Abstract
We consider the calculation of Nitzan–Kelly’s manipulability index in the impartial anonymous and neutral culture (IANC) model. We provide a new theoretical study of this model and an estimation for the maximal difference between manipulability indices in the IANC model and a basic model, the impartial culture (IC). The asymptotic behavior of this difference is studied with the help of the impartial anonymous culture (IAC) model. It is shown that the difference between the IAC and IANC models tends to zero as the number of alternatives or the number of voters grows. These results hold for any other probabilistic measure that is anonymous and neutral. Finally, we calculate Nitzan–Kelly’s index in the IANC model for four social choice rules and compare it with the IC model.
Similar content being viewed by others
References
Aleskerov F, Karabekyan D, Sanver R, Yakuba V (2011) On the degree of manipulability of multi-valued social choice rules. Homo Oecon 28(1/2):205–216 (Essays in Honor of Hannu Nurmi)
Aleskerov F, Kurbanov E (1999) Degree of manipulability of social choice procedures. In: Alkan A, Aliprantis CD, Yannelis NC (eds) Current trends in economics. Springer, Berlin
Barbera S (1977) The manipulation of social choice mechanisms that do not leave too much to chance. Econometrica 45:1572–1588
Berg S (1985) Paradox of voting under an urn model: the effect of homogeneity. Public Choice 47(2):377–387
Duggan J, Schwartz T (2000) Strategic manipulability without resoluteness or shared beliefs: Gibbard–Satterthwaite generalized. Soc Choice Welf 17(1):85–93
Eğecioğlu O, Giritligil AE (2011) The likelihood of choosing the Borda–Winner with partial preference rankings of the electorate. J Mod Appl Stat Methods 10(1):349–361
Eğecioğlu O, Giritligil AE (2013) The impartial, anonymous and neutral culture model: a probability model for sampling public preference structures. J Math Sociol 37(4):203–222
Favardin P, Lepelley D (2006) Some further results on the manipulability of social choice rules. Soc Choice Welf 26(3):485–509
Feller W (1957) An introduction to probability theory and its applications, 3rd edn. Wiley, New York
Gärdenfors P (1976) Manipulation of social choice functions. J Econ Theory 13:217–228
Gehrlein WV, Fishburn PC (1976) Condorcet’s paradox and anonymous preference profiles. Public Choice 26:1–18
Gehrlein WV, Lepelley D (2011) Voting paradoxes and group coherence. Springer, Berlin
Gibbard A (1973) Manipulation of voting schemes. Econometrica 41(4):587–601
Guilbaud GT (1952) Les théories de l’intéret general et le problèmelogique de l’agrégation. Econ Appl 5:501–584
Kelly J (1977) Strategy-proofness and social choice functions without single valuedness. Econometrica 45(2):172–175
Kelly J (1988) Minimal manipulability and local strategy-proofness. Soc Choice Welf 5(1):81–85
Kelly J (1993) Almost all social choice rules are highly manipulable, but few aren’t. Soc Choice Welf 10(2):161–175
Kuga K, Nagatani H (1974) Voter antagonism and the paradox of voting. Econometrica 42(6):1045–1067
Lepelley D, Valognes F (2003) Voting rules, manipulability and social homogeneity. Public Choice 116(1/2):165–184
Nitzan S (1985) The vulnerability of point-voting schemes to preference variation and strategic manipulation. Public Choice 47:349–370
Pattanaik P (1978) Strategy and group choice. North-Holland, Amsterdam
Peters H, Roy S, Storcken T (2012) On the manipulability of approval voting and related scoring rules. Soc Choice Welf 39(2–3):399–429
Pritchard G, Wilson M (2007) Exact results on manipulability of positional voting rules. Soc Choice Welf 29(3):487–513
Rodríguez-Álvarez C (2009) On strategy-proof social choice correspondences: a comment. Soc Choice Welf 32(1):29–35
Sato S (2008) On strategy-proof social choice correspondences. Soc Choice Welf 31(2):331–343
Satterthwaite M (1975) Strategy-proofness and Arrow’s conditions: existence and correspondence theorems for voting procedures and social welfare functions. J Econ Theory 10(2):187–217
Slinko A (2006) How the size of a coalition affects its chances to influence an election. Soc Choice Welf 26(1):143–153
Veselova YA (2014) The manipulability index in the IANC model. Clusters, orders, and trees: methods and applications. Springer, New York, pp 391–404
Acknowledgments
The author is grateful to F. Aleskerov for his scientific advice and encouragement, and also would like to thank D. Piontkovsky, D. Shvarts, A. Giritligil and participants of the Computational Social Choice session on the 11th Meeting of Society for Social Choice and Welfare (Indian Statistical Institute, New Delhi, India, August 17–20, 2012) for their helpful comments. The author would like to express her deep gratitude to two anonymous referees for all their comments and recommendations. They were extremely helpful and allowed to improve the presentation of results significantly. The author acknowledges the support the International Laboratory of Decision Choice and Analysis. The financial support from the Government of the Russian Federation within the framework of the Basic Research Program at the National Research University Higher School of Economics and within the framework the implementation of the 5–100 Programme Roadmap of the National Research University Higher School of Economics is acknowledged.
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1
Theorem 1
(ANECs with the minimal and maximal number of elements)
-
(a)
The minimal number of elements in an anonymous and neutral equivalence class is m!. This class is unique for the case of \(n\ge 3\).
-
(b)
The maximal number of elements in an anonymous and neutral equivalence class is m!n! if and only if \(m!>n\).
Proof
-
(a)
If profile \(\vec {P}\) consists of equal columns, then \(G_{\vec {P}} =\{(\tau _0 ,\sigma ):\sigma \in S_n\}\) and \(|G_{\vec {P}} |=n!\). Thus, \(|\theta _{\min } |\le m!n!/n!=m!\). For any \(\sigma \in S_n\) and \(\tau _1 \ne \tau _2\), it is impossible that both \((\tau _1,\sigma )\in G_{\vec {P}}\) and \((\tau _2 ,\sigma )\in G_{\vec {P}}\) because \((\vec {P}^{\sigma })^{\tau _1 }\ne (\vec {P}^{\sigma })^{\tau _2}\). Then, for any \(\vec {P}\), \(|G_{\vec {P}} |\le n!\). Hence, \(|\theta _{\min } |\ge m!n!/n!=m!\). Therefore, \(|\theta _{\min } |=m!\).
Let us consider a preference profile consisting of two columns, \(P_1 =(a_1 ,a_2 ,\ldots ,a_m )\) and \(P_2 =(a_2 ,a_1 ,a_3 \ldots ,a_m)\), where \(\tau \) is of order 2. Take \(\tau =(12)(3)\ldots (m)\) and \(\sigma =(12)\); then, \(|G_{\vec {P}} |=2=n!\). Thus, the number of ANECs with a minimal number of elements is greater than 1 when \(n=2\).
Now consider a profile \(\vec {P}\) with \(n\ge 3\) and at least two distinct columns, \(P_i\) and \(P_j\). Let \(\sigma \in S_n\) permute columns i and j, and fix column l, \(l\ne i\), \(l\ne j\). However, there is no permutation \(\tau \) s.t. \((\tau ,\sigma )\in G_{\vec {P}}\). Thus, \(|G_{\vec {P}} |<n!\).
We conclude that \(|G_{\vec {P}} |=n!\) with \(n\ge 3\) if and only if \(P_1 =P_2 =\cdots =P_n\). Since all such profiles belong to the same ANEC, it is unique.
-
(b)
First, we show that \(|\theta _{\max } |=m!n!\) implies \(m!>n\). Suppose, that \(|\theta _{\max } |=m!n!\), but \(m!\le n\). If \(m!<n\) for some \(\vec {P}\), then there exist i, j such that \(P_i =P_j\). Let \(\sigma _1\) permute only columns i and j. Then, \(g_0 =(\tau _0 ,\sigma _0 )\in G_{\vec {P}}\) and \(g_1 =(\tau _0 ,\sigma _1 )\in G_{\vec {P}}\), \(|G_{\vec {P}} |>1\) and \(|\theta _{\max } |<m!n!\). On the other hand, if \(m!=n\), consider the preference profile that consists of all m! different columns. Consider any permutation of order m, for example, (\(1 2 {\ldots } m\)), which splits up a set of m! columns into \(m!-1\) non-intersecting orbits. Again, \(|G_{\vec {P}} |>1\). Consequently, if \(|\theta _{\max } |=m!n!\) then \(m!>n\). Second, we prove that if \(m!>n\), then \(|\theta _{\max } |=m!n!\). Let L be the set of all linear orders. Suppose that there exist such m and n such that for each \(\vec {P}\) consisting of distinct columns, \(|G_{\vec {P}} |\ge 2\). Take any such \(\vec {P}\) and construct a set of preference profiles replacing its n-th column, \(P_n\), by any \({P}'\in L\backslash \{P_1 ,\ldots ,P_{n-1} \}\). We get
$$\begin{aligned} M=\{{\vec {P}}'=(P_1 ,\ldots ,P_{n-1} ,{P}'):{P}'\in L\backslash \{P_1 ,\ldots ,P_{n-1} \}\}. \end{aligned}$$The cardinality of this set is \(m!-n+1\). According to our assumption, for each \(\vec {P}\) in M, \(|G_{\vec {P}} |\ge 2\). This means that there exist \((\tau ,\sigma )\ne (\tau _0 ,\sigma _0)\), and \(\sigma \) does not have cycles of length 1. Then, for every \({P}'=(a_1 ,\ldots ,a_m)\), \((a_{\tau ^{i}(1)} ,\ldots ,a_{\tau ^{i}(m)} )\in \{P_1 ,\ldots ,P_{n-1} \}\), where \(i=1..t\) and t is the order of \(\tau \); however, this is impossible. Consequently, for any m and n, there exists a preference profile with only one stabilizing permutation, \((\tau _0 ,\sigma _0)\). The cardinality of the equivalence class \(\theta _{\max }\) consisting of such profiles is
$$\begin{aligned} | {\theta _{\max } }|=\frac{m!n!}{1}=m!n!. \end{aligned}$$
\(\square \)
Lemma 1
The number of fixed-points from \(\tilde{\Omega }_{m,n}\) for a permutation \(g=(\tau ,\sigma )\) is equal to
where \(d=LCM(\mu )\).
Proof
For some \(\vec {P}\in \tilde{\Omega }_{m,n}\) and some \(g\in G\), suppose that \(\vec {P}^{g}=\vec {P}\). Let us consider the permutation of alternatives \(\tau \). The order of this permutation, d, is equal to the least common multiple of cycle lengths, \(d=LCM(\mu )\). If a column is \((a_1 ,a_2 ,\ldots ,a_m)\), then after the action of permutation \(\tau \), it becomes \((a_{\tau (1)} ,a_{\tau (2)} ,\ldots ,a_{\tau (m)})\); after the second action of \(\tau \), it becomes \((a_{\tau ^{2}(1)} ,a_{\tau ^{2}(2)} ,\ldots ,a_{\tau ^{2}(m)})\); and so on. Finally, \((a_{\tau ^{d-1}(1)} ,a_{\tau ^{d-1}(2)} ,\ldots ,a_{\tau ^{d-1}(m)})\) again becomes \((a_1 ,a_2 ,\ldots ,a_m)\).
Next, for a given permutation \(\sigma \) of voters, we take any cycle \((i_1 \quad i_2 \ldots \quad i_k)\). Let the \(i_1 \)-th column be \((a_1 ,a_2 ,\ldots ,a_m)\). Then, after the action of permutation \(\sigma \), it becomes the column number \(i_2\). At the same time, \(\tau \) permutes alternatives and the column number \(i_2\) should be equal to \((a_{\tau (1)} ,a_{\tau (2)} ,\ldots ,a_{\tau (m)})\) because the preference profile after permuting both voters and alternatives should be the same. Thus, if the \(i_2\)-th column \((a_{\tau (1)} ,a_{\tau (2)} ,\ldots ,a_{\tau (m)})\) after permuting columns becomes the \(i_3\)-th column and \((a_{\tau (1)} ,a_{\tau (2)} ,\ldots ,a_{\tau (m)})\) is mapped to \((a_{\tau ^{2}(1)} ,a_{\tau ^{2}(2)} ,\ldots ,a_{\tau ^{2}(m)})\) by \(\tau \), then the \(i_3\)-th column should be \((a_{\tau ^{2}(1)} ,a_{\tau ^{2}(2)} ,\ldots ,a_{\tau ^{2}(m)})\), etc. Finally, the \(i_k\)-th column becomes the \(i_1\)-th column, and we conclude that it should be equal to \((a_{\tau ^{d-1}(1)} ,a_{\tau ^{d-1}(2)} ,\ldots ,a_{\tau ^{d-1}(m)})\) in order to be mapped to \((a_1 ,a_2 ,\ldots ,a_m)\) by permutation \(\tau \). Since repeated columns are not permitted, the length of the cycle \((i_1 \quad i_2 \ldots \quad i_k)\), k, should be equal to \(d=LCM(\mu )\). The same result holds for any other cycle of permutation \(\sigma \); in other words, the length of all cycles in \(\sigma \) is the same,\(\lambda _1 =\lambda _2 =\ldots =\lambda _\alpha \), or \(GCD(\lambda )=LCM(\lambda )\), and equals d.
Suppose a permutation \(g=(\tau ,\sigma )\) is such that \(\lambda _1 =\lambda _2 =\cdots =\lambda _\alpha =d\) and \(\sigma \) consists of \(\alpha =n/d\) cycles. The first column of the first cycle in \(\sigma \) is chosen from m! different columns. The rest of the columns in this cycle are determined by permutation \(\tau \). Then, the first column of the second cycle in \(\sigma \) can be represented by any of \(m!-d\) variants, as d columns are already used in the first cycle. The first column of the third cycle can be defined \(m!-2d\) different ways and so on. Inside any cycle, there cannot appear any column that is already used because a permutation \(\tau \) splits up the set of all different columns into non-intersecting orbits of columns that can be produced one from another by this permutation. Columns in every cycle form one of such orbits.
Finally, we obtain an exact formula for the number of fixed-points from \(\tilde{\Omega }_{m,n}\) for some permutation \(g=(\tau ,\sigma )\)
\(\square \)
Theorem 2
For any m and n such that \(m!>n\), the number of equivalence classes on \(\tilde{\Omega }_{m,n}\) is equal to
where \(S(\mu ,\lambda )\) is the statement “\(\lambda _1 =\lambda _2 =\cdots =\lambda _\alpha =d\)”.
Proof
Let \(\tilde{R}\) be the number of ANECs on \(\tilde{\Omega }_{m,n}\), and \(\vec {P}^{1},\vec {P}^{2},\ldots ,\vec {P}^{\tilde{R}}\) be the representatives of these classes. For any preference profile \(\vec {P} \quad | {\theta _{\vec {P}} } |\cdot | {G_{\vec {P}} } |=| G |\) holds. Then, take the sum of these equalities over representatives
Since \(| {G_{\vec {P}^{k}} } |=| {G_{\vec {P}^{l}} }|\) if \(\vec {P}^{k}\) and \(\vec {P}^{l}\) belong to the same ANEC, we can rewrite
The sum of stabilizing permutations over all preference profiles from \(\tilde{\Omega }_{m,n}\) is equal to the sum of all fixed-points from \(\tilde{\Omega }_{m,n}\) for all permutations
Using Lemma 1 and denoting the statement “\(\lambda _1 =\lambda _2 =\cdots =\lambda _\alpha =d\)” by \(S(\mu ,\lambda )\) we get
Since \(| {\tilde{F}_g }|\) depends only on the cycle type of permutation \(g=(\tau ,\sigma )\), we can take the sum over all partitions \(\lambda \) and \(\mu \) and multiply every component by the number of permutations of n with partition \(\lambda \) and the number of permutations of m with partition \(\mu , z_\lambda ^{-1} n!\) and \(z_\mu ^{-1} m!\), respectively.
\(\square \)
Corollary 1
For any m and n such that \(m!>n\):
-
(a)
The number of maximal ANEC satisfies the following inequality
$$\begin{aligned} \frac{2(m!-1)!}{(m!-n)!n!}-\tilde{R}(m,n)\le R_{\max } (m,n)\le \tilde{R}(m,n). \end{aligned}$$ -
(b)
If for m and n we have \(n>m\) and n is a prime number, then the number of maximal ANECs is equal to \(\tilde{R}(m,n)\).
Proof
-
(a)
The second inequality is obvious, because preference profiles from maximal equivalence class always consist of different columns for \(m!>n\). The total number of preference profiles consisting of pairwise different columns is
$$\begin{aligned} m!\cdot (m!-1)\cdot \cdots \cdot (m!+1-n)=\frac{(m!)!}{(m!-n)!}. \end{aligned}$$At the same time, it is the number of fixed points from \(\tilde{\Omega }_{m,n}\) for the identity permutation. Since for the identity permutation the statement \(S(\lambda , \mu )\) is true, \((m!)!/(m!-n)!\) is included in the sum \(m!n!\tilde{R}(m,n)\). The rest of the components of the sum, i.e.
$$\begin{aligned} m!n!\sum _\lambda {\sum _\mu {z_\lambda ^{-1} z_\mu ^{-1} \chi (S(\lambda ,\mu ))\mathop {\prod }\limits _{j=0}^{\alpha -1} {(m!-j\cdot d)} } } -\frac{(m!)!}{(m!-n)!} \end{aligned}$$form the sum of fixed points from \(\tilde{\Omega }_{m,n}\) for all permutations except the identity permutation. The problem is that the sets \(\tilde{F}_g\) intersect and we cannot find an exact number of preference profiles that have more than one stabilizing permutation. However, we can be sure that this number is not more than this sum. Consequently, the number of profiles having only one stabilizing permutation is not less than
$$\begin{aligned}&\frac{(m!)!}{(m!-n)!}-\left( {m!n!\sum _\lambda {\sum _\mu {z_\lambda ^{-1} z_\mu ^{-1} \chi (S(\lambda ,\mu ))\mathop {\prod }\limits _{j=0}^{\alpha -1} {(m!-j\cdot d)} } } -\frac{(m!)!}{(m!-n)!}} \right) \\&\quad = \frac{2(m!)!}{(m!-n)!}-m!n!\sum _\lambda {\sum _\mu {z_\lambda ^{-1} z_\mu ^{-1} \chi (S(\lambda ,\mu ))\mathop {\prod }\limits _{j=0}^{\alpha -1} {(m!-j\cdot t)} } } . \end{aligned}$$Dividing by the cardinality m!n! of the maximal ANEC, we get
$$\begin{aligned} R_{\max } (m,n)\ge \frac{2(m!-1)!}{(m!-n)!n!}-\tilde{R}(m,n). \end{aligned}$$ -
(b)
If n is a prime number, then \(\sigma \) either contains only one cycle of length n, or n cycles of length one. The first case means that the least common divisor of the \(\tau \) cycle lengths should also be n, but we assumed \(n>m\). Thus, this is impossible. The second case means that we only have an identity permutation in a stabilizer group of any profile from \(\tilde{\Omega }_{m,n}\), i.e. \(\tilde{\Omega }_{m,n}\) consists of preference profiles from maximal equivalence classes, and we have only one component in the sum \(\tilde{R}(m,n)\)
$$\begin{aligned} \tilde{R}(m,n)=R_{\max } (m,n)=\frac{(m!-1)!}{(m!-n)!n!}. \end{aligned}$$
\(\square \)
Lemma 2
For differences between any probabilistic models \(M_1, M_2\), \(M_3\), the triangular inequality holds
Proof
Let \(E_1\), \(E_2\) and \(E_3\) denote events that maximize the difference of probabilities between \(M_1\) and \(M_2\), \(M_2\) and \(M_3\), \(M_1\) and \(M_3\) respectively.
Since the absolute difference s maximal both for \(E_1\) and the complementary event \(E_1 ^{c}\), without loss of generality, suppose that all differences are positive for \(E_1\), \(E_2\) and \(E_3\). Then,
Therefore,
\(\square \)
Theorem 3
The share of maximal N-AECs and the number of elements in them tends to 1 as \(n\rightarrow \infty \) or \(m\rightarrow \infty \).
-
(1)
\(\mathop {\lim }\nolimits _{n\rightarrow \infty }\frac{m!R_r (m,n)}{K(m,n)}=1\) and \(\mathop {\lim }\nolimits _{m\rightarrow \infty }\frac{m!R_r (m,n)}{K(m,n)}=1\);
-
(2)
\(\mathop {\lim }\nolimits _{n\rightarrow \infty }\frac{R_r (m,n)}{R(m,n)}=1\) and \(\mathop {\lim }\nolimits _{m\rightarrow \infty }\frac{R_r (m,n)}{R(m,n)}=1\).
Proof
-
(1)
First prove that
Let us denote the sum in the denominator by \(\mathop {\sum }\nolimits _{\mu \in \Pi _m \backslash \{(1,1,\ldots ,1)\}} {A_\mu }\), where
Then
Suppose, that n / d is integer, and the binomial coefficient is non-zero
-
(2)
Fix the value of m and let \(n\rightarrow \infty \)
Now fix n and let \(m\rightarrow \infty \)
Consequently,
-
(3)
Now sum the terms \(m!A_\mu /K(m,n)\) over all partitions \(\mu \) from \(\Pi _m \backslash \{(1,1,\ldots ,1)\}\) and s.t. n / d is integer. Let us denote this set of partitions by \({\Pi }^{\prime }_{m}\). The number of elements in \({\Pi }^{\prime }_{m}\) is less or equal than in \(\Pi _m \backslash \{(1,1,\ldots ,1)\}\). Further, \(| {\Pi _m \backslash \{(1,1,\ldots ,1)\}} |<(m-1)!\) for \(m\ge 4\) and for \(m=3 | {\Pi _m \backslash \{(1,1,\ldots ,1)\}} |=(m-1)!=2\).
Therefore,
Finally,
Since
-
(4)
In this part the proof is done for \(n\rightarrow \infty \), for \(m\rightarrow \infty \) the proof is analogous.
Finally,
\(\square \)
Corollary 2
The difference between the IAC and IANC models \(\Delta _{\textit{IAC}-\textit{IANC}}\) tends to zero as \(n\rightarrow \infty \) or \(m\rightarrow \infty \).
Proof
\(\Delta _{\textit{IAC}-\textit{IANC}}\) is calculated as the difference of two components: (1) the share of N-AECs \(\vartheta \) whose cardinality is grater than average cardinality, \(av(|\vartheta |)\); (2) the share of AECs \(\rho \) which belong to such \(\vartheta , |\vartheta |>av(|\vartheta |)\). Suppose, the set \(\{\vartheta :\vartheta >av(|\vartheta |)\}\) consists of \(R_1\) N-AECs of cardinality m! (containing \(K_1\) AECs), \(R_2\) N-AECs of cardinality m! / 2 (containing \(K_2\) AECs), \(R_3\) N-AECs of cardinality m! / 3(containing \(K_3\) AECs) and so on. Then,
The next part of the proof is done for \(n\rightarrow \infty \), for \(m\rightarrow \infty \) the proof is analogous. By Theorem 3,
Summing up,
\(\square \)
Appendix 2
Rights and permissions
About this article
Cite this article
Veselova, Y. The difference between manipulability indices in the IC and IANC models. Soc Choice Welf 46, 609–638 (2016). https://doi.org/10.1007/s00355-015-0930-3
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-015-0930-3