Dynamics of Strategic Three-Choice Voting

Dynamics of Strategic Three-Choice Voting

D. Volovik Center for Polymer Studies and Department of Physics, Boston University, Boston, Massachusetts 02215 USA    M. Mobilia Department of Applied Mathematics, School of Mathematics, University of Leeds, Leeds LS2 9JT, U.K.111Permanent address from January 2009. Address from January 2009: M.Mobilia@leeds.ac.uk. Mathematics Institute & Warwick Centre for Complexity Science, The University of Warwick, Coventry CV4 7AL, U.K.    S. Redner Center for Polymer Studies and Department of Physics, Boston University, Boston, Massachusetts 02215 USA
July 14, 2019

In certain parliamentary democracies, there are two major parties that move in and out of power every few elections, and a third minority party that essentially never governs. We present a simple model to account for this phenomenon, in which minority party supporters sometimes vote ideologically (for their party) and sometimes strategically (against the party they like the least). The competition between these disparate tendencies reproduces the empirical observation of two parties that frequently exchange majority status and a third party that is almost always in the minority.

89.75.Fb (structures and organization in complex systems), 02.50.-r (probability theory)

I Introduction

A feature of governance in several countries with parliamentary elections—typically British Commonwealth countries and Britain itself—is that two major parties move in and out of power every few elections, while a smaller third party either has never or rarely governed. This lack of representation of the minority party occurs even though its vote fraction can be close to that of the major parties.

Governance is determined by the party (or coalition) that has the majority of members of parliament (MPs). Each MP is the candidate with the most votes in each parliamentary district election. This voting system makes it difficult for a minor party to gain representation that mirrors its vote fraction. For example, if one party (out of three) receives 30% of the vote in every district while the other two parties equally share the remaining 70%, then the minority party will have no MPs although it is supported by nearly 1 in 3 voters. As an illustration british-elections (), in the 1983 British election, the Conservative party won 42.4% of the popular vote and 397 seats (61.1% of 650 seats), the Labor party won 27.6% of the vote and 209 seats (32.2% of 650), while the traditional third-place Liberal party won 25.4% of the vote but only 23 seats (3.56% of 650). Similarly, Canada had two major parties for much of the century—Liberals (center-left), Progressive Conservatives (center-right), and a smaller, but still national scale, New Democratic Party (leftist) that has never governed canadian-elections (); sc ().

Figure 1: (a) British election results since 1830. (b) Canadian election results from 1935–90; here NDP/CC denotes the New Democratic Party and its Cooperative Commonwealth party antecedent. (c) Representative simulation results for the time evolution of the densities of the three states in the strategic voter model with state-dependent strategic bias for agents and parameter (see text for definition of ).

To understand how such a voting pattern can arise, consider the case of the traditional minority Liberal party in Britain. If their election fortunes seem promising, then supporters are likely to be galvanized to vote for their party. However, if Liberal prospects seem bleak and the diametrically opposed Conservative party appears strong, a Liberal may vote for the Labor party to forestall the Conservatives. This strategy clearly disfavors the minority party (Fig. 1). In fact, when the number of parties is larger than two, ambiguous outcomes for the voting preference, such as the Condorcet paradox cond (), can easily arise. There is also a richer range of phenomenology than in two-party voting galam ().

To account for the endemic weakness of a minority party, we introduce the “strategic” voter model that encapsulates the strategic/ideological dichotomy outlined above. The original voter model voter () is a paradigmatic non-equilibrium process that describes ordering in non-equilibrium systems and consensus in opinion dynamics Castellano_review (). In the voter model, agents are endowed with a 2-state opinion variable. The dynamics is defined by picking a voter at random and updating its opinion and repeating ad infinitum. In the update event, the agent adopts the opinion of a random neighbor. Neighbors can be defined on a complete graph (interaction with any other agent equiprobably), across the links of regular lattices VMlattice () or complex networks VMnets (), or on adaptive graphs adaptive (). Various extensions of the dynamics itself have been considered extensions ().

In the next section, we define two natural versions of strategic voting and determine their dynamics in the mean-field limit. Typically, the population is driven to a steady state where a single party is perpetually in the minority unless the strategic bias is extremely weak. In the symmetric-breaking steady state, discrete fluctuations ultimately cause the minority party to escape its minority status on a slow time scale. It appears possible that election results (Fig. 1) would exhibit these same oscillations if the data continued for a much longer time period.

Ii Strategic Voting

We consider a population of voters on a complete graph that have three possible opinions states: , and . The idealized complete graph is used because of its simplicity and analytical tractability. Each of these states is equivalent, in contrast to real political parties, and our model thus does not incorporate any subjective political attributes. We denote the fractions of the three types of voters as , , and . The state of each individual evolves by the following voter-like dynamics:

  • With rate , a voter spontaneously changes to another state equiprobably. This thermalization step ensures that consensus is never reached.

  • With rate , a random pair of voters in states and is picked and one member of this pair adopts the state of the other voter. In general, does not equal (see below).

In the mean-field limit (), the fractions of each voter species evolve according to the rate equations:


The terms proportional to account for spontaneous opinion changes, while the remaining terms account for voter-like updating. In the classic voter model voter (); Redner (), each , because a voter pair can change to or to equiprobably; thus the density of each species is conserved. In our model, we only require to conserve the total density.

For the strategic voter model, we take to be nonzero when one of or denotes a minority state. A simple choice is a strategic bias that is independent of the fractions of each species, viz:

(and cyclic permutations for and ). Thus if is in the minority, then an interaction favors the outcome rather than because and . Conversely, if is in the minority, an interaction favors rather than . If neither or are in the minority, then they undergo conventional voter dynamics in which their average densities do not change. While it would be politically more realistic to have non-symmetric interactions in which the minority species has a definite preference for one of the two non-minority species, this more complicated model does not yield any additional insights about strategic voting.

In the sector of the composition triangle (where is in the minority, see Fig. 2), the rate equations reduce to:


Here we use to simplify the -dependent terms and the last term in the equation for . Similar equations hold for the and sectors by cyclic permutations.

We will also study a more realistic strategic voting in which the strategic bias vanishes as the minority population approaches those of the other two states. In terms of election sentiment, if a minority party supporter believes that his/her party will win in an upcoming election, then there is every reason to vote ideologically with the party and not strategically. For simplicity we consider the case where the strategic bias varies linearly in the depth of minority status as quantified by the density-dependent strategic voting rate in Eqs. (II) (the sector), and analogously for the other sectors of the composition triangle.

Iii Solution to the Rate Equations

iii.1 Constant Strategic Bias

For a strategic bias with a fixed rate , we solve the last of Eqs. (II) by rewriting it in the factorized form


Here quantifies the relative importance of the strategic bias (with strong bias corresponding to small ), and

We now apply a partial fraction expansion to Eq. (3) to render it integrable by elementary means and the solution is


Thus converges to its steady-state value of exponentially in time. The value of sharply goes to zero as because the strategic bias dominates, while gradually approaches the limiting value of when (see Fig. 3(a)).

Figure 2: Composition triangle showing the stable (dots) and unstable (circles) fixed points of the rate equations. The triangle represents the locus of points in the density space. Each sector is demarcated by separatrices, and the local flow near one stable fixed point is shown. The heavy lines outline the sector where the density of is in the minority.

For the majority density, we write the first of Eqs. (II) as , where , with formal solution



Finally, we substitute the above result for into Eq. (5) and perform the integral to obtain


and similarly for . Here , and the steady-state densities are . It is easy to check that , so that the approach to the steady-state is dominated by in Eq. (6), while the minority species relaxes more quickly to stationarity: . The same considerations apply, mutatis mutandis, in the and sectors of the composition triangle.


Figure 3: Simulation results for the fixed point minority concentration for: (a) a population of agents with constant strategic bias as a function of , and (b) agents with state-dependent strategic bias as a function of . The numerical value , which agrees perfectly with the rate equation predictions, is determined by the time integral of the concentration divided by the observation time. The change in behavior when becomes a cusp as .

The non-trivial fixed points (dots in Fig. 2) are the basins of attractions for the , and sectors for any value of the basic parameter , and for almost all initial conditions (except for the special cases where all the densities initially equal or only two densities are nonzero).

iii.2 State-Dependent Strategic Bias

We now consider strategic bias with state-dependent rate in the rate equations (II). To simplify technical details, we assume (corresponding to an initial state ) throughout the evolution; thus the normalization condition now is . The rate equation for the minority density in the sector now is:


where , , and . By plotting the right-hand side of Eq. (7) as a function of , it is clear that the stable fixed point value is for . In this case, any initial state in the sector flows to the fixed point ; this behavior is qualitatively similar to that of constant strategic bias. Conversely, for the stable fixed point is , with and both greater than (Fig. 3, right). Thus a sufficiently weak strategic bias or a sufficiently strong voting uncertainty, as quantified by , leads to the fixed point becoming stable. This change in the stability of the fixed points as a function of the basic parameter is an unexpected feature of state-dependent strategic voting.

To solve for the time dependence, we perform a partial fraction expansion to transform the rate equation (7) to

This equation can be straightforwardly integrated and the result is



The main feature of this result is that the approach to the reactive steady state is still exponential in time, but the decay time can become quite long when the parameter . For example for , the asymptotic decay of the minority density is


Thus as where and approach each other, the approach to the fixed point becomes very slow. On the other hand, for , the densities all decay to their common value of as


Iv Influence of stochastic fluctuations

In a finite population, the stochasticity of the dynamics allows the system to eventually escape the basin of attraction of a fixed point. This escape corresponds to a minority party becoming one of the top two parties on a slow time scale. To understand this escape dynamics, we formulate a discrete version of the strategic voter model for a fixed population , with agents of species , of species , of species , with . This population evolves by the reactions:

for the case where is in the minority. Corresponding rules apply when and are in the minority.


Figure 4: Average time span for a single party to have minority status versus population for strategic voting with: (a) constant strategic bias with and (b) state-dependent strategic bias with ; this latter value is chosen to give the same steady-state minority densities for both models. Both time dependences are consistent with exponential growth in .

The stochastic dynamics of the strategic voter model may be analytically described by a master equation whose finite-size expansion Gardiner (); vanKampen () leads to a Fokker-Planck equation. The fixed points in the rate equation then correspond, for finite , to an effective attractor for the probability distribution. Because of finite-size fluctuations, the system can escape such a potential well and thereby move between different attractors. This barrier crossing corresponds to a change in the identity of the minority party. We can investigate this barrier crossing by rephrasing it as a first-passage problem Gardiner (); vanKampen (); Redner (). Following a standard calculation (as given, for example, in Ref. RMF ()), we find that changes in the identity of the minority party occur on a time scale that grows exponentially in  InPrep (), much longer than the time scale of fluctuations between the majority and second-place parties.

We have tested this prediction by numerical simulations. The time span over which a given state is in the minority as a function of the total population does appear to grow exponentially with for both constant and state-dependent strategic bias, but with a much smaller amplitude for the latter case (Fig. 4). For the more realistic case of a state-dependent strategic bias, simulations of single realizations exhibit changes in minority status on a time scale that can be tuned to be in a similar range as that of election data (Fig. 1).

V Discussion

We proposed a strategic voter model to describe the feature whereby two major parties dominate in certain parliamentary democracies, while a third party remains in the minority. Our model is based on a strategic bias that inherently disfavors the minority species. We studied two variants of strategic voting in which the strategic bias is either fixed or state dependent. For fixed bias, the minority species is doomed to remain eternally in the minority in the rate equation limit. For a state-dependent bias, a bifurcation occurs between a stable minority and equal densities of the three parties as the strength of the strategic bias varies.

It is worth mentioning that our three-choice voting model has an important distinction with predator-prey models that exhibit ”competitive exclusion”. This feature — known as Gause’s Law G () — states that two species that compete for the same resources cannot stably coexist if other ecological factors are constant. In our model, if we consider the opinions as different species, the dynamics allow for spontaneous mutations and additionally the interactions are not constant but state dependent. These features allow for a stable coexistence of all species.

For finite populations, stochastic fluctuations allow the minority party to escape its status on a time scale that grows exponentially with population size. The concomitant slow oscillations in the identity of the minority party can be readily seen in simulations, and this time evolution seems to roughly mirror real election data, especially if the latter could be extended over a much longer time scale. However, to quantitatively match the election data, it would be necessary to have a strategic bias that is still weaker than our state-dependent strategic voter model to reproduce both the density difference between majority and minority and the time scale for a change in minority status.

Acknowledgments: SR thanks the hospitality of the Complexity Science Centre at the University of Warwick, where this work began. We gratefully acknowledge financial support from NSF grant DMR0535503 (DV and SR), and the Swiss National Science Foundation through the Advanced Fellowship PA002-119487 (MM).


  • (1) http://en.wikipedia.org/wiki/UnitedKingdomgeneralelections.
  • (2) http://en.wikipedia.org/wiki/Canadianfederalelection.
  • (3) We ignore Canadian political parties that had only regional strength, such as the Social Credit party.
  • (4) M.-J.-A.-N. Caritat (Marquis de Condorcet) “Essai sur l’application de l’analyse a la probabilité des décisions rendues a la pluralité des voix” (1785); K. J. Arrow, Social Choice and Individual Values (John Wiley & Sons, New York, 1951).
  • (5) S. Gekle, S. Galam, and L. Peliti, Eur. Phys. J. B 45, 569 (2005).
  • (6) T. M. Liggett, Stochastic Interacting Systems: Contact, Voter, and Exclusion Processes, (Springer-Verlag, New York, 1999).
  • (7) C. Castellano, S. Fortunato, and V. Loreto, arXiv:0710.3256v1
  • (8) P. L. Krapivsky, Phys. Rev. A 45, 1067 (1992); L. Frachebourg and P. L. Krapivsky, Phys. Rev. E 53, R3009 (1996);E. Ben-Naim, L. Frachebourg and P. L. Krapivsky, Phys. Rev. E 53, 3078 (1996); I. Dornic, H. Chaté, J. Chave, and H. Hinrichsen, Phys. Rev. Lett. 87, 045701 (2001)
  • (9) V. Sood and S. Redner, Phys. Rev. Lett. 94, 178701 (2005). V. Sood, T. Antal, and S. Redner, Phys. Rev. E 77, 041121 (2008); F. Vazquez and V. M. Eguiluz, New Journal of Physics 10, 063011 (2008).
  • (10) S. Gil and D. H. Zanette, Phys. Lett. A 356, 89 (2006); P. Holme and M. E. J. Newman, Phys. Rev. E 74, 056108 (2006); F. Vazquez, V. M. Eguiluz, and M. San Miguel, Phys. Rev. Lett. 100, 108702 (2008); I. J. Benczik, S. Z. Benczik, B. Schmittmann, and R. K. P. Zia, Europhys. Lett. 82, 48006 (2008).
  • (11) M. Mobilia, Phys. Rev. Lett. 91, 028701 (2003); R. Lambiotte and S. Redner, J. Stat. Mech. L10001 (2007).
  • (12) C. W. Gardiner, Handbook of Stochastic Methods, (Springer, New York, 2002), 2nd ed.
  • (13) N. G. van Kampen, Stochastic Processes in Physics and Chemistry, (North-Holland, Amsterdam, 1997), 2nd ed.
  • (14) T. Reichenbach, M. Mobilia, and E. Frey, Phys. Rev. E 74, 051907 (2006).
  • (15) S. Redner, A Guide to First-Passage Processes, (Cambridge University Press, Cambridge, 2001).
  • (16) D. Volovik, M. Mobilia, and S. Redner, in preparation.
  • (17) G. F. Gause, The Struggle for Existence (Williams & Wilkins, Baltimore, MD, 1934); G. Hardin, Science 131, 1292 (1960).
Comments 0
Request Comment
You are adding the first comment!
How to quickly get a good reply:
  • Give credit where it’s due by listing out the positive aspects of a paper before getting into which changes should be made.
  • Be specific in your critique, and provide supporting evidence with appropriate references to substantiate general statements.
  • Your comment should inspire ideas to flow and help the author improves the paper.

The better we are at sharing our knowledge with each other, the faster we move forward.
The feedback must be of minimum 40 characters and the title a minimum of 5 characters
Add comment
Loading ...
This is a comment super asjknd jkasnjk adsnkj
The feedback must be of minumum 40 characters
The feedback must be of minumum 40 characters

You are asking your first question!
How to quickly get a good answer:
  • Keep your question short and to the point
  • Check for grammar or spelling errors.
  • Phrase it like a question
Test description