et al., 2021). Instead, they must rely on one or more proxy decision-makers
to decide about those aspects of their care.
A classic view, in the spirit of Dworkin (1994) is that the person at Time 1 should
be allowed to express their autonomous preferences for how they would like to be
treated at Time 2, using an AD or similar instrument, where these preferences represent
their “critical interests”—roughly, their interests in shaping their life course in a way that
is consistent with their deeper projects and values. Moreover, this view continues, the
preferences expressed at Time 1 should carry considerable, even decisive, weight in
determining the treatment of the person at Time 2, whose interests are now thought to
be more superficial. Jox (2020) draws the distinction like this:
The preferences we lose with dementia are the more abstract, long-reaching
ones that are rooted in our values, biography and self-image (e.g., the moral
preference to be a vegetarian), while the preferences that emerge are concrete,
short-term preferences rooted in our immediate perception (e.g., the wish to eat
meat that looks and smells deliciously). (p. 86)
According to this way of thinking, one reason to favor a person’s Time 1 “critical”
interests over their Time 2 “experiential” interests (e.g., when deciding about their
treatment), is that the critical interests might seem to better reflect the person’s true
self—who they really are deep down inside—as opposed to their more fleeting desires.
Indeed, when these desires have come about through a process of neurodegeneration,
they can seem to be inauthentic or alien to the self: a reflection of the disease and not
In the present chapter, we seek to better understand how laypeople reason about
the “true self" of a person with advancing dementia. We are also interested in how such
reasoning bears on laypeople’s views about the validity or invalidity of an AD regarding
that person’s treatment. Toward that end, we will report the results of two empirical
studies we undertook to gain insights into this relationship: namely, between judgments
about the true self and whether to follow an AD.
To justify our empirical approach, we draw on arguments from the emerging
literature on experimental philosophical bioethics, also known experimental bioethics or
simply “bioxphi” (Earp, 2019; Lewis, 2020; Mihailov et al., 2021). Bioxphi describes an
approach to normative theorizing in bioethics that employs evidence about laypeople’s
moral judgments as derived from psychological experiments. In a recent publication, we
proposed several means of drawing substantive normative conclusions from
argumentative premises that include such evidence (see Earp et al., 2021). We will
rehearse some of those proposals in a later section, where we explore potential
implications of our findings for competing normative positions in the debate about
The true self and the autonomous self
Before setting up our empirical work, we must first try to clarify some concepts—most
notably that of the “true self.” Some aspects of a person seem essential to who they are,
while other aspects may seem peripheral or superficial. It is common to refer to those
aspects of a person that seem more essential—more authentically them—as their “true
self” while other aspects of the person will not be so described (Strohminger et al.,
2017). For the purposes of this chapter, we do not take a stand on whether people
actually have “true selves.” We simply note that the concept is widely used and forms
an important part of folk psychology. How has the true self been identified?
Within certain strands of Western philosophy, there has been a strong
“rationalist” tendency to consider that a person’s true self is rooted in, or simply is, their
autonomous self: the part of them that is rational and reflective rather than instinctive or
supposedly “emotional” (Aristotle, 2014; Frankfurt, 1971). However, there are also
important traditions according to which a person’s spontaneous feelings, impulses, or
desires—unchecked by cold reason or repressive socialization—are most indicative of
who they really are (for discussion, see Newman et al., 2014).
Both perspectives may play a role in laypeople’s judgments about the true self. In
line with this, previous empirical work suggests that judgments about another’s true self
do not simply track that person’s rational beliefs, autonomous preferences, or explicit
commitments as against their unreflective desires or dispositions—or vice versa.
Instead, judgments about which aspect is more diagnostic of a person’s true self
depend in part upon one’s own moral attitudes or values. For example, in one study,
participants considered the case of a Christian pastor experiencing an internal conflict.
The pastor sincerely and reflectively believes that homosexuality is wrong, but he is
instinctually same-sex attracted. Which aspect of the pastor’s psyche reflects his true
self? Participants in this study were divided in their judgments. Politically conservative
participants viewed the pastor’s true self as the part of him that reflectively judges that
homosexuality is wrong, not the part of him that is spontaneously attracted to other men
(Newman et al., 2014). Politically progressive participants, however, made the opposite
judgment. And a number of follow-up studies have helped to explain why: in general,
participants tend to view another’s true self as the part that is consistent with the
participants’ own value-laden beliefs and priorities, whether or not that part is
associated with autonomy (Newman, Bloom, & Knobe, 2014; De Freitas et al., 2018;
Dranseika et al., 2022).
This evidence raises the possibility that a person could lose their autonomous
decision-making capacity—for example, in the context of neurodegeneration—while still
seeming to remain their “true self.” If so, their non-autonomous preferences at Time 2
could, in principle, reflect their true self as it exists at that time. And this, in turn, could
give us reason to respect those preferences even if they conflict with the person’s
autonomous preferences as recorded in an AD at Time 1 (for discussion, see Toomey,
Further arguments in favor of respecting “Time 2” preferences
A number of authors have raised arguments in this vein. For example, Dresser (1995),
Walsh (2020), and Jaworska (1999) hold that in the case of a Time 1/Time 2 conflict of
the kind described above, the apparent preferences or best interests of a person at
Time 2 often should be given greater and perhaps decisive weight, especially insofar as
the individual has undergone significant personal (e.g., cognitive) change since Time 1.
One way to defend this view is to say that the person at Time 1 and the person at
Time 2 are in some relevant sense the same person, albeit one whose preferences may
have changed over time. According to this view, the fact that the person at Time 2 does
not have the same degree of autonomy as the person had at Time 1 does not mean that
their preferences no longer matter. Instead, this view holds, it is the interests of the
person at Time 2 which are most relevant to treatment decisions affecting them at that
time, and so their (apparent) preferences – or what seems to be in their current best
interests – should be given decisive weight.
Another way to argue for the view is to say that, in some cases, the person at
Time 2 may be sufficiently different from the person at Time 1 that they are, again in
some relevant sense, simply “no longer the same person” as the one who signed the
AD. In that case, the person at Time 1 does not have final say over what happens to the
person at Time 2—they are different people—and so the preferences of the person at
Time 1 as expressed in the AD should not be decisive (see Degrazia, 1999). Rather, it
is again the preferences or best interests of the person at Time 2 that should be given
the most weight in deciding what should be done.
Unsurprisingly, there is no consensus among bioethicists about which of these
various contrasting views is correct or best justified. In previous work, we have argued
that in cases of persisting disagreement among “ethical experts” as to the normatively
appropriate response in a given situation, evidence about the moral judgments of
laypeople with regard to the situation can inform bioethical theorizing (Earp et al., 2021).
For example, if ‘commonsense’ moral judgments on a particular issue can be shown,
through experiments, to be robust rather than fickle or arbitrary, this can increase the
burden of proof on philosophical opponents of the commonsense view.
In this work, we assess laypeople’s judgments about whether the Time 2
preferences of a patient with dementia should be followed in the case of a conflict with
their (autonomous) preferences at Time 1, and whether these judgments relate to
perceptions of the patient’s true self. We describe this case in the following section.
A difficult case: the “happy demented patient”
Although advance directives can express various preferences regarding end-of-life care,
in the usual scenario, a person will state at Time 1 that they prefer not to be treated
under certain conditions that may materialize at Time 2—for example, if they are no
longer able to take care of themselves, or if they can no longer remember the names of
close family members, and they succumb to an illness that will be fatal if left untreated.
In the example of Alzheimer's disease, people may be affected both positively
and negatively, and may become unrecognizable to friends and family (Strohminger &
Nichols, 2015; Tobia, 2016, 2017). On the negative end, a parent or grandparent, say,
who had always been kind, dignified, and gentle-spirited, may seem to deteriorate into
an uncooperative individual who lashes out at those around them. On the positive side,
some individuals with dementia, although they may have unreliable memories and
experience various difficulties, nevertheless seem quite contented in their altered
condition, even happy.
A particularly tragic scenario is one in which a seemingly happy person with
dementia, who now lacks decision-making capacity, falls ill with a curable condition
such as pneumonia and (thereby) meets the criteria set forth in an AD for treatment to
In such a situation, caretakers and proxy decision-makers may face a dilemma:
they can honor the wishes of the person at Time 1, who, out of a concern for their sense
of dignity, perhaps, or for how they would like to be remembered, expressly preferred to
be allowed to die rather than be kept alive under these conditions; or they can decide in
favor of the apparent preferences (or current best interests) of the person at Time 2,
who seems happy and would presumably like to be treated so that they can continue to
enjoy the simple activities of daily life. Consider the real-life case of Mrs. Black,
described in Walsh (2020, p. 55):
At the age of eighty-five, Mrs. Black received a diagnosis of mid-stage dementia.
Due to the progression of her dementia, Mrs. Black often struggled to recall the
names and faces of family members. Nevertheless, she was noted by her nurses
at the residential aged care facility she lived at as being an exceptionally happy
woman. She took joy in her daily activities, particularly watching birds pass by
through the window. Whilst in care, Mrs. Black developed a serious bacterial
infection. She had an advance directive stipulating that if she were ever to suffer
an illness which resulted in her inability to recognize her family members, she
would not wish to receive any medical treatment to prolong her life. Consequently,
her son insisted that her advance directive be followed and so the directive was
implemented. She died shortly after. The entire medical team reported feeling
In the bioethics literature, discussion of dilemmas like these overwhelmingly has
been theoretical. Different theorists draw on abstract principles, as well as their own
intuitions, to make arguments about what should be done in the case of a conflict
between the preferences of a person at Time 1 and the apparent preferences or best
interests of the person at Time 2. But it has not been clear whether, or to what extent,
laypeople – that is, other stakeholders without any special training in philosophy or
bioethics – have similar intuitions about what should be done in such cases. And yet,
laypeople often are faced with real-world responsibility to make decisions about end-of-
life care, including whether to honor a person’s preferences as expressed in an AD. In
developing normative frameworks to guide such decision-making, then, it is not only the
intuitions and judgments of bioethicists and other putative experts that need to be
considered, but also those of a wider range of stakeholders (Savulescu et al., 2019,
2021). As we will now discuss in more detail, this suggests a role for empirical studies.
The need for empirical research
There is a long history within empirical bioethics of collecting data on the views of
various stakeholders. Typically, these studies have used surveys, focus groups, and
other methods to develop a rich, descriptive account of what those views are regarding
key issues. Once there is a good sense of what laypeople think or feel about a given
issue, a number of strategies have been devised to reach normative conclusions about
what should be done in light of those data (Davies et al., 2015).
“Experimental philosophical bioethics” or bioxphi (Earp, Demaree-Cotton, et al.,
2020; Earp et al., 2022; Lewis, 2020; Mihailov et al., 2021) draws inspiration from one
such strategy: the “consultative” strategy. In this strategy, information about the moral
views of laypeople regarding some live issue is collected, and then separately fed into
an argument, advanced by bioethicists, about what should be done with respect to that
issue. Often, this involves developing an ethical framework or recommendation that
accounts for the views collected (i.e., through a process of reflective equilibrium), where
these views are assumed to carry at least prima facie normative weight (Davies et al.,
Bioxphi is consultative in this sense, but it also has some distinctive features. In
addition to documenting the dominant tendencies in stakeholders’ judgments about
realistic bioethical cases, bioxphi research seeks to develop data-driven models of
these judgments by understanding how they come about (i.e., the cognitive processes
underlying them) and what influences them (i.e., the eliciting factors that give rise to
them). Primarily, bioxphi does this by employing experiment-based study designs drawn
from the cognitive sciences. For instance, it might be asked whether the presence or
absence of an AD (a candidate eliciting factor that can be manipulated in an
experiment) shapes participants’ judgments about the appropriate course of treatment—
or withholding of treatment—in a given end-of-life situation (for a similar design, see
Earp, Latham, et al., 2020). With respect to potential cognitive processes, it might then
be asked whether, or how, the influence of an AD on such judgments is related to some
other factor, such as people’s representations of a patient’s true self.
The present studies
In this chapter, we are continuing our investigation (building on Earp, Latham, et al.,
2020) into (a) the factors influencing laypeople’s judgments about whether to treat a
person at Time 2 based on information about their Time 1 preferences, and (b) how
such judgments bear on their assessment of the personal-identity relation (see Earp et
al., 2022) between the autonomous individual at Time 1 and the non-autonomous
individual with dementia at Time 2. We report two studies looking into these questions.
Ethical approval to conduct these studies was obtained from the Yale University
Institutional Review Board (#2000026057).
Using a case study modeled loosely on that of Mrs. Black (described above), we
described a patient (“Robin”) with a dementia diagnosis who, while still autonomous
(Time 1), expresses a preference not to be treated in case she falls ill after no longer
being able to take care of herself—i.e., upon losing her autonomy—as this is a status
she considers undignified (Time 2). However, by the time Robin loses her autonomy in
that sense, she seems to be living a happy life (see below for the precise wording of the
vignette). She then falls ill with a treatable condition. We asked participants whether she
should be treated and whether Robin was “still her true self” (for additional background
on the concept of a true self as it relates to these discussions, see Newman et al., 2014,
2015; Strohminger et al., 2017).
As a preliminary issue, we were interested in whether knowledge of Robin’s Time
1 preference as recorded in an AD (i.e., not to be treated at Time 2) would influence
participant judgments about treatment compared to a control condition in which the
same sequence of events played out but with no explicit information about Robin’s Time
1 preferences. Based on the results of a previous study from our group (Earp, Latham,
et al., 2020), we expected a strong baseline preference for treating Robin at Time 2
given that her condition was described as easily curable but fatal if left untreated; and
we expected that this preference would be significantly weakened by knowledge that
Robin at Time 1 wanted not to be treated under these conditions as stated in her AD.
With respect to the true self question, before we settled on a specific hypothesis,
we reasoned that the results relating to treatment could come out broadly in one of two
ways: either we would find that those participants who judged that Robin was “still her
true self” at Time 2 would be more likely to endorse following the AD—which called for
withholding treatment under the stated conditions—or they would be less likely to do so,
saying instead that Robin at Time 2 should be treated.
A line of reasoning in favor of the first possibility (“true self” judgments being
associated with greater endorsement of withholding treatment) rests on the idea that
participants would adopt a broadly Dworkinian perspective, as follows:
1. Robin at Time 1 autonomously signed an AD saying she did not want to
be treated at Time 2 under X conditions (i.e., if she fell ill and was no
2. If an individual at Time 2 is the same person as the individual who
autonomously signed an AD at Time 1, then the AD applies at Time 2
(when the individual is no longer autonomous) and should be followed.
3. Robin at Time 2, under X conditions, is “still her true self,” which includes
the concept of being the same person as Robin at Time 1.
4. Therefore, the AD applies at Time 2 and should be followed.
5. Therefore, treatment should be withheld from Robin.
However, such reasoning takes the judgment that someone at Time 2 is “still her true
self” as a mere proxy for the seemingly simpler idea that she is “the same person” as
the individual at Time 1. Yet the latter judgment is all that is needed, on a Dworkinian
approach, to conclude that the preferences of the autonomous person at Time 1 (as
recorded in an AD) still apply. But as we suggested in a previous section, participants
might judge that Robin is still her true self even when she is no longer autonomous. In
that case, they might think that the apparent preferences of this current true self should
trump the preferences of an earlier self, whether or not those selves are numerically
identical. In this case, they might reason roughly as follows:
1. Regardless of what someone autonomously states in an AD at Time 1, if a
person is still her true self at Time 2—despite lacking autonomy—and
meets various other conditions (e.g., seems to be happy apart from falling
ill and/or seems to want to keep on living and can easily be cured), then
she should be treated at Time 2.
2. Robin is still her true self at Time 2 and meets the various other
3. Therefore, Robin should be treated.
Based on the results of an informal pilot study, we ended up pre-registering the
prediction that, by whatever route, participants who judged that Robin was still her true
self would be more likely to conclude that she should be treated than that treatment
should be withheld, contrary to the “Dworkinian” line of reasoning spelled out above. By
contrast, we did not pre-register a hypothesis regarding participants who judged that
Robin was not her true self at Time 2. We describe the study and our findings next.
The study design, sampling plan, exclusion criteria, planned statistical analyses, and
hypotheses were pre-registered at aspredicted.org (#66397). All materials, scripts for
analysis, and original (anonymized) data are available at the Open Science Framework
(OSF) at https://osf.io/jqc53/.
Based on our pre-registered sampling plan, 798 adult respondents from the United
States, nationally representative for age, race, and gender, were recruited through the
Prolific platform and randomly assigned to one of two
conditions (control or advance
directive). Thirty-eight participants were excluded because they failed an attention
check, bot check, or completed the survey too quickly (under 3 minutes). This resulted
in a final sample of 760 (Mage = 46.5, SD = 16.4; 51% female, 48% male, 1% non-
binary). Participants were paid $1 for their completion of the survey.
We also recorded whether participants had an advance directive: Most
participants reported not having an advance directive (n = 545; 72%), some reported
having one (n = 188; 25%), and a small minority was unsure (n = 26; 3%).
Participants were presented with a short story about a person named Robin who, before
developing dementia, was dignified, modest, and polite. After the disease progressed,
Robin became loud, messy, and childlike but was nonetheless content with life. After
reading the story, participants were asked several questions designed to capture their
Data for a third condition, in which testimony about the character’s Time 1 preferences was presented,
but only verbally (i.e., not in the form of an advance directive) was also collected; however, to simplify the
presentation of results, we have decided not to include the data for this third condition in the present
chapter. The full, anonymized dataset with all three conditions is available at https://osf.io/jqc53/.
In a two-way ANOVA, we examined whether the effect of advance directives differed across these three
groups: the analysis suggested a potential two-way interaction between AD status and condition, F(2,
753) = 2.75, p = .064 (after controlling for the main effects of condition and AD status). We found that
participants were more likely to withhold treatment in the AD condition than in the control condition
whether they reported having an advance directive (B= -42.8, t = -9.39, p < .001), not having an advance
directive (B = -30.6, t = -11.48, p < .001) or being unsure (B = -28.4, t = -2.30, p = .022), with the effect
size being larger in the case of participants who had an AD.
intuitions regarding Robin’s treatment and identity.
The introductory paragraphs of the
story read as follows:
“Robin is an old friend of your entire extended family. In the prime of her life,
Robin was quietly dignified, modest, polite, and always attentive to the needs of
others. Now, after years of advancing dementia, Robin lives alone in a nursing
facility near your home. Because Robin can no longer make her own decisions,
you are Robin’s healthcare decision-maker.
These days, Robin is physically frail and almost childlike: laughing loudly about
amusing things, complaining loudly about aches and pains, and having
occasional outbursts of temper with nursing staff and fellow patients. Robin’s
clothes are often food-stained, and she has to wear diapers because she cannot
control her bladder. She often smells bad. But Robin is largely unaware of these
problems, and seems to be enjoying life. Her favorite activities include
mealtimes, watching children’s television shows, and being pushed through the
nursing home’s gardens in a wheelchair.”
The story then revealed to participants the nature of Robin’s medical situation:
“Robin’s doctor calls you and says that Robin has come down with pneumonia.
The options are to treat the illness, which would likely cure Robin, or instead to
allow the pneumonia to advance without treatment, keeping Robin comfortable
and permitting her death. The doctor will treat if you authorize it, but will not treat
if you don’t.”
Then, participants were randomly assigned one of two versions of the story. In Version
A (the control condition), they are given no information regarding Robin’s autonomous
Time 1 preferences for what should happen at Time 2. In Version B (the treatment
condition), they are told that Robin previously signed an advance directive which
outlines her treatment preferences:
Version A: “Robin seems happy in her current condition. You realize you never
had a specific conversation with Robin about what to do in these circumstances
(Robin not being able to care for herself).”
Participants were randomly given a version of the story in which Robin is referred to with either male or
female pronouns to account for any potential gender-biased patterns in the results. As Robin’s gender did
not significantly impact any of the reported results (all ps > .45), to avoid confusion, Robin will hereon be
referred to with female pronouns.
Version B: “Robin seems happy in her current condition. Before the dementia
diagnosis, Robin told you that she had filled out an advance directive stating that
she would prefer death to an undignified old age, not being able to care for
herself. (Advance directives are tools to permit patients to express their wishes
about how they’d like to be treated if they lose mental capacity, but there is no
legal obligation to obey them.)”
Participants were then asked to respond to questions regarding Robin’s treatment,
identity, and previous wishes, described below. They were also asked to answer a
simple question about the story to ensure they were paying attention. Then, for
purposes of exploratory analysis, they were asked to complete two additional survey
batteries: the Oxford Utilitarianism Scale (Kahane et al., 2018) and the empathy
subscale of the Interpersonal Reactivity Index (Davis, 1983); however, these data have
not been analyzed for purposes of this chapter.
Participants were asked to respond to three questions designed to assess their
intuitions regarding Robin’s treatment, identity, and previous wishes. The first question
Treatment: “Should Robin's pneumonia be treated?”
Participants were given a sliding scale ranging from 0 to 100, with 0 labeled “Definitely
should NOT be treated” and 100 labeled “Definitely SHOULD be treated,” and were
asked to record their response as a number along the scale. For the second question,
participants were asked to rate their agreement or disagreement with the following
True Self: "In her late stage of dementia, Robin is still her true self."
Participants were given a sliding scale ranging from 0 to 100, with 0 labeled “Completely
Disagree” and 100 labeled “Completely Agree” and were asked to record their response
as a number along the scale. In the third question, participants were asked to rate their
agreement or disagreement (using the same scale just described) with the following
Previous Wishes: "Robin's previous wishes, from when she still had the capacity
to make her own decisions, should fully determine what happens regarding her
Participants were then asked three similar questions designed to probe how weighty
they felt Robin’s Time 1 preferences were to not live an undignified life at Time 2. We
asked the following questions:
Before Robin lost the capacity to make her own decisions, how deeply did Robin
think about whether she had a preference for death over an undignified old age?
Before Robin lost the capacity to make her own decisions, how important was it
for Robin to be allowed to die rather than live in an undignified way?
Before Robin lost the capacity to make her own decisions, how strongly did she
prefer death to an undignified old age?
Responses were given on a scale ranging from 0 (Not at all) through 50 (Somewhat) to
100 (Extremely). We averaged responses to these three-items to form a composite
measure of the perceived weightiness (depth, importance, strength) of Robin’s Time 1
preference for non-treatment at Time 2 (Cronbach’s ɑ = .90). Finally, participants were
presented with a battery of demographic questions asking about their gender, age,
race/ethnicity, political orientation, and religiosity among other qualities, along with an
attention/comprehension check and a bot check (see the materials at
https://osf.io/jqc53/ for the full set of questions).
Summary statistics are presented in Table 1, including information on each of
four Welch’s t-tests comparing the results across experimental conditions.
Table 1. Mean Responses (and 95% Confidence Intervals) by Condition.
(n = 389)
(n = 371)
Should Robin be treated?
[82.1 - 88.6]
[49.2 - 55.6]
t(623.03) = -14.57, p < .001
Cohen’s d = -1.07
Weightiness of Robin’s
preference for non-
[41.9 - 45.8]
[78.8 - 82.9]
t(750.94) = 25.72, p < .001
Cohen’s d = 1.86
Is Robin still her true self?
[37.9 - 43.9]
[25.7 - 31.6]
t(756.76) = -5.55, p < .001
Cohen’s d = -0.40
Previous wishes should
t(773.34) = -1.4, p = .30
Cohen’s d = -0.08
As shown in Table 1, participants tended to agree that Robin’s previous wishes
should determine whether she receives treatment—and this belief did not differ across
conditions. Condition assignment did, however, affect decisions to treat (our first pre-
registered prediction), beliefs about whether Robin was still her true self, and the
perceived weightiness of Robin’s Time 1 preference for non-treatment at Time 2.
Treatment decisions: Pre-registered hypothesis 1
In line with our first pre-registered prediction, participants’ treatment
recommendations differed significantly by condition: participants in the control group,
who were provided with no information about Robin’s Time 1 preferences, tended to
recommend treatment. Meanwhile, those in the advance directive (AD) condition were
significantly less likely to recommend treatment. Notably, however, approximately half of
the participants in the AD condition still recommended treatment, despite being told
explicitly that Robin’s AD instructed that treatment should be withheld in such
Perceived strength of preference
We formed an index of the perceived weightiness of Robin’s Time 1 preference
not to receive treatment at Time 2 by computing the average of the three above-
mentioned items (depth, importance, and strength of preference; Cronbach’s ɑ = .90). A
significant difference in the perceived weightiness of Robin’s preference emerged
across conditions: in the AD condition, participants ascribed a very weighty Time 1
preference to be allowed to die at Time 2, while in the control condition, participants
ascribed a much less weighty preference, with the mean score in this condition (M =
43.9, SD = 21.2) falling below the scale midpoint (50, representing “Somewhat”), t(388)
= -5.71, p < .001, d = -0.29. This may suggest that, in the control condition, with no
recorded preference to the contrary (i.e., in the form of an AD), participants felt on
balance that Robin would prefer to be treated and continue living at Time 2,
notwithstanding the described loss of autonomy.
True self attributions
Beliefs about Robin’s true self differed significantly by condition: although
participants in both conditions, on average, denied that Robin was her true self at Time
2, this tendency was stronger among participants in the AD condition. This suggests
that judgments about a person’s (still being their) true self following neurodegeneration
are influenced at least in part by information about their autonomous preferences at
Analyses of covariance: pre-registered hypothesis 2
In the AD condition, participants were less likely to recommend treatment for
Robin at Time 2 and ascribed a weightier Time 1 preference to be allowed to die.
Indeed, an ANCOVA indicated that the perceived weightiness of Robin’s preference for
non-treatment predicted decisions not to treat in both conditions (BControl = -0.35, t = -
5.19, BAD = -0.97, t = -12.10, both ps < .001), and accounted for much of the effect of
condition (see Table 2). This suggests that most of the effect of the AD on treatment
decisions was explained by the perceived (greater) weightiness of the person’s Time 1
preferences regarding treatment at Time 2 (compared to the no AD baseline/control).
Moving now to true self judgments. Confirming our second pre-registered
prediction, we found that—even when controlling for both condition assignment and the
perceived weightiness of Robin’s Time 1 preference for non-treatment—beliefs about
Robin’s true self independently predicted treatment decisions (see Model 3 in Table 2).
Simple slopes analysis revealed that participants who judged that Robin was still her
true self at Time 2 were more likely to recommend treatment in either condition: BControl
= 0.14, t = 3.26; BAD = 0.43, t = 8.04, both ps < .001, contrary to the “Dworkinian” line of
reasoning outlined above. As indicated by the condition✕true self interaction, the
relationship was significantly stronger in the AD condition than in the control condition
(see Figure 1).
Table 2. Analyses of Covariance in Treatment Decisions.
Condition ✕ Preference
Condition ✕ True self
Note. All ps < .001.
Figure 1. Scatter plot of treatment decisions against true self judgments by
condition. Grouped density plots are displayed along both axes
Consistent with previous work from our group (Earp, Latham, et al., 2020), U.S.
participants in this sample (nationally representative for age, race, and gender) showed
a strong default preference, in the control condition, for treating a non-autonomous
person with dementia (“Robin”) described as suffering from readily curable pneumonia.
However, the existence of an AD instructing non-treatment significantly weakened this
default tendency, with about half of the participants indicating that Robin should be
treated (despite the AD) and half recommending that treatment be withheld. Notably,
this main effect of condition on treatment decisions was mostly explained by the
perceived weightiness of Robin’s Time 1 preference to be allowed to die at Time 2
under the stated circumstances (i.e., following loss of autonomy). This is consistent with
“substituted judgment” standards for end-of-life care, whereby an AD is interpreted as
especially good evidence of what the person, at Time 1, would have wanted to happen
in the current situation.
That being said, even when controlling for the effect of condition (AD versus
control) and the perceived weightiness of Robin’s Time 1 preference for non-treatment
at Time 2, we found a relationship between beliefs about Robin’s true self and treatment
recommendations. On average, participants judged that Robin was no longer her true
self after developing dementia. Yet, contrary to the Dworkinian prediction, the more
strongly participants judged that Robin was “still her true self” at Time 2
(notwithstanding her lack of autonomy), the stronger their recommendation for treatment
despite being informed that her AD requested non-treatment under the stated
On a broader note, the very fact that a large proportion of participants judged that
Robin was still her true self despite her lack of autonomy suggests that the “true self”
and the “autonomous self” are not synonymous among laypeople.
The results of our first study shed light on laypeople’s endorsement of treatment (versus
allowing a patient to die) and their judgments about a person’s true self following loss of
autonomy in a case of dementia. Among those participants who judged that Robin was
still her true self at Time 2, the overwhelming recommendation was that she be treated.
Whereas, among those who judged that Robin was no longer her true self at Time 2,
treatment recommendations were polarized: about half of these participants
recommended treatment (contrary to the AD), and half recommended that treatment be
withheld (consistent with the AD). To further explore the relationship between true self
judgments and treatment recommendations, we conducted a follow-up study.
The purpose of Study 2 was to unpack different potential senses of “true self”
that participants might have had in mind. In particular, we wanted to understand the
greater variability in treatment decisions among participants who denied that Robin
remained her true self at Time 2, compared to those who judged that she was still her
true self at Time 2. To that end, we conducted a replication of the “AD” condition from
Study 1, but added further questions teasing apart different potential senses of being
the same person, to see how these related to true self judgments and treatment
decisions. We also included exploratory questions assessing the extent to which
participants regarded Robin’s Time 1 (explicit) preferences versus Time 2 (implicit)
preferences as being more genuine or authentic.
Study 2 was an exploratory study, for which we did not make a pre-registration. Based
on available funding, 202 participants were recruited through MTurk to complete the
study. After excluding data from 19 respondents for failing to meet the criteria outlined
previously, we obtained a final sample of 183 participants (Mage = 34.2, SD = 11.36;
50% female, 45% male, 4% other/non-binary). Participants were paid $1 for their
completion of the survey.
Participants were presented with the same story from Study 1. We eliminated the
control group (Version A) and presented only the advance directive vignette (Version B).
Following the vignette, participants were asked a series of questions as described
below. The demographic questions remained unchanged from Study 1.
In Study 2, participants were presented with the same “treatment” and “true self”
questions as in Study 1, followed by additional questions meant to probe different
senses of being the same person over time. To frame these questions, participants
were presented with a hypothetical three-way debate between the characters Art, Bart,
and Cart (inspired by Tobia, 2015)—each one espousing a different concept of
identity—and asked to indicate which character they agreed with the most.
In the debate, Art espouses a total identity disruption view, according to which
Robin at Time 2 is metaphysically distinct from Robin at Time 1: that is, the two Robins
are not the same person, either in a superficial sense or in terms of a deeper true self.
Bart, by contrast, espouses the opposite view—an identity persistence interpretation:
Robin at Time 2 is the exact same person as Robin at Time 1, including in terms of a
deeper true self. Finally, Cart espouses a dual character view of identity (Knobe et al.,
2013; Knobe, 2022): although there is some, superficial or technical sense in which
Robin at Time 2 is the same person as Robin at Time 1, in a deeper sense, Robin is no
longer her true self. The debate was presented as follows:
Imagine you've just read this story about Robin aloud at your local café and three
inquisitive people – Art, Bart, and Cart – overhear you. Each of them offers their
opinion. Which person do you agree with the most?
Art says: “At the end of the story, the patient with dementia is simply not the
same person as Robin anymore. When you think about what it really means to
be Robin, deep down inside, it’s almost like Robin went out of existence and
some totally different person – a person with dementia – took Robin’s place.”
Bart says: “Of course the patient with dementia at the end of the story is the
exact same person as Robin. Moreover, when you think about what it really
means to be Robin, deep down inside, you would have to say that Robin at the
end of the story is still very much her true self.”
Cart says: “There is of course an important sense in which the patient with
dementia at the end of the story is still Robin, but when you think about what it
really means to be Robin, deep down inside, you would have to say that Robin at
the end of the story is just no longer her true self anymore.”
As a further exploratory measure, participants were then presented with a similar-style
question, with the aim of uncovering peoples’ judgments regarding the authenticity of
Robin’s preferences before and after developing dementia. Our thinking here was as
follows. One reason a person might endorse treatment at Time 2 despite instructions to
the contrary in an AD from Time 1, is because they judge that the preference as
recorded in the AD was not actually a reflection of the person’s true desires: perhaps
they think, for example, that the person was insufficiently careful in thinking about what
they really wanted to happen at Time 2. Or perhaps they think—as we suggested in the
introduction—that the person simply changed their mind between Time 1 and Time 2,
with their Time 2 preference being perfectly valid (i.e., deserving of respect or
consideration). To account for these kinds of possibilities, we described a second
debate between three new characters.
A character named Derik espouses a preference change view: Robin’s preference for
non-treatment at Time 1 was authentic, but by Time 2 she had simply changed her
preference, now authentically preferring treatment. Erik espouses a view according to
which only Robin’s preference at Time 2 was authentic; and Ferik espouses an original
preference view: Robin’s original preference for non-treatment at Time 1 was authentic,
and those preferences still apply at Time 2. The debate was presented as follows:
Imagine you've just read this story about Robin aloud at your local café and three
more inquisitive people – Derik, Erik, and Ferik – overhear you. Each of them
offers their opinion. Which person do you agree with the most?
Derik says: “Robin made an authentic decision in expressing a preference for
death to an undignified old age. What she wrote in the advance directive,
therefore, was absolutely her true preference at Time 1. However, by Time 2,
when the dementia had set in, Robin’s preference had simply changed. Even
though Robin at Time 1 – before getting dementia – would definitely regard her
existence at Time 2 as highly undignified, her preference for death under those
conditions shifted so that by Time 2, she had come to prefer life under the very
Erik says: “Robin did not make an authentic decision in expressing a preference
for death to an undignified old age. What she wrote in the advance directive,
therefore, was not her true preference, even at Time 1. By contrast, at Time 2,
when the dementia had set in, Robin's true preference was able to emerge. Even
though Robin at Time 1 – before getting dementia – would definitely regard her
existence at Time 2 as highly undignified, her supposed preference for death
under those conditions was inauthentic. By Time 2 it had become clear that her
true preference all along was for life under the very same conditions."
Ferik says: “Robin made an authentic decision in expressing a preference for
death to an undignified old age. What she wrote in the advance directive,
therefore, was absolutely her true preference at Time 1. Moreover, by Time 2,
when the dementia had set in, Robin's original preference had not changed. In
other words, Robin at Time 1 – before getting dementia – would definitely regard
her existence at Time 2 as highly undignified, and her preference for death under
those conditions still applied at Time 2, when those conditions occurred."
Next, participants were presented with a number of exploratory statements reflecting the
nature or importance of Robin’s preferences at Time 2 (i.e., following the loss of
autonomy). These statements, which were not analyzed for the present chapter (see
https://osf.io/jqc53/ for the full set of statements), were followed by a
comprehension/attention check and demographic measures as described in Study 1.
To assess whether participants favored particular views about Robin’s personal
identity and authentic preferences, we conducted multinomial proportion tests against
the uniform distribution (i.e., ⅓).
First, we examined whether participants endorsed a unitary view of Robin’s
identity—e.g., that Robin was still herself (persistence view), or that she was no longer
herself at all (disruption view)—or whether, alternatively, participants embraced a dual
character view of Robin’s identity.
A majority of participants (n = 99; 57%, 95% CI [49%, 64%]) preferred the dual
character description, proportion test c2 = 41.49, p < .001. The proportions of
participants who opted for the persistence (n = 38; 22%, 95% CI [14%, 29%]) and
disruption (n = 38; 22%, 95% CI [14%, 29%]) descriptions were significantly lower than
the null proportion of ⅓, ps = .001.
What did participants view as Robin’s authentic preference? A majority of
participants (n = 93; 53%, 95% CI [45%, 60%]) regarded Robin’s original wishes as the
only authentic preference, proportion test c2 = 30.02, p < .001. Some participants
endorsed the preference change view (n = 70; 40%, 95% CI [32%, 47%]), while rather
few (n = 14; 8%, 95% CI [1%, 16%]) regarded Robin’s later, Time 2 preference as being
the only authentic one.
Cross-tabulating the personal identity and authentic preference responses
revealed a significant association between beliefs about Robin’s personal identity and
her authentic preferences, c2(df = 4) = 11.91, p = .018. Of the nine response profiles,
two profiles revealed a frequency statistically greater than would be expected by chance
(i.e., 1/9, with ps < .001).
As described in Table 3, these response profiles corresponded to dual character
views of identity. While both profiles acknowledged that Robin at Time 2 was no longer
her true self, they differed in whether Robin’s inferred Time 2 preference (i.e., to be
treated and live) was seen as superseding the Time 1 preference recorded in the AD
(i.e., not to be treated and allowed to die). In one profile, participants reported that
Robin’s inferred preference to live at Time 2 was inauthentic—and therefore that Robin
should not be treated, in keeping with the instructions of the AD. In another, participants
judged that the preferences of Robin’s ‘shallow’ self at Time 2 were nevertheless
authentic (dual character-preference change view), and therefore that she should be
treated (see Figure 2).
A further two response profiles were non-significantly more frequent than chance
(ps > .20), and corresponded to unitary views of identity according to which either Robin
was no longer herself and her authentic preference was established at Time 1 (the
disruption view), or Robin was still herself and her authentic preference had shifted from
Time 1 to Time 2 (the persistence view). Replicating Study 1, participants with the
persistence view judged that Robin should be treated while participants with the
disruption view judged that she should not be treated.
In line with Study 1, treatment decisions depended on participants’ beliefs about Robin’s identity, F(2,
172) = 6.29, p = .002, h2p = .073. Participants were more likely to recommend treatment if they endorsed
the persistence view (M = 74.5, 95% CI [63.8, 85.2]) than if they endorsed the disruption view (M = 52.1,
95% CI [41.4, 62.7]), t = 2.94, p = .010, or the dominant, dual character view (M = 53.2, 95% CI [46.3,
59.3]), t = -3.67, p = .003. The latter two groups, both of which include the judgment that Robin at Time 2
is no longer her true self, did not differ, t = 0.18, p = .98.
Thus, the decision to treat (whether or not Robin is “still her true self” at Time 2)
seems to correlate strongly with the view that Robin at Time 2 (authentically) prefers to
be treated, notwithstanding what she may have autonomously preferred at Time 1.
Table 3. Response profiles and mean treatment decisions.
At Time 2, Robin was no longer her true
self though she remained herself in some
‘shallow’ sense. Only her wishes at Time 1
reflect an authentic preference.
At Time 2, Robin was no longer her true
self though she remained herself in some
‘shallow’ sense. Her authentic preference
changed from Time 1 to Time 2.
At Time 2, Robin was no longer herself.
Only her wishes at Time 1 reflect an
At Time 2, Robin was still herself. Her
authentic preference had changed from
Time 1 to Time 2.
Figure 2. Scatter plot of treatment decisions against true self judgments by
profile, with overlaid 95% confidence ellipses.
In Study 2, we replicated the relationship between true self beliefs and treatment
recommendations in the presence of an AD. Judgments that Robin was still her true self
were associated with decisions in favor of treatment. In contrast, participants who
thought Robin was no longer her true self manifested greater disagreement about
whether Robin should be treated. By appealing to the notion of a dual character concept
of personhood, Study 2 helped us understand why.
Broadly speaking, participants endorsed a dual character concept of identity: for
most, the dementia patient was still Robin in a superficial sense, but no longer Robin in
a deeper sense. This understanding of Robin’s bifurcated self raised a further question
when considering what the doctor should do: i.e., whether the preferences of Robin at
Time 2 or Time 1 should guide proxy decision-making. Many participants reasoned that
only the wishes of Robin’s true self at Time 1 constituted a genuine or authentic
preference, and therefore that the doctor ought to follow the AD (and withhold
treatment). Meanwhile, other participants considered that the wishes of Robin at Time
2—even though she may no longer be her true self—nevertheless constitute an
authentic preference. From this perspective, the doctor ought to provide treatment, in
order to honor the current wishes of Robin (i.e., at Time 2).
In this work we addressed a recurring question about the validity of advance directives
(ADs). What should be done in the case of a conflict between a person’s autonomous
preference at Time 1 (when preparing an AD) and the current preferences or apparent
best interests of that person at Time 2 (having experienced a loss of autonomy)? To
inform this debate, we examined how a large sample of US adults (nationally
representative for age, race, and gender), who could potentially face the real-world
responsibility of making proxy decisions about end-of-life care, reason about this
conflict. We conducted two studies which revealed the extent to which ADs guide
people’s proxy decision-making, and on what basis.
Our study revealed an impact of ADs on proxy recommendations, which was
largely explained by people’s inferences about the patient’s relevant preferences. This
impact was, however, modest—with people disagreeing substantially about whether to
apply the AD or not, especially when they judged that the patient at Time 2 was no
longer her true self. Thus, our studies documented a role for beliefs about the patient’s
personal identity, unpredicted by existing theory.
Since ADs record people’s preferences for how they themselves would like to be
treated, one might expect that identity persistence constitutes a prerequisite for the AD
to apply. In other words, the AD may apply as long as the patient is still herself—but
could not apply to entirely different persons or even radical transformations of the
patient herself. Our studies revealed the opposite pattern: When people judged that the
patient was the same person as the one who wrote the AD (in the sense of still being
her true self) they disregarded her prior request and recommended treatment instead.
Our follow-up study helped to explain why: If Robin was still herself at Time 2, then her
prior request was voided by her current (authentic) preference.
That being said, most participants embraced a more sophisticated view of
Robin’s identity: they acknowledged that she remained herself in some (i.e., shallow or
technical) sense, while simultaneously recognizing that, in a deeper sense, Robin was
no longer her true self. It was precisely this dual character view of Robin’s identity
(Knobe, 2022) that engendered such profound disagreement about the import of her
Some participants considered that only the preferences of Robin’s true self at
Time 1 were authentic, and therefore recommended applying the AD—against Robin’s
current inferred preferences or apparent best interests. Other participants considered
that the preferences of Robin’s less-autonomous self at Time 2 could be genuine, and
therefore recommended treatment—against the wishes of Robin at Time 1. In this way,
a dual character view of personal identity may help to explain the extent of
disagreement among proxy decision-makers when considering whether to follow an AD.
Our study can be contextualized in terms of the normative debate about the
validity of ADs. Dworkin (1994) argued that ADs ought to hold decisive weight over a
person’s medical treatment—as long as they capture their critical interests, i.e., those
interests that reflect their true or enduring self, which Dworkin seems to equate to the
person’s autonomous self. In these circumstances, a person’s experiential interests
(e.g., to keep living a pleasant but “undignified” life) ought to be overridden in favor of
their critical interests (e.g., to end their life if they can no longer take care of
Our data suggest that many laypeople are not persuaded by this line of
reasoning. First, a large proportion of participants in our study judged that Robin at Time
2 was still her true self and had authentic preferences that should be followed, even
though she was no longer autonomous. These participants thus seemed to disagree
with the Dworkinian position that the true self must be autonomous (although they might
still think that a non-autonomous true self could represent one’s critical interests).
Second, even among those participants who judged that Robin was technically the
same person at Time 2, but no longer her true self, participants did not uniformly
conclude that the AD should apply and that treatment should be withheld. Some
participants reached that conclusion; but many others recommended honoring Robin’s
non-autonomous, Time 2 “experiential” interests, rather than doing what would align
with her autonomously endorsed “critical” interests as recorded in the AD at Time 1.
However, whether participants’ judgments about personal identity, in either the
deep or shallow sense, played a causal role in their proxy decision-making for Robin is
uncertain on the basis of these studies. For example, rather than true self judgments
driving treatment decisions, it could be that, for some participants, a desire not to
withhold treatment from an easily curable patient influences them to judge that the
patient, despite appearances, is “still her true self” after all (such that her current
preferences can appropriately be given greater weight). Future work should attempt to
manipulate these variables to see where the causal direction lies. Such work should
also explore what specific factors contribute to the perception that someone is or is not
their true self upon losing decision-making capacity as a result of developing dementia.
Whether there are normative implications to our findings remains to be seen.
Certainly, the findings seem to be in tension with what we have been calling
“Dworkinian” accounts of ADs, according to which a person’s Time 1 (autonomous)
preferences, as recorded in an AD, should control their Time 2 treatment following loss
of autonomy, just so long as the person at Time 2 is the same person as the one who
signed the AD. The normative recommendation of such accounts does not accord with
the judgments of many of our participants who judged both that (a) the person at Time 2
was (in one sense or another) the same person as the one at Time 1, but (b) the
preferences of the person at Time 1 should not determine what happens at Time 2. A
bioethicist who wanted to argue that the Dworkinian view is, nevertheless, correct, might
need to explain why so many laypeople are getting it wrong. For example, such a
bioethicist might want to undertake a “debunking” strategy (see Earp et al., 2021;
Mukerji, 2019), conducting further experiments to show that the judgments of laypeople
about this case are influenced by inappropriate (that is, normatively irrelevant or
Alternatively, one might pursue what we have called the “parsimony” strategy for
drawing normative conclusions from premises that include empirical information (i.e.,
about people’s moral judgments). According to this strategy, “If relevant stakeholders
consistently make a judgment p which encodes moral claim M, then M has prima facie
normative weight” (Earp et al., 2021). If our results turn out to be robust and replicable,
then, according to this strategy, the following tentative conclusion would be justified:
insofar as a patient with dementia is judged by relevant stakeholders to be her true self
following a loss of autonomy, the normatively correct thing to do is to treat her under the
described conditions, even if this conflicts with what is instructed by her AD. Whereas,
when a patient is judged not to be her true self, a “pluralism” strategy could be pursued.
This would reflect the fact that participants in our studies who denied that Robin was still
her true self sharply disagreed about whether she should be treated. This strategy holds
that in cases of “persistent disagreement between two or more conflicting, yet pro tanto
reliable, judgments, these judgments may all have comparable normative weight,” at
least as a starting point for further bioethical inquiry (Earp et al., 2021).
Rather than drawing a specific normative conclusion, we have sought to illustrate
how insights into the cognitive processes that guide proxy medical decision-making,
when coupled with certain normative premises, can give rise to new empirically
informed arguments in the debate about the validity of advance directives (ADs). Of
course, these findings must be treated with caution: they should be regarded as
tentative until they are replicated by others; and they cover just one type of scenario—
the case of a “happy demented patient” with an easily curable condition whose AD calls
for withholding of treatment. Multiple features of this case could be manipulated in future
work, providing greater insight into the specific factors that drive the judgments of
laypeople: the AD could call for treatment rather than non-treatment; the patient could
be unhappy rather than happy; the condition could be burdensome or risky to treat
rather than easily curable, and so on. Additionally, qualitative research into the
reasoning processes that underlie proxy decision-making may complement the
evidence obtained through the quantitative methods we have relied upon here. In
deciding what should be done in various cases, however, we suggest that information
about the factors that influence the judgments of relevant stakeholders should be taken
into account (whether as part of an attempt to vindicate those judgments or to call them
into question). We hope that work in experimental philosophical bioethics – or bioxphi –
will continue to contribute to this end.
Thank you to Joshua Knobe and members of the Knobe lab for extensive feedback on
these findings and ideas.
Aristotle. (2014). Aristotle: Nicomachean Ethics (R. Crisp, Ed.). Cambridge University
Boumans, J., van Boekel, L. C., Verbiest, M. E., Baan, C. A., & Luijkx, K. G. (2022).
Exploring how residential care facilities can enhance the autonomy of people with
dementia and improve informal care. Dementia, 21(1), 136–152.
Davies, R., Ives, J., & Dunn, M. (2015). A systematic review of empirical bioethics
methodologies. BMC Medical Ethics, 16(15), 1–13.
Davis, M. H. (1983). Measuring individual differences in empathy: Evidence for a
multidimensional approach. Journal of Personality and Social Psychology, 44(1),
Degrazia, D. (1999). Advance directives, dementia, and ‘the someone else problem.’
Bioethics, 13(5), 373–391. https://doi.org/10.1111/1467-8519.00166
Dresser, R. (1995). Dworkin on dementia: Elegant theory, questionable policy. Hastings
Center Report, 25(6), 32–38. https://doi.org/10.2307/3527839
Dworkin, R. (1994). Life’s Dominion: An Argument About Abortion, Euthanasia, and
Individual Freedom. Knopf Doubleday Publishing Group.
Earp, B. D. (2019, August 1). Introducing bioxphi. The New Experimental Philosophy
Earp, B. D., Demaree-Cotton, J., Dunn, M., Dranseika, V., Everett, J. A. C., Feltz, A.,
Geller, G., Hannikainen, I. R., Jansen, L. A., Knobe, J., Kolak, J., Latham, S.,
Lerner, A., May, J., Mercurio, M., Mihailov, E., Rodríguez-Arias, D., López, B. R.,
Savulescu, J., … Tobia, K. (2020). Experimental philosophical bioethics. AJOB
Empirical Bioethics, 11(1), 30–33.
Earp, B. D., Latham, S. R., & Tobia, K. P. (2020). Personal transformation and advance
directives: An experimental bioethics approach. American Journal of Bioethics,
20(8), 72–75. https://doi.org/10.1080/15265161.2020.1781964
Earp, B. D., Lewis, J., Dranseika, V., & Hannikainen, I. R. (2021). Experimental
philosophical bioethics and normative inference. Theoretical Medicine and
Bioethics, 42(3), 91–111. https://doi.org/10.1007/s11017-021-09546-z
Earp, B. D., Lewis, J., Skorburg, J. A., Hannikainen, I. R., & Everett, J. A. C. (2022).
Experimental philosophical bioethics of personal identity. In K. P. Tobia (Ed.),
Experimental Philosophy of Identity and the Self (pp. 183–202). Bloomsbury.
Frankfurt, H. G. (1971). Freedom of the will and the concept of a person. Journal of
Philosophy, 68(1), 5–20. https://doi.org/10.2307/2024717
Jaworska, A. (1999). Respecting the margins of agency: Alzheimer’s patients and the
capacity to value. Philosophy & Public Affairs, 28(2), 105–138.
Jox, R. J. (2020). Living will versus will to live? How to navigate through complex
decisions for persons with dementia. American Journal of Bioethics, 20(8), 85–
Kahane, G., Everett, J. A. C., Earp, B. D., Caviola, L., Faber, N. S., Crockett, M. J., &
Savulescu, J. (2018). Beyond sacrificial harm: A two-dimensional model of
utilitarian psychology. Psychological Review, 125(2), 131–164.
Knobe, J. (2022). Personal identity and dual character concepts. In K. Tobia (Ed.),
Experimental Philosophy of Identity and the Self (p. in press). Bloomsbury.
Knobe, J., Prasada, S., & Newman, G. E. (2013). Dual character concepts and the
normative dimension of conceptual representation. Cognition, 127(2), 242–257.
Lewis, J. (2020). From x-phi to bioxphi: Lessons in conceptual analysis 2.0. AJOB
Empirical Bioethics, 11(1), 34–36.
Mihailov, E., Hannikainen, I. R., & Earp, B. D. (2021). Advancing methods in empirical
bioethics: Bioxphi meets digital technologies. American Journal of Bioethics,
21(6), 53–56. https://doi.org/10.1080/15265161.2021.1915417
Mukerji, M. (2019). Expermental Philosophy: A Critical Study. Rowman & Littlefield.
Newman, G. E., Bloom, P., & Knobe, J. (2014). Value judgments and the true self.
Personality and Social Psychology Bulletin, 40(2), 203–216.
Newman, G. E., De Freitas, J., & Knobe, J. (2015). Beliefs about the true self explain
asymmetries based on moral judgment. Cognitive Science, 39(1), 96–125.
Reiheld, A. (2022). Methods of Bioethics. In J. M. Reynolds & C. Wieseler, The
Disability Bioethics Reader (1st ed., pp. 50–60). Routledge.
Savulescu, J., Gyngell, C., & Kahane, G. (2021). Collective Reflective Equilibrium in
Practice (CREP) and controversial novel technologies. Bioethics, online ahead of
Savulescu, J., Kahane, G., & Gyngell, C. (2019). From public preferences to ethical
policy. Nature Human Behaviour, published online(August 26), 1–3.
Strohminger, N., Knobe, J., & Newman, G. (2017). The true self: A psychological
concept distinct from the self. Perspectives on Psychological Science, 12(4),
Strohminger, N., & Nichols, S. (2015). Neurodegeneration and identity. Psychological
Science, 26(9), 1469–1479. https://doi.org/10.1177/0956797615592381
Tobia, K. P. (2015). Personal identity and the Phineas Gage effect. Analysis, 75(3),
Tobia, K. P. (2016). Personal identity, direction of change, and neuroethics.
Neuroethics, 9(1), 37–43. https://doi.org/10.1007/s12152-016-9248-9
Tobia, K. P. (2017, September 19). Change becomes you. Aeon.
Toomey, J. (2021a). “As long as I’m me”: From personhood to personal identity in
dementia and decision-making. Canadian Journal of Bioethics, 4(1), 57–69.
Toomey, J. (2021b). Understanding the perspectives of seniors on dementia and
decision-making. AJOB Empirical Bioethics, 12(2), 101–112.
Veit, W., Earp, B. D., Browning, H., & Savulescu, J. (2021). Evaluating tradeoffs
between autonomy and wellbeing in supported decision making. American
Journal of Bioethics, 22(11), 21–24.
Walsh, E. (2020). Cognitive transformation, dementia, and the moral weight of advance
directives. American Journal of Bioethics, 20(8), 54–64.