Comments to the Author

This paper proposes a number of ideas in support of approval voting. The paper does not
recall any of the arguments potentially raised "in favor" of approval voting in the literature,
but suggests to augment it by three "merits" of AV. Firstly, AV avoids the problem of
independence of irrelevant alternatives raised by Arrow. Secondly, AV does not enable using
agenda-setting to produce results that would be inconsistent with majoritarianism or
pluritarianism. Thirdly, AV is not subject to majority cycles.

Every question based on promising intuitions is worth investigating; however, not all the
arguments developed in the paper are entirely convincing, and the conclusions are
unfortunately not as conclusive as pretended. It is quite uncomfortable to review a paper
submitted in a scientific journal in which a number of distinct intuitions are presented as
personal deep convictions, introduced without any link with existing literature on the issues
at stake, and in which the proofs are sketched on limited cases, rather than based on sound
methods standardly used to tackle every cases.

Although one would expect less unproved subjective and normative assertions in a scientific
paper, I have very much appreciated that the author makes clear from the beginning which
properties of voting rules she or he values, allowing readers to not concur (From now on, I'll
use the epicene form "the authors"). The authors notably appreciate that voters faithfully
follows instructions for voting; certain rule may more or less incite them to do so. They shall
retain the assumption that voters will follow instructions to vote in AV for every candidate
they would appreciate to actually be elected.

Second part seems to concern alpha-consistency (as in Sen) rather than IIA (as in Arrow):
this could be straightforwardly be tackled as such, whereas this becomes explicit after a
couple of pages. Notations (Ox or X) are varying and specification misleading (why is time
needed while the issue concerns different contexts and not different stages of contexts or
preferences), and we hardly understand why this question is not considered from the point
of view of function of choices over various opportunity sets. As a result, the argument and
the conclusion appear quite fragile. Reformulating the same argument in a short and
straightforward demonstration for both versions of the consistency condition would certainly
be more convincing.

Third part considers that, when a voter votes in approval voting, he may vote for every
candidates he would fancy being elected (as assumed from the beginning), but that a
number of debates concerning approval behaviors may raise.

p.1, The authors defend their view that "aggregating principle's should capture "group
preferences in worlds in which voting rules were strictly followed". We understand it would
be nasty from part of the voter to vote in favor of x when one prefers y, which precludes any
strategic voting. This idea may also apply to approval voting (however necessarily at a
different level): a voter should approve x and y when she would appreciate x or y to be
elected, disregarding whether she likes x better than y.

Despite this apparently clear distinction between approvable and not approvable
candidates, it is not always obvious to identify how many candidates should eventually be
approved for given preferences but for different contexts. The paper studies with a couple of
(outstanding) examples whether the number of approved candidates would or would not
matter, and concludes "such distortions of democracy cannot occur under AV."

The basis of this view is controversial. An alternative view is that we can hardly believe that
there is such thing as a natural cut for each voter between the approvable candidates and
the non-approvable candidates ; that we do not even know whether individual preferences
are actually / naturally pre-orders or dichotomous preferences. A voter who would concur
that two candidates could be acceptable winner (hence be approved of) may well have
preferences for one candidate (x) over the other one (y). Depending on the voting context,
approving y and x may elect y, while approving only x may elect x. As voting in an
aggregating process requires to contribute to a collective process, a good way of respecting
the rules of the voting method and influence the result may precisely be to vote
strategically, i.e. to vote for x or x and y depending on the context.

An important contrast between what the paper defends and this perspective (i.e. the huge
literature in social choice theory) is that the latter supports that strategic voting, by
construction, is inherent to any rules including AV (whichever our subjective judgment over
strategic behaviors), while the former qualifies it as a "distortion of democracy".

A second contrast with this literature is that computational social choice theory speaks of k-
level approval voting to summarize the issue: one individual may approve of 1,2,3, … n
candidates and eventually approves exactly k candidates. See for instance the Handbook of
Computational social choice theory by Ulle Endriss et al. From then on, these papers have
analyzed all sorts of paradoxes in every cases; they notably compute the probability of
incentive to strategic behavior vs. strategy proofness in every potential contexts, every
potential n and k, and many diverse assumptions on preferences profile; some papers also
consider the case of majority cycles (as rapidly sketched in the fourth part).

The least we could expect is that a paper which tackles these problem mention that such
results already exist, and describes how this paper does augments this existing literature.

More generally, the common knowledge in social choice theory is that the arrovian
impossibility (including cycles) is raised by the focus on ordinal preferences (rankings as pre-
orders) and the impossibility of comparisons. Going beyond these restriction will almost
inevitably result in a solution, among which any inclusion of some cardinality – as in AV, or
as in evaluative procedures in compliance with utilitarianism.

For all these reasons, I can hardly assert the paper provides novel insights on approval
voting. However, the illustrations chosen in this submitted paper are particularly relevant
and meaningful, and it might be worth insisting on their particular contributions to the
existing literature. The authors do have a potential to make their different (important)
arguments clear to a number of readers, in so far as the formal and computational literature
may sound quite indigestible to many.

Although I cannot support the publication of this paper, I would suggest, if I may, that every
argument could now be taken separately. Separate thorough contributions where the
existing literature is even minimally recalled, but where their results are properly analyzed
and nicely illustrated as it is done here, would be (salutary) welcome in scientific
publications.
***************************************************************************

This paper argues that Approval Voting (AV) has three desirable characteristics: independence
from irrelevant alternatives, resistance to strategic agenda-setting, and resilience against
majority cycles. Section II, on Independence of Irrelevant Alternatives (IIA), examines different
definitions of IIA and whether AV conforms to those definitions. No formal proofs or references
to other work that proves the conjectures that are made is offered. Section III considers a
classic example and states a theorem. No proof or references to other work that proves this
conjecture is offered. Section IV, considers an example. It offers no general conjecture.
As a paper in the Social Choice theory literature, I found this to be weak, and do not support
publication. The conjectures made in this paper are found elsewhere, especially in the work of
Brams and Merrill. No formal proofs are offered, and one section (on majority cycles) consists
of an example. In addition, other standards for evaluation creep into the presentation. For
example, at several points (e.g., pages 8-9) the text appeals to a standard of “rationality,” but
does not define what rationality is. Do you mean strict transitivity of individuals’ preferences?
Were this paper in another journal, such as Public Choice or Social Choice and Theory, these
weaknesses would be sufficient for rejection.

This paper, however, may be aiming for a different audience than those who work on social
choice theory, namely, the legal scholars and social scientists who read the Election Law
Journal. In this respect, I found the paper needing more of an argument. Is the author
advocating for this mode of voting? Is this paper arising because of increased interest in AV in
practical settings, such as local elections or local budgeting? Is there an obvious alternative
system against which this is compared, such as first-past-the-post or instant runoff voting?
And, do those other systems also have these same characteristics.

There is a massive literature on Approval Voting and also on IIA, both in individual choice theory
from psychology and in social choice. And, there is a massive literature on social choice and
alternative voting systems. It would be unreasonable to expect a paper to engage fully with
these literatures. However, I found this paper to be extremely light on engagement with the
literature. The paper needs to do a better job situating the discussion in the general literature.
Is the author responding to a specific line of critique of AV? If so, what exactly is this critique
and how do the conjectures presented in the paper respond to that critique?

As an essay on AV and social choice theory for a legal or general audience, I found the essay
lacking in justification of standards and connection to ideas that motivate the paper. Why are
these three characteristics particularly desirable? What are the undesirable characteristics of
AV, and why would these three characteristics make AV desirable even in the face of its
undesirable characteristics?

The concept of IIA finds much of its justification in Arrow’s original work on this subject. Arrow
and, later Gibbard and Satterthwaite, proved that no voting system has the following
properties: universal domain, transitive ranking, Pareto principle, IIA, and non-dictatorship
(non-manipulability). No one of these criteria in social choice theory is privileged over another.
Yet, this paper, by putting forth 3 criteria seems to be favoring at least one of these (IIA) over
others. If IIA was the criterion by which voting systems are chosen, then social choice theory
would not have faced the problems it has. Sen, Schwartz and others have weakened and
generalized these criteria and found that there are a wide class of similar “impossibility
theorems.” And, other authors have proposed other criteria. Saari and VanNewenhizen (1987)
criticize AV for leading to indeterminacy. AV may have all of these desirable properties, but
reaching a decision may not be one of them. Cox (1984) and Green-Armitage (2010 and 2014)
show that AV is susceptible to insincere and strategic voting (as distinct from strategic agenda
manipulation). Cox (1985) shows that AV has a strong median voter result (even in multiple
dimensions), so even if it is resistant to cycling it is even more likely to suffer from violations of
non-dictatorship. The present paper describes AV from the perspective of a couple of
properties, but does not situate it in a more extensive set of values that normally are
considered in social choice theory. The paper reads very much as a fragment of the
consideration of AV, rather than as a complete treatment for a broader audience.

All of that literature is, of course, in the shadow of Kenneth Arrow and the principles he
enumerated. I found the paper to be disconnected from recent work on social choice theory
that seeks to view voting from a different set of principles. Notably, Patty and Penn (2014)
argue that legitimacy and stability are principles through which voting systems should be
evaluated. The paper, then, is rooted in the Arrowian tradition, and I do not discern a
substantial contribution to that line of inquiry. It would be very interesting to assess AV
through newer lenses, such as Patty and Penn, but that is not the enterprise undertaken in this
paper.

In sum, I did not find that this paper breaks any new ground regarding AV or other voting
systems; nor does it engage sufficiently with specific problems of voting and the literature on
voting to offer an essay that would be of general interest to the election law community.