-------------------------------------------------------------------
THE SMALL IMPROVEMENT ARGUMENT, EPISTEMICISM, AND INCOMPARABILITY
-------------------------------------------------------------------

             Edmund Tweedy Flanigan; John Halstead
                             2018

[This  is the  penultimate  draft  of an  article  that appears  in
Economics &  Philosophy, Vol 34,  iss. 2, pp. 199-219.  Please cite
that version: http://doi.org/10.1017/S0266267118000019.]

## Introduction

Which is more impressive - St Paul's Cathedral or the Eiffel Tower?
It seems that  neither is more impressive than the  other. Are they
equally impressive? If they were,  then a minute improvement to the
impressiveness of St Paul's would  make it more impressive than the
Eiffel Tower. But a minute  improvement does not seem sufficient to
shift  the  balance. So,  the  two  cannot be  equally  impressive.
Therefore, none  of the  trichotomy of comparative  relations 'more
impressive than,'  'less impressive than' and  'equally impressive'
apply between  the Eiffel Tower  and St  Paul's. This is  the Small
Improvement Argument (SIA).[1]

For  some SIAs,  our  failure  to have  confidence  in  any of  the
components of the trichotomy can be explained by what we might call
our 'contingent ignorance' about the properties of the two options,
as when there is  some fact of the matter that  we just haven't yet
found out. Others can be  explained by our conceptual incompetence,
such as when  we haven't thought hard enough  about the comparative
concept in  question. However,  there are  some 'hard  cases' which
cannot be  explained in this  manner. The comparison of  the Eiffel
Tower and St Paul's is arguably  one example: even once we have all
the information and are sufficiently conceptually competent, we may
still be  unable to conclude  that one  is more impressive  or that
they  are equally  impressive. We,  along  with a  number of  other
philosophers, believe that these hard cases are borderline cases of
vague comparative predicates (Broome 1997).  Just as we cannot come
to a definitive answer in these hard comparisons, we cannot come to
a definitive answer about  when a man is bald, even  if we know all
the facts about  the number of hairs  on his head and  how they are
dispersed,  and even  if we  completely understand  the concept  of
baldness. The reason for this, in  our view, is that these are both
instances of vagueness.

Ruth Chang (2002) has notably  rejected this claim. She argues that
hard cases (which she calls 'superhard cases') can be distinguished
from  borderline  cases  by  considering some  of  the  permissible
practical responses  to the two types  of case. We will  not engage
with  that argument  here  for  reasons of  space  and because  our
arguments would  largely repeat those made  elsewhere (Broome 1997;
Wasserman 2004;  Gustafsson 2013; Williams 2016).  Instead, we take
it for granted here that the hard cases raised in some SIAs exploit
vagueness, and  our conclusion  is accordingly conditioned  on that
assumption.

The  question  that follows  is,  what  is  the correct  theory  of
vagueness, and  what are its  implications for the SIA?  Almost all
vagueness-based accounts of the SIA have thus far assumed the truth
of supervaluationism, one leading theory of vagueness. According to
supervaluationism,  sentences involving  borderline cases  of vague
predicates are  neither true nor false.  Supervaluationist accounts
of the SIA thus  say of the above case that it  is neither true nor
false that the Eiffel Tower is  more impressive than St Paul's, nor
that  it is  less  impressive, and  nor that  the  two are  equally
impressive. However, supervaluationism  is not universally accepted
as  a theory  of vagueness,  and it  has some  problematic features
(Williamson 1994,  chap. 5).  For example, it  implies that  a true
disjunction can have no true disjuncts. With respect to the SIA set
out above, for instance, it says that the disjunction 'St Paul's is
either more impressive than, less impressive than, or as impressive
as  the Eiffel  Tower' is  true, but  that none  of the  individual
disjuncts is  true.[2] This seems problematic.  Indeed, the natural
thing to say in response to such a theory may be that it simply has
the wrong  account of what  'or' means; we  need a theory  which is
consistent with 'or,' and supervaluationism  does not fit the bill.
Appeals  to semantics  of this  kind command  widespread assent  in
other domains. For example, the  most popular response to the claim
that  the relation  'better than'  is intransitive  is simply  that
this  must be  wrong, as  a  semantic matter  (Huemer 2013).  Thus,
supervaluationism's hegemony in debates  about the SIA is certainly
open to question.[3]

Although  this  of course  does  not  count as  decisive  criticism
of  supervaluationism,   it  does   give  us  prima   facie  reason
to  explore  viable   alternatives  to  it.  One   of  the  leading
alternatives to  supervaluationism is epistemicism. In  contrast to
supervaluationism, epistemicism holds  that borderline propositions
have exactly  one truth  value - true  or false -  but that  we are
incurably ignorant of it. Consider the example of a plump man, Jim,
who is a borderline case of  'is fat.' The epistemicist denies that
it  is  neither true  nor  false  that  Jim  is fat.  Rather,  this
proposition has  exactly one truth  value, but we cannot  know what
that truth value is.

In the wake  of persuasive recent defences of  epistemicism and its
growing philosophical popularity, epistemicism  at least deserves a
seat at  the table in  discussions about vagueness  (Sorensen 1988;
Williamson 1994).[4] Whether those  defences succeed is, of course,
a judgment about which reasonable  people will differ, but we think
it is  clear that epistemicism  cannot, in view of  these defences,
simply be ignored.

Yet  in spite  of  epistemicism's large  and growing  philosophical
popularity, an epistemicist account of the  SIA has yet to be fully
developed in the literature. Our goal  here is to fill this gap. We
argue that on  an epistemicist vagueness-based account  of the SIA,
items are comparable in small improvement cases. In other words, if
epistemicism is true, in  small improvement cases, incomparabilists
confuse  our ignorance  of a  ranking with  the non-existence  of a
ranking.  (Note  that  our  argument only  establishes  that  items
are  comparable  in  small  improvement  cases,  and  therefore  is
compatible  with the  possibility  that items  are incomparable  in
other  cases not  involving  vagueness. For  example,  it might  be
argued  that  any  amount  of profound  artistic  contemplation  is
absolutely incomparable with any amount  of base pleasure.) This is
potentially important  for axiology. If epistemicism  is true, then
the SIA does  not show the betterness ordering to  be incomplete in
small improvement cases.

Our argument  is doubly  conditional: if all  hard cases  raised in
SIAs are borderline cases of  vague predicates, and if epistemicism
is true, then  items are comparable in small  improvement cases. We
make  no attempt  to defend  either  part of  the antecedent  here.
Nonetheless, as we have argued above, there is sufficient reason to
explore what follows if both parts of the antecedent are true.

The  paper  is structured  as  follows.  In  Section 2,  we  define
comparability, formalise  the SIA,  and define borderline  cases of
vague predicates.  In Section  3, we introduce  in more  depth, and
discuss the implications of, supervaluationism and epistemicism for
the  SIA. Since  epistemicism is  deductively logically  consistent
with both comparabilism and incomparabilism about small improvement
cases, we  develop and  discuss versions of  both, arguing  that we
ought to accept epistemic  comparabilism. Interestingly, one of the
arguments  we use  also  shows  that even  if  hard  cases are  not
borderline cases of vague predicates, if epistemicism is true, then
items cannot  be on  a par.  Lastly, we  consider the  treatment of
higher  order hard  cases by  supervaluationism, epistemicism  (our
preferred  theory), and  Chang's  parity view.  We  argue that,  as
things stand in the literature, only the epistemicist view provides
a compelling account of how to account for such cases.

## Comparability and the Small Improvement Argument

### Comparability

How one defines comparability depends upon certain assumptions, and
we  use a  definition  with which  some  philosophers disagree.  We
believe  that what  Ruth Chang  (2002) has  called the  'Trichotomy
Thesis'  is  true, and  will  take  it  for granted  here.  Chang's
argument against the  Trichotomy Thesis rests in part  on the claim
that  not all  SIAs can  be  explained by  contingent ignorance  or
vagueness. As we have mentioned  above, we believe that others have
provided decisive criticism of this view.[5]

 Trichotomy Thesis: Two items, x and y, are comparable in terms of
 their Fness, if and  only if x is Fer than, less F  than, or as F
 as y.

According to the Trichotomy Thesis, two options are incomparable in
terms of their  Fness if and only if  it is not true that  x is Fer
than, less F than, or as F as y. Comparabilism can be understood as
the  claim  that for  any  two  items in  F's  domain,  one of  the
trichotomy  of 'Fer  than,' 'less  F than,'  or 'as  F as'  applies
between them.  Incomparabilism is the denial  of comparabilism. For
instance,  a comparabilist  about the  creativity of  artists would
claim that, for  all pairs of artists, one is  either more creative
than,  less  creative  than,  or  as  creative  as  the  other.  An
incomparabilist about the creativity of artists would deny this.

Note that on  our definition, it is  sufficient for incomparabilism
that  all of  the components  of the  trichotomy are  not true.  On
some logics,  propositions can  be neither  true nor  false (Broome
1997).  We believe  that,  as  a semantic  matter,  if  all of  the
components of the  trichotomy are not true, that  is sufficient for
incomparability.  The  claim that  it  is  false  that one  of  the
components holds  is stronger (given  that false implies  not true)
and  so is  also sufficient.[6]  If all  of the  components of  the
trichotomy are not true between a pair of items, then the items are
not ordered  in terms  of their  Fness, and it  is the  ordering of
items in terms of Fness that  the concept of comparability ought to
capture.

One final thing to note about our definition is that it allows that
it might be  the case that between two comparable  items, we do not
or cannot know, in a trivial  sense, which of the components of the
trichotomy is true. Some such cases are straightforward. Suppose we
have to discern which of two carrots  is longer, but one of them is
in  a  locked  safe to  which  we  do  not  have access.  We  might
ordinarily say "the  length of these two  carrots is incomparable."
However, it is not, on the present understanding of 'incomparable.'
It is true  that one carrot is  longer than the other  or that they
are equally long. That we cannot  measure them does not change this
fact. They are, then, in  the relevant sense, comparable. We should
not  confuse incomparability  with ignorance  about the  ranking of
options.

### The Small Improvement Argument

We can now set out the  Small Improvement Argument. A range of SIAs
have been presented in the literature. Consider this version, which
we believe  makes the case  particularly strongly. Suppose  that we
are comparing a  painter, Francis, with a musician,  Kate, and that
it  seems as  though  it is  not  true that  one  is more  creative
than  the  other.  Now  consider Francis+,  who  is  slightly  more
creative than  Francis: suppose that  one of his  frescos manifests
slightly more creativity. If Francis  and Kate were exactly equally
creative, then Francis+ would be  more creative than Kate. However,
intuitively  it is  not true  that Francis+  is more  creative than
Kate, since a small improvement  to one of Francis's frescoes could
not tip the balance in this way. Therefore, Francis and Kate cannot
be equally creative. Therefore, it is not true that Francis is more
creative than, less  creative than, or as creative as  Kate. By the
Trichotomy Thesis,  Francis and Kate are  therefore incomparable in
terms of their creativity.

The SIA may be generalised and formalised as follows:

Premises

 P1. 'Equally F' is a transitive relation.

 P2. For some F, and some options x, x+, and y in F's domain:

 (a) It is not true that x is Fer than y

 (b) It is not true that y is Fer than x.

 (c) It is not true that x+ is Fer than y.

 (d) x+ is Fer than x.

 P3. If it is not true that x is Fer than or less F than y, and it
 is  not true  that  x and  y  are equally  F, then  x  and y  are
 incomparable with respect to F.

Argument

 [Suppose for reductio that]:

 4. x and y are equally F

 [From P1, P2 (d) and 4]:

 5. x+ is Fer than y.

 [But this contradicts P2 (c). So]:

 6. It is not true that x and y are equally F.

 [From P2 (a) and (b) and 6]:

 7. It is not true that x is Fer  than or less F than y, and it is
 not true that they are equally F.

 [From P3 and 7]:

 C. x and y are incomparable with respect to F.

Our  target is  P2. We  argue that,  assuming that  epistemicism is
true, it is not possible for P2 (a), (b), (c) and (d) to be true at
the same  time. If  epistemicism is true,  incomparabilists confuse
our ignorance of the truth of P2 (a), (b) and (c) with the truth of
all of P2 (a), (b), (c) and (d).[7]

### Hard cases and borderline cases

In our view, many SIAs which  have been presented in the literature
involve a  significant amount  of contingent  epistemic limitation,
not unlike our example of  the locked-away carrots. SIAs which only
involve contingent ignorance  are uncontroversially unsound because
they confuse our merely contingent  ignorance of a ranking with the
non-existence of  a ranking.  Similarly, other SIAs  exploit poorly
understood concepts.  It may sometimes  be the case,  for instance,
that two soldiers seem incomparable  in terms of their bravery, but
only  because we  do not  understand  the concept  of bravery  well
enough. However, it is widely  believedthat there remain hard cases
in  which no  amount of  further empirical  checking or  conceptual
analysis would allow us to determine which of the components of the
trichotomy  applies. The  creativity  SIA set  out  above seems  to
present an example of such a hard case.

As we  have said, we believe  that these hard cases  are borderline
cases  of  vague  predicates.  Before   we  proceed,  it  is  worth
being  clear  about what  it  means  to  say  that something  is  a
borderline  case  of  a  vague  predicate.  There  is  disagreement
both  about  the definition  of  vagueness  and of  borderlineness.
Sensitivity to sufficiently small changes but sensitivity to larger
ones  ('tolerance'), susceptibility  to  the  sorites paradox,  and
susceptibility  to  borderline  cases  have all  been  defended  as
definitive of  vagueness, and have  also been subject  to criticism
(Sorensen 1988; Greenough  2003; Bueno and Colyvan  2012). In light
of this controversy, we remain neutral on the sufficient conditions
for  something  to be  a  borderline  case  of a  vague  predicate.
However, we  follow the most  popular account of  borderlineness by
holding that  it is  a necessary condition  of borderline  cases of
vague  predicates that  they are  resistant to  inquiry (Williamson
1994,  2; Keefe  and Smith  1997; Sorensen  2015).[8] A  case is  a
borderline case  only if  we cannot  determine whether  a predicate
applies or does  not apply by reducing our  contingent ignorance or
by improving our conceptual competence.

## Vagueness and Ignorance

We will  now briefly  review the implications  of supervaluationism
for  the SIA  before  moving  on to  consider  the implications  of
epistemicism.

### Supervaluationism

Before  beginning,  we  should  make clear  that  what  follows  is
simply a  discussion, rather than  a critique, of  the implications
of  a  supervaluationist  account  of  the SIA  for  the  truth  of
incomparabilism. Our claim is that the implications of that view on
this question are ambiguous,  but we do not take this  to be a flaw
per se of  a supervaluationist account of the SIA.  We do, however,
also  happen  to believe  that  the  source  of this  ambiguity  --
supervaluationism's rejection  of classical logic and  semantics --
is  a good  reason to  disprefer supervaluationism  as a  theory of
vagueness  more generally.  But establishing  that goes  beyond the
scope of  our argument here, which  is conditional on the  truth of
epistemicism.

Supervaluationism is  the only  theory of  vagueness that  has been
fully developed in relation to comparability and the SIA, though it
nonetheless  has  an  ambiguous relationship  to  comparability.[9]
According to supervaluationism, in borderline cases 'x is Fer than,
less F  than or as  F as y'  is true, but  all of the  disjuncts of
this  disjunction are  neither true  nor false.  This is  explained
in  the  same  way  that the  supervaluationism  explains  the  law
of  the  excluded  middle  (Broome  1997,  82).  According  to  the
supervaluationist, vague concepts can  be admissibly 'sharpened' in
numerous ways.  For instance, some admissible  sharpenings of 'more
creative  than'  will  draw  the boundary  between  'more  creative
than' and  'not more  creative than' in  different places.  For the
borderline  case 'Francis  is  more creative  than  Kate,' on  some
admissible sharpenings, Francis is more creative than Kate, whereas
on others  he is not.  Supervaluationism dictates that  'Francis is
more  creative than  Kate'  is therefore  neither  true nor  false.
'Francis is more creative than Kate' is only true (or as it is also
said  in  the  literature  is 'supertrue')  when  Francis  is  more
creative than Kate on all admissible sharpenings.

However,  even  in  cases  when  'Francis  is  more  creative  than
Kate'  is  neither true  nor  false,  i.e.  is not  supertrue,  the
disjunction 'Francis is more creative  than, less creative than, or
as creative as Kate' is true on all sharpenings, i.e. is supertrue,
because all  sharpenings entail a complete  ordering. Therefore, on
supervaluationism, the disjunction is true, even though each of its
disjuncts is neither true nor false.

The truth  status of  the disjunction makes  supervaluationism seem
like  a  form  of  comparabilism  about  small  improvement  cases,
whereas  the  truth  status   of  the  individual  disjuncts  makes
supervaluationism seem like a  form of incomparabilism. This puzzle
is  explained  by  the  fact   that  the  intuitive  definition  of
comparabilism follows  the dictates  of classical  semantics, which
does not allow a true disjunction  to have no true disjuncts. Since
supervaluationism rejects this principle of classical semantics, it
is  not  clear  whether  supervaluationism is  a  comparabilist  or
incomparabilist theory about small improvement cases.[10]

This  ambiguity   can  also  be   brought  out  by   examining  the
formalisation  of the  SIA we  set out  above. In  our presentation
of  the argument,  we  included truth  predicates  in the  premises
themselves, and thus constructed  a 'metalinguistic' version of the
SIA. Here is P2 (a), (b) and (c):

 P2. For some F, and some options x, x+, and y in F's domain:

 (a) It is not true that x is Fer than y

 (b) It is not true that y is Fer than x.

 (c) It is not true that x+ is Fer than y.

We arrive at P2 (a), (b) and  (c) in the following way. Consider P2
(a). Because 'x is Fer than y'  is a borderline case, it is neither
true nor false that  x is Fer than y, and by  consequence it is not
true that x  is Fer than y.  The same reasoning applies  for P2 (b)
and (c).

On  classical  logic  and  semantics, merely  appending  the  truth
predicate to a premise cannot  affect the soundness of the argument
of  which the  premise  is a  part. For  instance,  if my  argument
depends on the premise

P. S

then revising the premise to instead claim that

P'. 'S' is true

would make no  difference to the argument's soundness.  But this is
not  so on  supervaluationism. According  to supervaluationism,  P2
(a), (b) and (c) are true, whereas versions of these premises which
do not include truth predicates would be neither true nor false. To
see this, consider  a non-metalinguistic version of  P2, P2', which
does not include truth predicates in the premises:

 P2'. For some F, and some options x, x+, and y in F's domain:

 (a) x is not Fer than y

 (b) y is not Fer than x

 (c) x+ is not Fer than y

Now consider P2' (a). On  supervaluationism, since P2' (a) concerns
a borderline case, it is neither true nor false. That is, in such a
case, 'x is not Fer than y'  is neither true nor false. But for the
SIA to be  sound, its premises must be true.  Similar remarks apply
to P2'(b)  and P2'(c). The  premise is therefore  false. Therefore,
supervaluationism entails the unsoundness of the non-metalinguistic
SIA.  By  contrast,  each  component (a-c)  of  the  metalinguistic
premise  P2  is  true  according  to  supervaluationism;  and  thus
supervaluationism supports the soundness of the metalinguistic SIA.

This is  another way of bringing  out supervaluationism's ambiguous
relationship  to  comparability, and  also  shows  why we  use  the
metalinguistic  version  of  the  SIA:   since  (or  so  we  argue)
epistemicism   implies   comparabilism,  the   metalinguistic   SIA
highlights the  fact that  there is a  distinction between  the way
that supervaluationism and epistemicism  treat some versions of the
SIA.

### Epistemicist comparabilism or incomparabilism?

We now turn to the  implications of epistemicism for comparability.
First,  we  will  provide  a  brief  account  of  epistemicism.  As
is  well known,  epistemicism claims  that borderline  propositions
are  classically truth-functional  but are  distinguished by  being
propositions whose truth  value cannot be known.  For instance, the
epistemicist claims that there is a  precise point as we take hairs
way at which 'Tom is bald'  becomes true where before it was false,
though we  do not and cannot  know where this point  is (Williamson
1994, 198-201).  The principle  of bivalence is  true: propositions
have  exactly one  truth value  --  true or  false; no  proposition
can  be  neither true  nor  false  (Williamson 1994,  187-89).  The
epistemicist explains the inquiry resistance of borderline cases by
appealing to our ignorance: no amount of further empirical checking
would enable us to know whether Tom is bald or not -- our epistemic
deficit is  non-contingent and incurable. Contingent  ignorance can
be resolved  by further  empirical checking, whereas,  according to
the  epistemicist, the  ignorance  manifested  in borderline  cases
cannot.

It might  appear that  epistemicism straightforwardly  entails that
options  are comparable  in small  improvement cases.  This is  not
so. Epistemicism  is (deductively)  logically compatible  with both
comparabilism  about small  improvement cases  and incomparabilism.
Consider  first   what  we  call   epistemicist  comparabilism.[11]
Epistemicist comparabilism says that for all x and y in F's domain,
x is either Fer than y, less F than y, or x and y are equally F and
that only one  of the disjuncts is true. For  example, suppose that
Francis is  slightly less  creative than Kate.  There is  then some
slightly  more creative  version of  Francis, Francis+,  who is  as
creative  as Kate,  and  a slightly  more  creative version  again,
Francis++, who is more creative  than Kate. However, we cannot know
the  truth value  of any  of these  three comparative  propositions
because the  trichotomy of 'more  creative than,' 'as  creative as'
and  'less creative  than'  are  vague. Epistemicist  comparabilism
posits a  single point of  precise equal creativity, and  says that
items either side  of this point are either more  creative than, or
less creative  than the items  used as the standard  of comparison.
Hence,  the comparative  ordering  of options  is complete  (unless
there is  some other  non-small improvement-based  argument showing
the ordering to be incomplete).

Epistemicist comparabilism can be  visually represented by Broome's
standard  configuration  (Broome 1997).  Suppose  that  there is  a
linear scale of  Fness, and that the standard, x,  is in the middle
of this  scale, and that  there is a  set of options  (y_1, y_2...,
y_(n)) ordered in terms  of their Fness. Epistemicist comparabilism
posits that only one of the  options in the y-set, y_(e) is exactly
equally as F as x. The subset of options that are less F than y are
also less F than  x, and the subset of options that  are Fer than y
are also Fer  than x. The grey box indicates  the set of borderline
cases for which  we are incurably ignorant which  of the components
of the trichotomy holds:

However, as  Broome (1999,  152) has  pointed out,  epistemicism is
also (deductively) logically  compatible with incomparabilism about
small improvement cases. Epistemicist incomparabilism posits a zone
of incomparability  sandwiched by two zones  of comparability. More
precisely, it  posits that  there is  a subset,  S_1, of  y options
which are Fer than  x, and a subset, S_3, which are  less F than x.
However, it  posits that, on  the scale  of Fness there  is another
subset, S_2, of y options sandwiched between S_1 and S_3, which are
neither Fer than nor less F than x. Provided the options in S_2 and
x are not equally F, the options in S_2 and x are incomparable with
respect  to F.  There  is  a sharp  transition  from  the zones  of
comparability into the zone  of incomparability. However, it cannot
be known when this transition  occurs because each component of the
trichotomy  is vague.  This  is incomparabilism  with an  incurable
epistemic deficit.

A first pass at  visually representing epistemicist incomparabilism
by the standard configuration is presented below. (As we discuss in
our criticism of epistemicist incomparabilism below, it is actually
subject to a qualifying  proviso, which makes visual representation
difficult.) Suppose  again that there  is a linear scale  of Fness,
and one item, x, is in the  middle of this scale, and that there is
a set  of options (y_1,  y_2..., y_(n))  ordered in terms  of their
Fness. The grey box indicates the set of borderline cases for which
we  are  incurably ignorant  of  which  of  the components  of  the
trichotomy holds:

Should  we   accept  epistemicist  comparabilism   or  epistemicist
incomparabilism?  We   cannot  think  of  any   reasons  to  favour
epistemicist incomparabilism over epistemicist comparabilism. There
are, however,  two abductive  reasons to believe  that epistemicist
comparabilism is true.

Firstly,  epistemicist  comparabilism  posits  a  single  point  of
precise  equality. The  epistemicist  incomparabilist treatment  of
equality, by  contrast, is problematic.  It appears that  there are
two possible treatments. The first would  be to claim that there is
a solitary  point within  the gap between  the application  of 'Fer
than' and 'less F than' (that  is, within S_2) at which options are
equally F and are therefore  comparable. However, it cannot be true
that there  is a solitary  point of  equality and that  options are
incomparable. If x and y_(e) are  equally F, then all other options
(apart from  y_(e)) in the  sequence of items (y_1,  y_2,... y_(n))
that are ordered in terms of their  Fness, must be Fer than or less
F than x. The  subset of options that are Fer  than y_(e) must also
be Fer than x because x and  y_(e) are equally F; and the subset of
options that  are less  F than  y_(e) must  also be  less F  than x
because  x and  y_(e) are  equally F.  Precise equality  crowds out
incomparability.

The second  way for epistemicist incomparabilism  to treat equality
would be to  claim that no pairs  of items can be  equally F within
the gap between  the zones of comparability (that  is, within S_2).
If this were  true, however, the question of whether  any two items
were equally  F could  never be inquiry  resistant, since  it would
have a determinate  answer ('no'), and therefore  there could never
be borderline cases of 'equally  F.' As discussed earlier, this has
the consequence  that 'equally  F' cannot  be vague.[12]  There are
reasons to believe that this is a mistake.

It seems  that the question of  whether predicates of the  form 'is
equally  F'  apply  sometimes  can be  inquiry  resistant.  Suppose
again  that  we   are  assessing  whether  Kate   and  Francis  are
equally  creative. Intuitively,  it  seems that  once  we have  all
the  information and  are  conceptually competent,  we might  still
be  unable  to settle  with  complete  confidence whether  the  two
are  equally creative;  we  would  not be  certain  that they  were
not  equally  creative. But  such  certainty  is what  epistemicist
incomparabilism requires, since on that view, the fully determinate
answer to whether any two items are equally F is 'no.' Epistemicist
comparabilism is,  by contrast, compatible with  the intuitive lack
of  certainty  about this  case.  Given  that  there is  no  deeper
theoretical reason to  accept epistemicist incomparabilism's stance
on  this  issue,  it  is  hard  to see  why  we  would  accept  the
incomparabilist  view. Indeed,  all  proponents of  vagueness-based
accounts of  the SIA have  hitherto assumed that predicates  of the
form 'is equally F' can  be vague: according to supervaluationists,
in borderline cases,  all components of the  trichotomy are neither
true nor false.  This suggests that there  is significant intuitive
support for the proposition that predicates of the form 'is equally
F' can be vague.

Secondly, epistemicist incomparabilism  as we have set  it out says
that there is  a zone of incomparability that is  sandwiched by two
zones of  comparability. However,  this characterisation  should be
qualified;  the  story  is  still more  complicated.  This  can  be
demonstrated by what we call the 'Monadic-Dyadic Argument.'

This  argument begins  by noting  that, according  to epistemicism,
there is what  we will call a precise 'monadic  threshold' at which
the  truth value  of a  statement  with a  vague monadic  predicate
switches from true  to false. This is, for  instance, the threshold
at  which,  as  his  creativity  improves,  'Francis  is  creative'
switches from  being false to true.  If this is true,  then in many
cases 'Fer  than' will  apply within  the gap in  which all  of the
components of the trichotomy are supposed  to be false. This can be
shown using the following principle:

 Monadic-dyadic principle: If x is F and y is not F, then x is Fer
 than y

We take  the monadic-dyadic principle  to be uncontroversial  for a
large class of  comparatives (though there may  be some exceptions,
which we return to below).[13] If  one person is creative, then she
is more creative than all people  who are not creative. Now imagine
a musician, Elton, and a singer,  Freddie, who stand just on either
side of the  monadic threshold of 'is creative':  Elton is creative
but Freddie is not.  Therefore, given the monadic-dyadic principle,
Elton  is more  creative  than Freddie.  However,  since Elton  and
Freddie are  very close to one  another in terms of  creativity and
manifest creativity in different ways, 'Elton is more creative than
Freddie' is also  a borderline case, and so we  cannot know whether
or not it is true. Therefore, 'Elton is more creative than Freddie'
is  true,  but we  cannot  know  whether or  not  it  is true.  Yet
according  to epistemic  incomparabilism, 'Elton  is more  creative
than Freddie'  is false  because, in virtue  of being  a borderline
comparative case, it lies within a zone of incomparability.

Thus, epistemicist incomparabilism must be qualified. It should say
that  for some  borderline cases  of some  vague comparatives,  two
items are  incomparable, except  when they are  either side  of the
monadic  threshold.[14] In  that case,  one  item is  Fer than  the
other, so the items are comparable. These cases, it is important to
note, will  be pervasive. There  is an infinite number  of possible
cases in which  two items are either side of  the monadic threshold
and are a  borderline case of the comparative form  of that monadic
predicate. Therefore, there are points  of comparability in what we
initially said must be a zone of incomparability.

Furthermore, if Elton  and Freddie stand only  marginally on either
side of  the monadic  threshold, then Elton  is only  slightly more
creative than Freddie. But if this is true, then it seems as though
the following must also be true: there is some small improvement to
Freddie's creativity  which would  make him  as creative  as Elton,
where before he  was not. But as we argued  above, precise equality
crowds out incomparability. If this line of argument is sound, then
epistemicist  incomparabilism  collapses  entirely.  There  may  be
other  ways to  qualify the  theory so  that it  allows pockets  of
comparability in the zone of incomparability, but doing so seems ad
hoc and lacks deeper theoretical or intuitive justification.

(It is worth noting in passing  that, if epistemicism is true, this
argument also counts against  tetrachotomist views, such as Chang's
parity view.  Chang would  argue that the  comparison of  Elton and
Freddie  is a  'superhard' case  of 'more  creative than'  and that
therefore they  are on a  par with respect to  creativity. However,
the Monadic-Dyadic Argument shows that in fact one is more creative
than  the other.  Thus, if  epistemicism  is true,  then there  are
points of comparability in what was  initially posited to be a zone
of parity. If epistemicism is true, this seems like a fatal flaw of
all  tetrachotomist  theories. Note  that  this  argument does  not
assume or imply that hard  cases of comparison are borderline cases
of vague dyadic  comparatives. Rather, it moves  from an assessment
of vague  monadic predicates to the  claim that items are  not on a
par in terms of a shared property. This is compatible with it being
true  that hard  cases of  comparison are  not borderline  cases of
vague dyadic comparatives.)

As   we   mentioned   above,  the   monadic-dyadic   principle   is
uncontroversial  for a  wide  range of  comparatives. However,  one
possible  response  to the  Monadic  Dyadic  Argument is  to  argue
that  the  monadic-dyadic principle  is  not  true for  axiological
betterness.[15]  There  is  a   great  deal  of  controversy  about
whether  goodness  can  be  reduced to  betterness  or  vice  versa
(Gustafsson  2014), and  some  of this  controversy  may extend  to
the monadic-dyadic  principle. For  example, Gustafsson  (2017) has
defended  'blank  critical   range  utilitarianism'  in  population
ethics, according  to which the monadic-dyadic  principle is false.
In  brief, on  this theory,  lives can  be good,  bad, neutral,  or
'blank'  for the  person living  them. If  a life  is blank,  it is
neither good, bad,  nor neutral. Due to the possibility  of a blank
life and the resultant  (non-vagueness-based) incompleteness in the
ranking of  the goodness of  lives, it  is possible that  some good
lives are not better than blank  lives, even though blank lives are
not good (Gustafsson 2017, 15).

While  it  is  true  that  the  Monadic-Dyadic  Argument  fails  if
Gustafsson's axiology  is correct,  we would offer  two rejoinders.
Firstly, if the  blank critical range axiology is  true, then there
are still  no reasons to believe  that epistemicist incomparabilism
rather than comparabilism is true -- we have merely lost one reason
that favours  comparabilism. And  if the  critical range  theory is
false and some other plausible axiology that is compatible with the
monadic dyadic principle is  true, then the Monadic-Dyadic Argument
alone is sufficient to refute epistemicist comparabilism. The blank
critical  range axiology,  moreover,  is  highly controversial,  so
epistemicist incomparabilism is also  for this reason less probable
than  epistemicist comparabilism.  This point  generalises for  all
controversial  axiologies which  are  putatively incompatible  with
the  monadic-dyadic principle,  not just  the blank  critical range
axiology.

Secondly, the monadic-dyadic principle can be amended to avoid this
counterexample. The  blank critical  range theory only  implies the
falsity of the monadic-dyadic principle as it applies to lives, but
not to  other items  that we  can compare  in terms  of axiological
betterness, such as  experiences. All that is  needed to decisively
refute  epistemicist incomparabilism  is  for some  version of  the
Monadic-Dyadic  Argument  to  commit the  incomparabilist  to  some
pockets  of  comparability  in  the  zone  of  incomparability.  In
general, it seems unlikely that  there could be an argument showing
that the  monadic-dyadic principle fails  to hold for  all possible
items that we can compare in terms of axiological betterness; it is
plausible that it  will always be possible to  develop some version
of  the  monadic-dyadic principle  which  is  sufficient to  refute
epistemicist incomparabilism.

In sum,  if we believe  epistemicism is true,  then incomparabilism
about small  improvement cases is  drained of all of  its intuitive
force. Incomparabilism was  initially intuitively plausible because
it provided  an explanation of  the phenomena in  small improvement
cases. If we can explain  those phenomena with appeal to unknowable
boundaries  and unknowable  precise  equality, there  is no  reason
to  accept  the  intuitive   and  theoretical  problems  that  come
with a  conjunction of  unknowable boundaries  and incomparabilism.
Unknowable  boundaries ought  to  be treated  as  a substitute  for
incomparability.[16]  If  epistemicism  is   true,  then  in  small
improvement  cases  incomparabilists  confuse our  ignorance  of  a
ranking with the lack of a ranking.

### Sharp boundaries and higher-order vagueness

There is one  other point about epistemicist  comparabilism that is
worth  making. Many  people take  the idea  of an  unknowable sharp
boundary to be so counter-intuitive  as to render epistemicism as a
wider theory of vagueness completely implausible. There must, these
people believe,  be some more plausible  non-epistemicist theory of
vagueness, which does not unambiguously entail comparabilism. Sharp
boundaries, on this view, are a liability.

In  fact,  however, this  intuition  provides  no leverage  against
epistemicism, as  things stand in the  literature on comparability.
This is because all existing  treatments of small improvement cases
posit sharp boundaries. The only difference is that they move them.
Not only that, having moved  them, they treat them differently from
the first-order  boundaries they  attempted to  avoid in  the first
place.

According to Broome's (1997) early form of supervaluationism, there
is  a  precise  and  knowable  answer to  the  question  'how  many
frescoes  must Francis  paint for  'Francis is  more creative  than
Kate'  to  shift  from  being  neither  true  nor  false  to  being
true?'[17] Similarly,  according to  Chang's parity view,  for some
predicates'Fer than'  and 'as F  as,' there is no  sharp transition
from 'x is  Fer than y' to 'x is  as F as y,' but there  is a sharp
transition from  'x is Fer than  y' to 'x is  on a par with  y with
respect to  F' which, it  also seems, is precise  and knowable.[18]
The problem is that it seems to  be just as hard to determine where
these boundaries are as it  does to determine where the first-order
boundaries are:  hard cases recur  at higher orders. If  Broome and
Chang were to  say that these boundaries are  sharp but unknowable,
then  they  would  be  committed  to  an  epistemicist  account  of
higher-order hard cases. But then, it  is difficult to see why they
would not just  accept an epistemicist account  of first-order hard
cases.

Thus, two  of epistemicism's main  rivals posit sharp  and knowable
boundaries.[19] This  seems at least  as unbelievable as  the sharp
and unknowable  boundary posited by epistemicist  comparabilism. It
is  the  unknowability  of  the  boundary  that  accounts  for  the
intuitive inquiry  resistance of  hard cases,  hard hard  cases and
so  on.  In this  respect,  existing  rivals  fail to  explain  the
phenomena. They treat higher-order  hard cases differently from the
first-order  cases  raised  by  the  SIA,  despite  the  apparently
identical phenomenology of first-order and higher-order cases.

By contrast,  epistemicists happily accept  higher-order vagueness.
For  the  epistemicist,  higher-order vagueness  is  understood  as
ignorance of our ignorance at lower orders of vagueness (Williamson
1994, 3). The zone of borderline cases has sharp boundaries, but we
cannot know where they are  because there are borderline borderline
cases. For instance,  we cannot know which is  the first borderline
case of the  monadic 'is bald' or the dyadic  'more creative than.'
Therefore, it seems  to be a dialectical  advantage of epistemicist
comparabilism that it bites the  first 'sharp boundary bullet' that
comes its way.

Could these  rival views adapt  to better account  for higher-order
cases? We find  it hard to see how Chang  could plausibly alter her
account to  deal with higher-order  hard cases. In accord  with the
spirit of parity,  she could propose that in the  hard cases of the
application of 'on  a par,' there is a zone  of 'super-parity.' But
such a move seems to be prima facie unattractive. Supervaluationism
seems to  be better  placed in  this regard  since supervaluational
accounts  of  higher-order vagueness  have  been  presented in  the
vagueness literature (Williamson 1999). This has, though, yet to be
done in  the literature on  comparability. Moreover, even if  it is
done in the  future, supervaluationism has to  posit knowable sharp
boundaries somewhere  (Sorensen 2001, 82-83). In  sum, epistemicist
comparabilism  is not  only worth  taking seriously  in the  debate
about the  SIA, it has  significant advantages over  existing rival
accounts.

It  is   true  that  some   philosophers  deny  the   existence  of
higher-order vagueness (Wright 2010).  Nonetheless, as things stand
in the  debate on comparability, epistemicist  comparabilism has an
advantage on the higher-order vagueness front. The phenomenology of
first-order borderline  cases and putative  higher-order borderline
cases does appear  to be the same: further inquiry  will not enable
us to know whether a  predicate applies at higher-orders.[20] Thus,
it  certainly  seems as  though  an  explanation is  required  from
proponents of non-epistemicist theories of comparability, which has
not  yet  been developed.  They  must  either explicitly  deny  the
possibility of higher-order vagueness or  explain why they treat it
differently from first-order vagueness.

## Conclusion

Many have  thought that the hard  cases raised in some  SIAs simply
cannot be  cases of ignorance. How  can they be, given  that all of
the relevant information  is in, yet the cases  remain resistant to
inquiry? Since they are not cases of ignorance, it must be the case
that each component  of the trichotomy fails  to hold. Epistemicism
about vagueness provides  an escape from this train  of thought. If
these hard cases  are borderline cases of vague  predicates, and if
epistemicism is true, then one  of the components of the trichotomy
is  true  but  we  cannot know  which.  The  unknowable  boundaries
posited by epistemicism force  us towards comparabilism about small
improvement cases. Furthermore, even if hard cases are not cases of
vagueness, if  epistemicism is  true, then options  cannot be  on a
par.

We have  argued for several  further conclusions as well.  For one,
epistemicist comparabilism has an important advantage over existing
rival accounts of the SIA. Epistemicist comparabilism appears to be
better equipped than  any existing rival theory to  account for all
hard  cases:  it can  account  for  hard  cases, hard  hard  cases,
and  so  on. Rival  theories,  on  the  other  hand, fail  to  cope
with higher-order  hard cases..  For another,  our argument  may be
practically important for axiology: many  have thought that the SIA
provides  one argument  for  the incompleteness  of the  betterness
ordering, by  showing that  some options  can be  incomparable with
respect to  goodness. But if  our thesis  is correct, then  the SIA
fails on this front.

## Acknowledgments

We would like  to thank Tim Williamson and John  Broome for helpful
comments on a  much earlier version of this paper.  We are grateful
to an  audience at the  White Rose Early Career  Ethics Researchers
Conference  at the  University  of York  for  helpful comments  and
suggestions,  and to  Roy  Sorensen for  advice. Johan  Gustafsson,
Michelle Hutchinson, and Marinella  Capriati also provided incisive
and  supererogatory comments.  We  would especially  like to  thank
Fabienne  Peter  and  two  anonymous  reviewers  at  Economics  and
Philosophy for their penetrating criticisms.

## References

Broome,   John.   1997.   Is   Incommensurability   Vagueness?   In
Incommensurability, Incomparability,  and Practical  Reasoning, ed.
Ruth  Chang. Cambridge,  Mass:  Harvard  University Press.  ------.
1999.  Ethics out  of  Economics.  Cambridge: Cambridge  University
Press.  ------. 2004.  Weighing  Lives.  Oxford: Oxford  University
Press.

Bueno, Otavio,  and Mark Colyvan.  2012. "Just What  Is Vagueness?"
Ratio 25 (1): 19-33. doi:10.1111/j.1467-9329.2011.00513.x.

Chang,   Ruth.   1997.   "Introduction."   In   Incommensurability,
Incomparability,  and  Practical  Reason,  edited  by  Ruth  Chang.
Cambridge, Mass;  London: Harvard  University Press.  ------. 2002.
"The Possibility of Parity." Ethics 112 (4): 659-88.

Dunaway, Billy. 2016. "Ethical  Vagueness and Practical Reasoning."
The Philosophical Quarterly. doi:10.1093/pq/pqw038.

Elson, Luke. 2014. "Heaps and  Chains: Is the Chaining Argument for
Parity a Sorites?" Ethics 124 (3): 557-71.

Espinoza, Nicolas. 2008. "The Small Improvement Argument." Synthese
165 (1): 127-139.

Greenough, Patrick.  2003. "Vagueness: A Minimal  Theory." Mind 112
(446): 235-81. doi:10.1093/mind/112.446.235.

Gustafsson, Johan E. 2013. "Indeterminacy and the Small-Improvement
Argument." Utilitas 25 (4): 433-45.

------. 2014. "Neither 'Good' in  Terms of 'Better' nor 'Better' in
Terms of 'Good.'" Nous 48 (3): 466-73. doi:10.1111/nous.12038.

------. 2017. "Population Axiology and  the Possibility of a Fourth
Category of Absolute Value."

Hawthorne,  John,  and  Andrew  McGonigal. 2008.  "The  Many  Minds
Account  of  Vagueness."  Philosophical Studies  138  (3):  435-40.
doi:10.1007/s11098-007-9110-3.

Huemer, Michael.  2013. "Transitivity,  Comparative Value,  and the
Methods of Ethics." Ethics 123 (2): 318-45. doi:10.1086/668905.

Hyde, Dominic. 1997. "From Heaps and  Gaps to Heaps of Gluts." Mind
106 (424): 641-660.

Keefe, Rosanna,  and Peter Smith. 1997.  "Introduction: Theories of
Vagueness."  In Vagueness:  A Reader,  by Rosanna  Keefe and  Peter
Smith. Cambridge, Mass; London: MIT Press.

Messerli,   Michael,   and   Kevin  Reuter.   2016.   "Hard   cases
of    comparison."    Philosophical    Studies.    Available    at:
doi:10.1007/s11098-016-0796-y.

Raz,  Joseph.  1986.  The   Morality  of  Freedom.  Oxford:  Oxford
University Press.

Rescher, Nicholas. 2009. Unknowability:  An Inquiry into the Limits
of Knowledge. Lanham, Md.: Lexington Books.

Sorensen,  Roy.  1988.  "Vagueness, Measurement,  and  Blurriness."
Synthese 75 (1): 45-82.-

-----. 2001. Vagueness and Contradiction. Oxford: Clarendon Press.

------.   2015.   "Vagueness."   In   The   Stanford   Encyclopedia
of   Philosophy,  edited   by   Edward  N.   Zalta,  Winter   2013.
http://plato.stanford.edu/archives/win2013/entries/vagueness/.

Sousa, Ronald B. de. 1974. "The  Good and the True." Mind 83 (332):
534-51.

Wasserman,   Ryan.   2004.   "Indeterminacy,  Ignorance   and   the
Possibility of Parity." Philosophical Perspectives 18 (1): 391-403.
doi:10.1111/j.1520-8583.2004.00034.x.

Williams, J. Robert G.  2016. "Indeterminacy, Angst and Conflicting
Values." Ratio 29 (4): 412-33. doi:10.1111/rati.12141.

Williamson, Timothy. 1994. Vagueness. London: Routledge.

------.    1997.   "Precis    of    Vagueness."   Philosophy    and
Phenomenological Research 57 (4): 921-28. doi:10.2307/2953810.

------. 1999.  "On the  Structure of Higher-Order  Vagueness." Mind
108 (429): 127-143.

Wright, Crispin. 2010. "The Illusion of Higher-Order Vagueness." In
Cuts and  Clouds: Vagueness,  Its Nature and  Its Logic,  edited by
Richard  Dietz and  Sebastiano Moruzzi.  Oxford: Oxford  University
Press.

Biographical Information

Edmund Tweedy  Flanigan is a  PhD Candidate in Political  Theory at
Harvard University. His doctoral  research focuses on political and
legal  obligation  and  authority.  He also  holds  an  M.Phil.  in
Political Theory from the University of Oxford.

John Halstead  is a  researcher at Founders  Pledge in  London,. He
holds a DPhil  in Political Philosophy from  Oxford, and previously
taught at the Blavatnik School of Government.

## Notes

[1] The argument was initially introduced in (de Sousa 1974). For a
classic discussion and overview of the SIA, see (Chang 1997).

[2] We explain why this is in section 3.

[3] This  is but one  of many criticisms of  supervaluationism. For
classic texts both critical and  in favor of supervaluationism, see
(Williamson 1994; Keefe and Smith 1997; Sorensen 2001).

[4] A number of  other prominent philosophers endorse epistemicism,
including John  Hawthorne (Hawthorne and McGonigal  2008), Nicholas
Rescher (2009),  and Patrick  Greenough (2003). Dunaway  (2016) has
explored the implications of epistemicism for practical reasoning.

[5] Elson (2014) provides strong  criticism of another crucial part
of Chang's argument against the Trichotomy Thesis. In section 3, we
show  that even  if  hard  cases are  not  cases  of vagueness,  if
epistemicism is true,  then there are very strong  reasons to doubt
the possibility of  parity, and that parity struggles  to deal with
higher-order hard cases.

[6] Some  philosophers argue that borderline  propositions are both
true and false (Hyde 1997). We bracket these accounts here.

[7]  Note  that   in  order  to  highlight   the  contrast  between
epistemicism and supervaluationism, we include the truth predicates
in the premises  themselves. We explain why we do  this in the next
section.

[8] One  of the virtues of  this necessary condition is  that it is
ecumenical between  different theories of vagueness.  The condition
is not  sufficient because  there appear  to be  other propositions
that are resistant  to inquiry, but which are  not borderline cases
of vagueness. For example, Goldbach's Conjecture might be resistant
to inquiry, but  this has nothing to do  with vagueness (Williamson
1997, 926).

[9] Broome  (1997) first  developed supervaluationism  with respect
to  comparability.  His  discussion   at  pp.  88-89  reflects  the
ambiguity  we  discuss here.  Some  philosophers  have argued  that
supervaluationism is inconsistent  with incomparability (Gustafsson
2013; Espinoza 2008).

[10]   In  his   later   work,  Broome   develops   a  version   of
supervaluationism  which is  about  assertability  rather than  the
truth.  He  argues  that  the   new  theory  is  incompatible  with
incomparabilism (Broome 2004, chap. 12  and 14). However, since the
new  theory  is about  assertability,  and  the theory  severs  the
logical  connection  between assertability  and  truth,  it has  no
implications  for  the  SIA  or  incomparabilism  (or  the  sorites
paradox).

[11]  Note again  that  this  theory only  implies  that items  are
comparable in small improvement cases, and is compatible with there
being incomparability for other reasons.

[12]  This is  intuitive, but  the explanation  for it  is somewhat
cumbersome. For all pairs of  items and all comparative predicates,
there  are   two  possibilities.  Firstly,  there   are  incomplete
orderings, characterised by vagueness, for  which there is one item
x, and a  subset of items, S_2 (y_(k),  y_(k+1)..., y_(n)), ordered
in terms of their Fness, and all members of this subset are neither
Fer than nor  less F than x.  For each comparison between  x and an
item in  S_2, we  know that  equally F  never applies.  Thus, these
cases  are  not inquiry  resistant.  Secondly,  there are  complete
orderings that are not characterised by vagueness. For all pairs of
items  in these  orderings, if  we had  perfect knowledge  and were
conceptually  competent,  then  reducing contingent  ignorance  and
improving conceptual competence would enable  us to know whether or
not 'equally F' applies.

[13] For example, does the fact that Nigel is dead and Brian is not
dead entail that Nigel is deader  than Brian? Our intuition is that
this  comparative  is  nonsensical,  but others  may  disagree.  We
discuss more difficult  cases below, but we  claim nonetheless that
the principle holds for a very wide range of comparatives.

[14] Of course, not all borderline cases of comparatives are either
side of the monadic threshold. For example, Mozart and Michelangelo
were both creative. But this is  a point about the structure of the
epistemicist incomparabilist view.

[15]  We thank  a  reviewer  for pushing  us  on  this point.  That
reviewer  also suggested  the  following  counterexample: Assume  a
hedonic axiology and imagine two states of affairs, S[Jane is happy
to  degree 10]  and S*[Jane  is happy  to degree  20 or  unhappy to
degree 10]. S, let  us say, is a good state;  and S*, plausibly, is
not. But because  of the uncertainty of S*, it  is not obvious that
S  is  better than  S*.  Indeed  they  may seem  incomparable.  Our
response to this  is chiefly that we take  relations of axiological
betterness to hold  between states of affairs,  rather than between
uncertain prospects, and  therefore that this cannot  show that the
monadic-dyadic  principle  fails  to  hold, as  a  principle  about
axiological  betterness.  (This is  only  one  of several  possible
rejoinders; if S*  is instead understood as a  disjunctive state of
affairs, we would reply that we hesitate to accept the existence of
such  states.) Nonetheless,  as we  discuss  in the  main text,  we
recognise  that on  some plausible  axiologies, the  monadic-dyadic
principle may be false.

[16]  Moreover, if  epistemicism is  true, then  our argument  here
shows that 'comparable  to' is not vague. If  epistemicism is true,
we know that  one of the components of the  trichotomy applies. So,
'comparable to' has no borderline cases and is therefore not vague.
Elson  (2014) discusses  the  possibility that  'comparable to'  is
vague.

[17]  In  his  later  work,   Broome  acknowledges  that  there  is
higher-order vagueness, but says he excludes it from his theory for
the sake of simplicity (Broome 2004,  180-81). It is not clear that
this  is justifiable.  After  all, we  could  rule our  first-order
borderline cases for the sake of  simplicity as well, but we do not
do so because we want to account for the phenomena.

[18]  Chang  argues  that incomparabilism  is  implausible  because
it  says  that  there  is  a precise  transition  from  a  zone  of
comparability  to  one  of incomparability  (Chang  2002,  673-79).
Chang's theory has  an exactly analogous feature.  For criticism of
Chang's argument see (Elson 2014).

[19] Raz is not explicit about this, but the same seems to apply to
his  arguments for  hard  incomparability  and for  incomparability
grounded in semantic indeterminacy (Raz 1986, chap. 13).

[20] Indeed,  as we  mentioned above,  Broome already  accepts that
there is higher-order vagueness, and Chang argues that there cannot
be a  sharp transition from  a zone of  comparability to a  zone of
incomparability.  Thus, she  accepts  something very  close to  the
intuition underlying higher-order vagueness.