Abstract
Scoring elimination rules (SER), that give points to candidates according to their rank in voters’ preference orders and eliminate the candidate(s) with the lowest number of points, constitute an important class of voting rules. This class of rules, that includes some famous voting methods such as Plurality Runoff or Coombs Rule, suffers from a severe pathology known as monotonicity paradox or monotonicity failure, that is, getting more points from voters can make a candidate a loser and getting fewer points can make a candidate a winner. In this paper, we study three-candidate elections and we identify, under various conditions, which SER minimizes the probability that a monotonicity paradox occurs. We also analyze some strategic aspects of these monotonicity failures. The probability model on which our results are based is the impartial anonymous culture condition, often used in this kind of study.
Similar content being viewed by others
Notes
See also Miller (2016) for a (more or less) real-world example.
The IC assumption considers that each voter chooses independently and with the same probability (1/6) one of the six possible rankings on the three candidates. By contrast, the IAC assumption introduces some degree of dependence in voters’ preferences (see Gehrlein 2006, for more on these two assumptions).
Felsenthal and Tideman (2014) and Miller (2016) refer to this kind of situation (\(\hbox {MLP+LMP}\)) as “double monotonicity failure”. Felsenthal and Tideman (2014) have shown that “all prominent voting methods that are vulnerable to monotonicity failure can also display double monotonicity failure” when n is large enough.
A noticeable exception is double monotonicity paradox, for which the vulnerability of NPER is lower than the vulnerability of both PER and BER.
Notice that the results from Table 1 suggest that the optimal value of \(\lambda \) lies strictly in the interior of the unit interval for MLP, LMP and GMP. Such a conclusion does not hold for \(\hbox {MLP+LMP}\): we will see in the next section that the probability of \(\hbox {MLP+LMP}\) is minimized when \(\lambda =1\).
Here, the 5-volume of E denotes the relative volume of E, i.e. is the volume of E relative to its affine span.
Volumes are computed here with respect to the Lebesgue measure on \(\mathbb {R}^{5}\). Note that, since all probabilities calculated in this paper are the quotient of two volumes, we obtain the same results by using any measure that is a multiple of the standard Lebesgue measure.
Felsenthal and Tideman (2013) call dynamic voters those voters who change their preference in the voting situation under consideration.
The Condorcet Efficiency of a voting rule F is defined as the probability that F elects the Condorcet Winner, given that such a candidate exists. See e.g. Gehrlein (1982).
This is one of the closeness measures used by Miller (2016).
It is important to point out that the frequency distributions of the various values of parameter \(\upalpha \) are not similar for the three voting rules under consideration. Consequently, a direct comparison of the results obtained for PER, NPER and BER could be misleading and should be conducted with caution. The reader is referred to the working paper version of our article for more on this issue: https://goo.gl/KXFRGW.
Recall that the Condorcet Efficiency of a voting rule is defined as the conditional probability that the rule elects the Condorcet winner, given that a Condorcet winner exists.
References
Barvinok A (1994) Polynomial time algorithm for counting integral points in polyhedral when the dimension is fixed. Math Oper Res 19:769–779
Brams SJ, Fishburn PC (1983) Paradoxes of preferential voting. Math Mag 56:207–214
Doron G, Kronick R (1977) Single transferable vote: an example of perverse social choice function. Am J Polit Sci 21:303–311
Cohen J, Hickey T (1979) Two algorithms for determining volumes of convex polyhedra. J ACM 26:401–414
Felsenthal DS, Tideman TN (2013) Varieties of failure of monotonicity and participation under five voting methods. Theory Decis 75:59–77
Felsenthal DS, Tideman TN (2014) Interacting double monotonicity failure with direction of impact under five voting methods. Math Soc Sci 67:57–66
Fishburn PC (1982) Monotonicity paradoxes in the theory of elections. Discret Appl Math 4:119–134
Gehrlein WV (1982) Condorcet efficiency and constant scoring rules. Math Soc Sci 2:123–130
Gehrlein WV (2006) Condorcet’s Paradox. Springer, New york
Gehrlein WV, Lepelley D, Moyouwou I (2015) Voters’ preference diversity, concepts of agreement and Condorcet’s paradox. Quality Quant 49(6):2345–2368
Gehrlein WV, Plassmann F (2014) A comparison of theoretical and empirical evaluation of the Borda Compromise. Soc Choice Welf 43:747–772
Lepelley D, Chantreuil F, Berg S (1996) The likelihood of monotonicity paradoxes in run-off elections. Math Soc Sci 31:133–146
Lepelley D, Louichi A, Smaoui H (2008) On Ehrhart’s polynomials and probability calculations in voting theory. Soc Choice Welf 30:363–383
Lepelley D, Moyouwou I, Smaoui H (2016) Borda elimination rule and monotonicity paradoxes in three-candidate elections. Econ Bull 36:1722–1728
Miller NR (2016) Monotonicity failure in IRV elections with three candidates: closeness matters. Working paper downloadable at http://userpages.umbc.edu/~nmiller/
Plassmann F, Tideman N (2014) How frequently do different voting rules encounter voting paradoxes in three-candidate elections? Soc Choice Welf 42:31–75
Smith HJ (1973) Aggregation of preferences with variable electorates. Econometrica 41:1027–1041
Wilson MC, Pritchard G (2007) Probability calculations under the IAC hypothesis. Math Soc Sci 54:244–256
Author information
Authors and Affiliations
Corresponding author
Additional information
We thank our referees for their very useful comments. A version of this paper including some further developments and details is available on the CEMOI website at: https://goo.gl/KXFRGW.
Appendices
Appendix 1: Analytical representations for the vulnerability of scoring elimination rules (\(\mathbf {F}_{\mathbf {\lambda }})\) to monotonicity paradoxes with 3 candidates and large electorate
R1. For \(\lambda \in \left[ 0,\frac{1}{2} \right] \),
For \(\lambda \in \left[ \frac{1}{2},1 \right] \) ,
R2. For \(\lambda \in \left[ 0,\frac{1}{2} \right] \),
For \(\lambda \in \left[ \frac{1}{2},1 \right] \),
R3. For \(\lambda \in \left[ 0,\frac{1}{2} \right] \),
For \(\lambda \in \left[ \frac{1}{2},1 \right] ,\)
R4. For \(\lambda \in \left[ {0,}\frac{{1}}{{2}} \right] \),
For \(\lambda \in \left[ \frac{1}{2},1 \right] \),
R5. For \(0\le \lambda \le \frac{1}{2},\)
For \(\frac{{1}}{{2}}{\le \lambda \le 1,}\)
R6. For \(0\le \lambda \le \frac{1}{2}\),
For \(\frac{{1}}{{2}}{\le \lambda \le 1,}\)
R7. For \(0\le \lambda \le \frac{1}{2},\)
For \(\frac{{1}}{{2}}{\le \lambda \le 1,}\)
R8. For \(0\le \lambda \le \frac{1}{2},\)
For \(\frac{{1}}{{2}}{\le \lambda \le 1,}\)
R9. \(\hbox { For } 0\le \lambda \le \frac{1}{2}\),
For \(\frac{{1}}{{2}}{\le \lambda \le }\frac{{2}}{{3}},\)
For \(\frac{{2}}{{3}}{\le \lambda \le 1,}\)
R10. For \(\le \lambda \le \frac{1}{2},\)
For \(\frac{{1}}{{2}}{\le \lambda \le }\frac{{2}}{{3}},\)
For \(\frac{{2}}{{3}}{\le \lambda \le 1,}\)
R11 For \(0\le \lambda \le \frac{1}{2},\)
For \(\frac{{1}}{{2}}{\le \lambda \le }\frac{{2}}{{3}},\)
For \(\frac{\mathrm {2}}{\mathrm {3}}\mathrm {\le \lambda \le 1,}\)
Appendix 2: Proof of Proposition 4.2
Let x be a voting situation at which a wins against b in the runoff and loses against c after being moved up by some voters in their rankings. Then by Proposition 4.1, MLP occurs under \(F_{\lambda }\) in favor of voters changing their preferences if and only if there exists \(t\in \left[ 0,x_{4} \right] \) such that:
with \(z=(x_{1}+t,x_{2},x_{3},x_{4}-t,x_{5}+x_{6},0)\). Note that (MLP7) is equivalent to (MLP1), (MLP2) and (MLP3). Then we only have to prove that given (MLP1), (MLP2) and (MLP3), (MLP8) holds if and only if (MLP4) and (MLP5) hold for \(\lambda \in \left[ 0,\frac{1}{2} \right] \); and that (MLP4) and (MLP6) hold for \(\in \left[ \frac{1}{2},1 \right] \). Clearly, \(S_\lambda \left( z,a \right) >S_\lambda \left( z,c \right) \) is a consequence of \(S_\lambda \left( x,a \right) >S_\lambda \left( x,c \right) \) since some voters move a up from x to z. Therefore given (MLP1), (MLP2) and (MLP3), (MLP8) is now equivalent to \(S_\lambda \left( z,a \right) >S_\lambda \left( z,c \right) \) and cM(z)a. That is
Taking into consideration the sign of the coefficient \(2\lambda -1\) and the fact that \(t\in ]0,x_{4}]\), it appears that t exists if and only if for \(\lambda \in [ 0,\frac{1}{2} [\), max \((\frac{T_{1}}{2\lambda -1},0)<t<\) min \((\frac{T_{2}}{2},x_{4})\) and for \(\lambda \in ] \frac{1}{2},1 ]\), \(<t<\) min \((\frac{T_{1}}{2\lambda -1},\frac{T_{3}}{2},x_{4})\). Now, for \(\lambda \in [ {0,}\frac{{1}}{{2}}[\), there exists t such that max(\(\frac{T_{1}}{2\lambda -1},0)<t\mathrm {<}\) min \((\frac{T_{2}}{2},x_{4})\) if and only if:
Since (MLP9) and (MLP10) are respectively equivalent to (MLP4) and (MLP5), to complete the proof for \(\lambda \in [ 0,\frac{1}{2} [\), we have to prove that (MLP11) and (MLP12) can be discarded. We omit (MLP12) as it has no influence on the 5-dimensional volume computed with \(0\le x_{\mathrm {4}}\). To see that (MLP12) is redundant, we simply rewrite \(q_{7}\) as a sum of non positive terms:
Similarly, for \(\lambda \in ] \frac{1}{2},1 ]\), there exists t such that \(<t\mathrm {<}min(\frac{T_{1}}{2\lambda -1},\frac{T_{2}}{2},x_{4})\) if and only if:
As mentioned above, (MLP15) has no influence on the evaluation of the 5-dimensional volume computed for \(x_{j}\ge 0\), \(j=1,\ldots , 6\). Finally (ML13) and (MLP14) are respectively equivalent to (MLP4) and (MLP6). For \(\lambda =\frac{\mathrm {1}}{2}\) , (MLP1), (MLP2), (MLP3) and (MLP4) hold from (MLP7) and (MLP8). Moreover, (MLP5) and (MLP6) are now equivalent. \(\square \)
Rights and permissions
About this article
Cite this article
Lepelley, D., Moyouwou, I. & Smaoui, H. Monotonicity paradoxes in three-candidate elections using scoring elimination rules. Soc Choice Welf 50, 1–33 (2018). https://doi.org/10.1007/s00355-017-1069-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-017-1069-1