Democracy, Borders, & Legitimacy This is an argument about the relationship between democratic inclusion and instrumental justifications of democracy. I show that any account of democracy that relies on the outcomes of democracy processes to demonstrate democracy's value must have a congruent account of inclusion. Not only that, but because different accounts of democracy rely on different and incompatible accounts of inclusion, these accounts of democracy are themselves incompatible. This paper is unique in that is simultaneously a philosophical argument and a computer simulation. Written in literate coffeescript, the code described in the paper can also be run by the coffeescript compiler to demonstrate the argument being described by the paper. The reader may generate both the argument in PDF, or the simulation and results in HTML with the command coffee paper.coffee.md . Best viewed in HTML with interactive graphs, the following links outline paper, installation, and dependency requirements. A PDF version is also available with static images. ― Introduction ― There are many ways one might justify democracy. One common approach is to point to the desirable outcomes that democratic procedures bring about. Democracy in this light leads to good policy. Sometimes the desirability of these outcomes is judged according to some independent criteria. According to these accounts, there is some external standard for measuring the quality of a decision. Epistemic accounts of democracy like Condorcet's Jury Theorem or David Estund's Epistemic Proceduralism are like this. Truth exists independently of our beliefs; democratic processes track the truth; so this gives us reason to value democracy. Other times, we might judge the quality of democratic outcomes against some agent relative criteria. Rather than rely on something external, the desirability of a particular result is assessed against some internal standard. It isn't the collective decisions of democracy that matter per se but how those collective decisions correspond with individual choices. We see this in utilitarian justifications of democracy where majority voting maximises the expected utility of voter preferences. But one needn't be a utilitarian to employ such an approach - Jean-Jacques Rousseau, for example, argued that majority rule realises the general will of the people and this gives us reasons value it. And for others still, it is not the particular content of democratic decision making that matters so much as the e!ects that democratic processes have on the participants. Democracy changes people. This is what John Stuart Mill, had in mind when he argued that democracy transforms the moral character of its participants. Democratic participation makes good citizens. 1 2 3 4 5 6
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Democracy, Borders, & Legitimacy
This is an argument about the relationship between democratic inclusion and instrumental justifications
of democracy. I show that any account of democracy that relies on the outcomes of democracy processes
to demonstrate democracy's value must have a congruent account of inclusion. Not only that, but
because different accounts of democracy rely on different and incompatible accounts of inclusion, these
accounts of democracy are themselves incompatible.
This paper is unique in that is simultaneously a philosophical argument and a computer simulation.
Written in literate coffeescript, the code described in the paper can also be run by the coffeescript compiler
to demonstrate the argument being described by the paper. The reader may generate both the argument in
PDF, or the simulation and results in HTML with the command coffee paper.coffee.md .
Best viewed in HTML with interactive graphs, the following links outline paper, installation, and
dependency requirements. A PDF version is also available with static images.
― Introduction ―
There are many ways one might justify democracy. One common approach is to point to the desirable
outcomes that democratic procedures bring about. Democracy in this light leads to good policy.
Sometimes the desirability of these outcomes is judged according to some independent criteria.
According to these accounts, there is some external standard for measuring the quality of a decision.
Epistemic accounts of democracy like Condorcet's Jury Theorem or David Estund's Epistemic
Proceduralism are like this. Truth exists independently of our beliefs; democratic processes track the
truth; so this gives us reason to value democracy.
Other times, we might judge the quality of democratic outcomes against some agent relative criteria.
Rather than rely on something external, the desirability of a particular result is assessed against some
internal standard. It isn't the collective decisions of democracy that matter per se but how those
collective decisions correspond with individual choices. We see this in utilitarian justifications of
democracy where majority voting maximises the expected utility of voter preferences. But one needn't
be a utilitarian to employ such an approach - Jean-Jacques Rousseau, for example, argued that majority
rule realises the general will of the people and this gives us reasons value it.
And for others still, it is not the particular content of democratic decision making that matters so much
as the e!ects that democratic processes have on the participants. Democracy changes people. This is
what John Stuart Mill, had in mind when he argued that democracy transforms the moral character of
its participants. Democratic participation makes good citizens.
Spaces are partitioned into polities. A polity represents a unit of political association that holds some
degree of sovereignty regarding specific issues, such as a nation-state, province, or local council. Most
political spaces are partitioned in some way - the world is divided into countries, countries are divided
into states or provinces, states are divided into electorates etc.
How we partition a space - how we decide who will be included in which political association - forms
the crux of the Boundary Problem and there are many competing theories concerning how to partition.
Amongst accounts of democratic inclusion we find proposals to group agents according to nationality,
cultural or linguistic salience, degree of economic or social interdependence, by who is a!ected by a
policy or issue, or even to not partition at all.
While numerous accounts of inclusion exist in political theory, the combination of actual possible
partitions of any space is orders of magnitude greater. To simplify the model and remain agnostic
about particular theories of inclusion while capturing a wide variety of possible agent compositions, a
stochastic algorithm to divide the space into di!erent polities will be used to generate a random sample
of polities.
The partitioning algorithm recursively divides the largest polity of the space at a random point until
the desired number of polities have been produced - each characterised by a di!ering number and
composition of agents.
Space::partition = (k) -> @polities = [@agents] while k > 1 k = Math.min k, @agents.length polity = @polities.shift() cut = Math.floor( Math.random() * (polity.length - k) ) + 1 @polities.push polity[...cut] @polities.push polity[cut..] @polities.sort (a,b) -> b.length-a.length k-- this
Democracies make collective decisions - it's why they exist. The decision procedure for our model
democracy will be a naive majority vote by each polity on a binary issue. This is the simplest decision
mechanism to model and it will assume that agents vote sincerely and deterministically according to
their belief.
vote = (polity) -> votes = {} for agent in polity if votes.hasOwnProperty agent.belief then votes[agent.belief]++ else votes[agent.belief] = 1 max = Math.max.apply null, (num for belief, num of votes) votes.winner = belief for belief, num of votes when num is max votes
With our model now defined, the relationship between accounts of democracy and accounts of
inclusion can now be explored by running Monte Carlo simulations of the model for various
combinations of agent, clusterings, and partitions.
At the beginning of each simulation, a space is created with a fixed agent profile, clustering factor, and
polity number. During each run, the space is partitioned into the desired number of polities which then
vote, with the results being recorded for statistical analysis.
This partition-vote-measure loop is repeated 1000 times, generating in a probability density function
17
18
19
of the votes in each polity for each belief-cluster-partition tuple. These results are then assessed against
three classes of instrumental accounts of democracy to examine the conditions under which accounts
of inclusion a!ect accounts of democracy.
― Content Independent Outcomes ―
One common justification for democracy is epistemic - that democracy has instrumental value as a
truth tracking processes. Some of these epistemic accounts are relatively simple and Condorcet's Jury
Theorem is one such example. Given a better-than-even chance of any voter being correct on some
choice, then the likelihood of a majority vote being being correct on that choice approaches certainty as
the number of voters increases.
Others, like Estlund 's Epistemic Proceduralism, are more nuanced. Rather than valuing democracy on
the basis of the outcome of a specific vote, democracy has legitimacy because much like a jury, its
decision processes have a tendency to produce correct decisions. Even when the majority is wrong on a
particular matter, the majority decision is still morally binding, so long as majority voting remains the
more reliable procedure than alternatives.
While di!erent, all epistemic accounts rely on the claim that democratic procedures, on average, are
better at determining the correct result than alternate ones. Further more, these results are correct
independently of the decision procedure used.
This type of justification requires some way of comparing the epistemic performance of di!erent
decision procedures and one intuitive way to do this is to assess them against some base line metric. An
obvious candidate for such a metric is the likelihood that any randomly selected voter holds the correct
belief or votes correctly - what I will call the epistemic base rate of a political space. We can then judge the
epistemic performance of a decision procedure against the epistemic base rate. Call this di!erence a
procedure's epistemic virtue.
In our simulation, the epistemic virtue of a simple majority vote can be measured as the frequency
based probability that a polity votes correctly given the initial conditions of the space. During each
trial, the number of polities that voted correctly is measured and the impact of di!ering partitions
assessed. The correct answer to the question our agents vote on is known in advance because we
stipulate it when we create agents with either right or wrong belief.
Space::virtue = () -> correctVotes = 0 for polity in @polities election = vote polity correctVotes += 1 if election.winner is 'right' correctVotes / @polities.length
Running a Monte Carlo simulation hundreds of thousands of times for a range of partition numbers (5
20
21
to 30), cluster factors (0.0 to 1.0) and epistemic base rates (0.5 to 1.0) yields a four dimensional data cube
of probability distributions for the expected correct majority vote in each base-rate-partition-cluster
tuple.
We can examine the results from any perspective or slice of the data cube. In the graph below, we see
the impact of repartitioning on epistemic virtue at an epistemic base rate of 0.6 from the perspective of
clustering. The vertical axis represents epistemic virtue and the horizontal axis, the degree of
clustering. Each line represents the number of polities the space was partitioned into, ranging from 5 to
30 partitions.
The likelihood of any randomly selected voter in the space being correct is 0.6. When agents are
uniformly distributed across the space by belief (i.e. no clustering of agent belief is present), the
epistemic virtue of majority voting - the likelihood that the majority vote of any polity is the correct
choice - is very high (0.82-0.97). This quickly deteriorates as clustering of agent belief increases
however, with no epistemic virtue of majority voting evident once agent clustering reaches 0.5. This
holds true for all levels of partition numbers and epistemic base rates > 0.5, although the e!ect
diminishes as diversity across the space diminishes.
Examining the same data from the perspective of partition number, we see little impact of partition
number on epistemic virtue for higher cluster factors, and only limited impact for low levels of
clustering, concordant with Condorcet's Theorem. This relationship holds for all epistemic base rates
between 0.51 and 0.99.
These results indicate that the epistemic virtue of majority voting is dependent not only on individual
agent belief having a greater than 50% likelihood of being correct, but also on how those agent beliefs
are distributed across the political space. A key stipulation of the Jury Theorem is that there must a
better than average chance of any voter being correct - we might call this the competency requirement. This
simulation shows that when clustering of agent belief is present, even if the competency requirement
is met for the space as a whole, it is not necessarily met for every partition of that space. Furthermore,
as clustering of agent belief across the space increases, the likelihood that every partition of the space
satisfies the competency criteria decreases.
The relationship between content-independent justifications of democracy and democratic inclusion is
now clearer. Epistemic justifications of democracy require accounts of inclusion that ensures that:
1. Polities are su#ciently large for the Condorcet e!ect to emerge.
2. Any partition of a space that satisfied the competency requirement must also satisfy the
competency requirement.
Base Rate 0.60
5 10 15 20 25 30partitions
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
0.85
0.90
0.95
1.00
epis
tem
ic v
irtue
Epistemic virtue by partition number
23
When the distribution of voter belief is uniform, where we draw the boundaries of our democracies is
largely irrelevant for content-independent accounts of democracy. Any account of democratic
inclusion will do. But when the distribution of voter belief is clustered, where we draw boundaries of
our democracies becomes very important. In this case, any compatible account of democratic inclusion
will need to demonstrate that each polity satisfies the competency requirement for epistemic
justifications.
We can summarise this finding by stating that the Boundary Problem only becomes a problem for
content-independent justifications of democracy when homogeneity of voter belief or competence is
high within polities but low between them. In these instances, how we bound the demos and draw
political borders is critical. Content-independent justifications require accounts of inclusion that
generate su#ciently large, externally homogeneous polities. Polities needs to be large and have similar
voter composition otherwise only some of them can be justified by content-independent accounts.
― Content Relative Outcomes ―
Not all instrumental accounts of democracy are epistemic however. Utilitarians justify democracy on
the grounds that it promotes the greatest happiness. Majority voting maximises the expected utility of
voter preferences when each individual has an equal chance of preferring each of two alternatives.
But one needn't be a card carrying utilitarian to employ such an approach. Rousseau argued that
majority rule realises the general will of the people and this gives us reasons to obey, while social choice
theorists hold that majority voting realises individual choice when collectively binding decisions must
be made (Kenneth O May, ).
While these accounts of democracy di!er in many ways, they all share a similarity in that the value of
democracy stems from some content-relative criteria - of fidelity between individual preference and
collective outcomes. It is not the contents of the outcome of a democratic process that matters per se,
but rather how well this collective outcome matches the wants, preferences, or intent of individual
participants.
We can judge these content-relative outcomes by defining the fidelity of a democratic procedure as the
likelihood that an individual's preference is the same as, or compatible with, the majority outcome.
Formalising fidelity as individual-collective choice equivalence we get:
Space::fidelity = () -> winners = 0 population = 0 for polity in @polities election = vote polity for key, val of election winners += val if key is election.winner population += val unless key is 'winner' winners / population
24
25
26
Running the same Monte Carlo simulation with the same parameters as the epistemic simulation yields
a di!erent set of results to those of the content-independent one. Below, we see the fidelity of
individual preference to majority vote for a distribution of agents with a 60:40 preference for some
choice A or B over a range of clustering and partition variables, viewed by degree of clustering.
Again, when viewed by degree of clustering, the e!ect of polity composition on preference realisation
is stark. When agents are uniformly distributed by preference across the space (i.e. when clustering is
low), the likelihood of an individual's preference being realised by majority vote is identical to that of
any two random agents preference being the same (i.e. the preference base rate).
As clustering of agent preferences across the space increases however, the fidelity between individual
and majority preference increases significantly. At its most extreme, there is near certainty that any
individual preference will be realised by a majority vote of a polity when the distribution of agents
across the political space is fully clustered - when agents are completely segregated by preference. This
relationship holds for all preference base rates, although it is more pronounced when the preference
base rate - the ratio of competing preferences - is lower.
In contrast with the epistemic simulation of democracy however, the impact of agent clustering on
outcome quality is reversed. Majority voting has the greatest likelihood of fidelity with individual
preference, and therefore greatest value from a content-relative perspective, when agents are highly
clustered. This contrasts sharply with the content-independent perspective where the greatest
for b in bases b = 500 + b*50 for p in partitions p = p * 5 for c in clusters c = c / 10 sim = simulateDemocracy metric, {"#{choices[0]}": b, "#{choices[1]}": 1000-b}, p, c, 1000 results.push { "baserate": b, "partitions": p, "clustering": c.toFixed(2), "#{name} #{metric}": ave(sim.trials).toFixed(15) } process.stdout.write "Running #{results.length * 1000} #{name} trials\r" save name, results
Finally, we capture terminal inputs to start up the simulation. To run the simulation of epistemic
democracy, use the command coffee paper.coffee.md epistemic . This will execute the code
embedded described in the paper above.
process.argv.forEach (val, index, array) -> runSimulation('epistemic', 'virtue', ['right', 'wrong']) if val is 'epistemic' runSimulation('preference', 'fidelity', ['red', 'blue']) if val is 'preference' runSimulation('moral', 'character', ['good', 'bad']) if val is 'character'
1. The term justification has been used in a variety of ways within political theory. Sometimes it is
used normatively to describe the content of reasons that might legitimate democratic authority.
Other times it is used descriptively to to describe the giving of such reasons. In this paper I use
justifications of democracy and accounts of democracy synonymously to mean any normative theory of
democracy's value.↩2. Marquis de Condorcet, ‘Essay on the application of mathematics to the theory of decision-
making’, Reprinted in Condorcet: Selected Writings, Keith Michael Baker, ed 33 (1976). ↩3. David M Estlund, Democratic Authority: A Philosophical Framework (Princeton University Press,
2009). ↩4. Douglas W Rae, ‘Decision-rules and individual values in constitutional choice’, American Political
Science Review 63, 01 (1969): 40–56. ↩5. The Social Contract: & Discourses (JM Dent & sons, Limited, 1920).↩6. Considerations on Representative Govrnment, 1862.↩7. Of course not all accounts of democracy are instrumental. Sometimes democracy is justified
intrinsically by appealing to substantive ideals like equality or justice. Unless explicitly stated,
democracy in this paper refers exclusively those instrumental conceptions where outcomes do the
normative of explaining democracy's value.↩8. Timothy Feddersen and Wolfgang Pesendorfer, ‘Elections, information aggregation, and strategic
voting’, Proceedings of the National Academy of Sciences 96, 19 (1999): 10572–4. ↩9. The Boundary Problem of Democratic Theory has gone by a number of names in political theory.
Robert Alan Dahl, , Democracy and Its Critics (Yale University Press, 1989), p. 193 has called it "the
problem of the unit" while R.E. Goodin, , ‘Enfranchising all a!ected interests, and its
alternatives’, Philosophy & public affairs 35, 1 (2007): 40–68, at p. 42 refers to it as the problem of
"constituting the demos". Most other theorists have settled on "the Boundary Problem".↩10. See F.G. Whelan, , ‘Prologue: Democratic theory and the boundary problem’, Liberal democracy 25
(1983) for the seminal analysis of the challenges that the Boundary Problem presents as well as
more recent work by Dahl, op. cit, Gustaf Arrhenius, , ‘The boundary problem in democratic
theory’, Democracy Unbound: Basic Explorations I, 2005: 14–28, Lars Bergström, , ‘Democracy and
political boundaries’, The Viability and desirability of global democracy, Stockholm Studies in Democratic
Theory 3 (2007): 14–32, Goodin, op. cit, David Miller, , ‘Democracy’s domain’, Philosophy & public
affairs 37, 3 (2009): 201–28, Hans Agné, , ‘Why democracy must be global: Self-founding and
democratic intervention’, International Theory 2, 03 (2010): 381–409, Arash Abizadeh, , ‘On the
demos and its kin: Nationalism, democracy, and the boundary problem’, American Political Science
Review 106, 04 (2012): 867–82, Johan Karlsson Scha!er, , ‘The boundaries of transnational
democracy: Alternatives to the all-a!ected principle’, Review of International Studies 38, 2 (2012):
321–42, Sarah Song, , ‘The boundary problem in democratic theory: Why the demos should be
bounded by the state’, International Theory 4, 1 (2012): 39–68, and Eva Erman, , ‘The boundary
problem and the ideal of democracy’, Constellations 21, 4 (2014): 535–46
11. Arrhenius op. cit. p1. ↩12. Whelan op. cit. p13. ↩13. Robert Alan Dahl, After the Revolution?: Authority in a Good Society (Yale University Press, 1970)
<http://books.google.com.au/books?id=5ENQAQAAIAAJ> pp59-60. ↩14. See Goodin, op. cit, Agné, op. cit, Abizadeh, op. cit, Ben Saunders, , ‘The democratic turnout
“problem”’, Political Studies 60, 2 (2012): 306–20, and Erman, op. cit for arguments from the
cit, and Paulina Ochoa Espejo, , ‘People, territory, and legitimacy in democratic states’, American
Journal of Political Science 58, 2 (2014): 466–78 for work concerned with how the Boundary Problem
a!ects democratic legitimacy and territorial states.↩16. Literate programming is a paradigm of computer programming that explains the logic of a
system in essay like form, interspersed with snippets of source code that a compiler executes. The
aim is to prioritise human understanding of the program logic over compiler e#ciency. See
Donald Ervin Knuth, , ‘Literate programming’, The Computer Journal 27, 2 (1984): 97–111 for an
introduction to the topic.↩17. As Goodin, op. cit. , p.55 observes, "Virtually (maybe literally) everyone in the world — and indeed
everyone in all possible future worlds — should be entitled to vote on any proposal or any
proposal for proposals".↩18. The possible number of di!erent partitions is the sum of binomial coe#cients of agents and the
number groups they are partitioned into. This increases exponentially as the number of agents
and groups increases, making a simulating all possible partitions within a reasonable time frame
is beyond the capacity of current desktop computing.↩19. Monte Carlo simulations are class of computational algorithms that rely on statistical sampling
from repeated simulation trials to generate numerical results. see George S Fishman, Monte Carlo:
Concepts, Algorithms, and Applications (Springer, 1996) for a full treatment of their use in computer
simulation.↩20. See Christin List and Robert E Goodin, ‘Epistemic democracy: Generalizing the condorcet jury
theorem’, Journal of Political Philosophy 9, 3 (2001): 277–306. ↩21. op. cit.↩22. A general principle about polity likeness is observable. For any given distribution of agents across
a space, the more internally homogeneous a polity is, them more externally heterogeneous it
must be, and vice versa. This is most pronounced when agent diversity within the space is high,
i.e. when epistemic or preference base rates are close to 0.5, but the relationship deceases as the
diversity in the decreases, i.e. when the base rates approach 0.0 or 1.0.↩23. Condorcet's Jury Theorem posits that the likelihood of majority rule selecting the correct
outcome increases as the number of voters increases (assuming voters have an independent
better than even chance of voting correctly). We should expect to see a gradual decrease in
epistemic virtue as the same number of voters in the space are partitioned into increasing
numbers of polities.↩24. Rae op. cit. ↩25. op. cit.↩26. ‘A set of independent necessary and su#cient conditions for simple majority decision’,
Econometrica: Journal of the Econometric Society, 1952: 680–4.↩27. op. cit. , sec.1.8.3.↩28. op. cit. Ch 3.↩29. Content indi!erent justifications are not typically used in isolation. Both Mill and Rousseau used
these content indi!erent justifications in conjunction with other instrumental justifications of
democracy such as strategic value, truth divination, and preference articulation.↩