## Abstract

A voting center is in charge of collecting and aggregating voter preferences. In an iterative process, the center sends comparison queries to voters, requesting them to submit their preference between two items. Voters might discuss the candidates among themselves, figuring out during the elicitation process which candidates stand a chance of winning and which do not. Consequently, strategic voters might attempt to manipulate by deviating from their true preferences and instead submit a different response in order to attempt to maximize their profit. We provide a practical algorithm for strategic voters which computes the best manipulative vote and maximizes the voter’s selfish outcome when such a vote exists. We also provide a careful voting center which is aware of the possible manipulations and avoids manipulative queries when possible. In an empirical study on four real world domains, we show that in practice manipulation occurs in a low percentage of settings and has a low impact on the final outcome. The careful voting center reduces manipulation even further, thus allowing for a non-distorted group decision process to take place.We thus provide a core technology study of a voting process that can be adopted in opinion or information aggregation systems and in crowdsourcing applications, e.g., peer grading in massive open online courses.

This is a preview of subscription content, access via your institution.

## Notes

Notice that \(CL({{\mathbf {Q}}})\) is a composition of \(CL(Q_i)\), i.e. transitive closures of individual voter response sets.

Notice that such a pair always exists.

## References

Aziz H, Brill M, Fischer F, Harrenstein P, Lang J, Seedig HG (2015) Possible and necessary winners of partial tournaments. J Artif Intell Res 54:493–534

Aziz H, Lev O, Mattei N, Rosenschein JS, Walsh T (2016) Strategyproof peer selection: mechanisms, analyses, and experiments. In: AAAI, pp 397–403

Balakrishnan S, Chopra S (2012) Two of a kind or the ratings game? Adaptive pairwise preferences and latent factor models. Front Comput Sci 6(2):197–208

Bannikova M, Naamani-Dery L, Obraztsova S, Rabinovich Z, Rosenschein JS (2016) Between fairness and a mistrial: consensus under a deadline. In: The 10th workshop on advances in preference handling (MPref), at the 25th international joint conference on artificial intelligence (IJCAI-2016)

Betzler N, Hemmann S, Niedermeier R (2009) A multivariate complexity analysis of determining possible winners given incomplete votes. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), vol 9, pp 53–58

Branzei S, Caragiannis I, Morgenstern J, Procaccia AD (2013) How bad is selfish voting? In: Proceedings of the AAAI conference on artificial intelligence, pp 138–144

Bredereck R, Faliszewski P, Niedermeier R, Talmon N (2016) Large-scale election campaigns: combinatorial shift bribery. J Artif Intell Res 55:603–652

Brewka G (1991) Nonmonotonic reasoning: logical foundations of commonsense. Cambridge University Press, Cambridge

Brewka G, Dix J, Konolige K (1997) Non-monotonic reasoning: an overview. CSLI Publications, Stanford

Brill M, Conitzer V (2015) Strategic voting and strategic candidacy. In: Proceedings of the AAAI conference on artificial intelligence, vol 15, pp 819–826

Capuano N, Loia V, Orciuoli F (2017) A fuzzy group decision making model for ordinal peer assessment. IEEE Trans Learn Technol 10(2):247–259. https://doi.org/10.1109/TLT.2016.2565476

Chen YL, Cheng LC, Huang PH (2013) Mining consensus preference graphs from users’ ranking data. Decis Support Syst 54(2):1055–1064

Conitzer V, Sandholm T (2005) Communication complexity of common voting rules. In: ACM EC, pp 78–87 https://doi.org/10.1145/1064009.1064018

Conitzer V, Walsh T, Xia L (2011) Dominating manipulations in voting with partial information. In: Proceedings of the AAAI conference on artificial intelligence, pp 638–643

Damart S, Dias LC, Mousseau V (2007) Supporting groups in sorting decisions: methodology and use of a multi-criteria aggregation/disaggregation DSS. Decis Support Syst 43(4):1464–1475

Dasgupta A, Ghosh A (2013) Crowdsourced judgement elicitation with endogenous proficiency. In: Proceedings of the 22nd international conference on World Wide Web. ACM, New York, pp 319–330

de Borda JC (1781) Mémoire sur les élections au scrutin. Histoire de l’Academie Royale des Sciences

Ding N, Lin F (2013) Voting with partial information: What questions to ask? In: Proceedings of the 12th international conference on autonomous agents and multiagent systems (AAMAS), pp 1237–1238

Endriss U, Obraztsova S, Polukarov M, Rosenschein JS (2016) Strategic voting with incomplete information. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), pp 236–242

Farquharson R (1969) Theory of voting. Yale University Press, Yale

García S, Fernández A, Luengo J, Herrera F (2010) Advanced nonparametric tests for multiple comparisons in the design of experiments in computational intelligence and data mining: experimental analysis of power. Inf Sci 180(10):2044–2064. https://doi.org/10.1016/j.ins.2009.12.010

Gibbard A (1973) Manipulation of voting schemes. Econometrica 41(4):587–602

Grandi U, Loreggia A, Rossi F, Venable KB, Walsh T (2013) Restricted manipulation in iterative voting: condorcet efficiency and Borda score. In: International conference on algorithmic decision theory (ADT). Springer, Berlin, pp 181–192

Jiang J, An B, Jiang Y, Lin D (2017) Context-aware reliable crowdsourcing in social networks. IEEE Trans Syst Man Cybern Syst 99:1–16

Kamishima T, Kazawa H, Akaho S (2005) Supervised ordering—an empirical survey. In: Proceedings of the 5th IEEE international conference on data mining (ICDM). IEEE Computer Society, New York, pp 673–676

Kendall MG (1938) A new measure of rank correlation. Biometrika 30(1–2):81–93

Konczak K, Lang J (2005) Voting procedures with incomplete preferences. In: Proceeding of the IJCAI-05 multidisciplinary workshop on advances in preference handling, vol 20, Citeseer

Kukushkin NS (2011) Acyclicity of improvements in finite game forms. Int J Game Theory 40(1):147–177

Laffont JJ (1987) Incentives and the allocation of public goods. In: Auerbach AJ, Feldstein M (eds) Handbook of public economics, vol 2. Elsevier, London, pp 537–569 (Chap 10)

Lev O, Rosenschein JS (2012) Convergence of iterative voting. In: Proceedings of the 11th international conference on autonomous agents and multiagent systems-volume 2, international foundation for autonomous agents and multiagent systems, pp 611–618

Lev O, Rosenschein JS (2016) Convergence of iterative scoring rules. J Artif Intell Res 57:573–591

Lu T, Boutilier C (2011) Robust approximation and incremental elicitation in voting protocols. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), pp 287–293

Lu T, Boutilier C (2013) Multi-winner social choice with incomplete preferences. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), pp 263–270

Mattei N, Walsh T (2013) Preflib: A library of preference data http://preflib.org. In: Proceedings of the 3rd international conference on algorithmic decision theory (ADT 2013). Lecture notes in artificial intelligence. Springer, Berlin

Meir R, Polukarov M, Rosenschein JS, Jennings NR (2010) Convergence to equilibria of plurality voting. In: Proceedings of the AAAI conference on artificial intelligence, pp 823–828

Meir R, Lev O, Rosenschein JS (2014) A local-dominance theory of voting equilibria. In: Proceedings of the 15th ACM EC, pp 313–330

Miller GA (1956) The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychol Rev 63(2):81

Naamani-Dery L, Golan I, Kalech M, Rokach L (2014a) Preference elicitation for group decisions. In: Group decision and negotiation conference 2014, Toulouse, France, June 10–13, 2014. Proceedings, pp 193–200

Naamani-Dery L, Kalech M, Rokach L, Shapira B (2014b) Reaching a joint decision with minimal elicitation of voter preferences. Inf Sci 278:466–487. https://doi.org/10.1016/j.ins.2014.03.065

Naamani-Dery L, Golan I, Kalech M, Rokach L (2015a) Preference elicitation for group decisions using the borda voting rule. Group Decis Negot 41(6):1015–1033

Naamani-Dery L, Obraztsova S, Rabinovich Z, Kalech M (2015b) Lie on the fly: iterative voting center with manipulative voters. In: Yang Q, Wooldridge M (eds) Proceedings of the 24th international joint conference on artificial intelligence, IJCAI 2015, Buenos Aires, Argentina, July 25–31, 2015. AAAI Press, New York, pp 2033–2039. http://ijcai.org/Abstract/15/288

Naamani-Dery L, Kalech M, Rokach L, Shapira B (2016) Reducing preference elicitation in group decision making. Expert Syst Appl 61:246–261

Obraztsova S, Rabinovich Z, Madunts A (2014) Faustian dynamics in Sarkar’s social cycle. In: Proceedings of the 21st European conference on artificial intelligence (ECAI-2014), pp 1071–1072

Obraztsova S, Elkind E, Polukarov M, Rabinovich Z (2015a) Strategic candidacy games with lazy candidates. In: Proceedings of the 24th international joint conference on artificial intelligence (IJCAI-2015), pp 610–616

Obraztsova S, Markakis E, Polukarov M, Rabinovich Z, Jennings NR (2015b) On the convergence of iterative voting: How restrictive should restricted dynamics be? In: Proceedings of the AAAI conference on artificial intelligence, pp 993–999

Obraztsova S, Lev O, Polukarov M, Rabinovich Z, Rosenschein JS (2016) Non-myopic voting dynamics: an optimistic approach. In: The 10th workshop on advances in preference handling (MPref), at the 25th international joint conference on artificial intelligence (IJCAI-2016)

Pini MS, Rossi F, Venable KB, Walsh T (2007) Incompleteness and incomparability in preference aggregation. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), pp 1464–1469

Polukarov M, Obraztsova S, Rabinovich Z, Kruglyi A, Jennings NR (2015) Convergence to equilibria in strategic candidacy. In: Proceedings of the 24th international joint conference on artificial intelligence (IJCAI-2015), pp 624–630

Rabinovich Z, Obraztsova S, Lev O, Markakis E, Rosenschein JS (2015) Analysis of equilibria in iterative voting schemes. In: Proceedings of the AAAI conference on artificial intelligence, pp 1007–1013

Reijngoud A, Endriss U (2012) Voter response to iterated poll information. In: Proceedings of the 11th international conference on autonomous agents and multiagent systems (AAMAS), pp 635–644

Reyhani R, Wilson M (2012) Best reply dynamics for scoring rules. In: Proceedings of the European conference on artificial intelligence (ECAI), pp 672–677

Roughgarden, T., & Schrijvers, O. (2017, December). Online prediction with selfish experts. In Proceedings of the 31st international conference on neural information processing systems, Curran Associates Inc., pp 1300–1310

Satterthwaite MA (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

Walsh T (2007) Uncertainty in preference elicitation and aggregation. In: Proceedings of the AAAI conference on artificial intelligence, pp 3–8

Walsh T (2011) Where are the hard manipulation problems? J Artif Intell Res 42:1–29

Wang W, Jiang J, An B, Jiang Y, Chen B (2017) Toward efficient team formation for crowdsourcing in noncooperative social networks. IEEE Trans Cybern 47(12):4208–4222. https://doi.org/10.1109/TCYB.2016.2602498

Xia L, Conitzer V (2011) A maximum likelihood approach towards aggregating partial orders. In: Proceedings of the international joint conference on artifical intelligence (IJCAI), pp 446–451

Yuen MC, King I, Leung KS (2011) A survey of crowdsourcing systems. In: Privacy, security, risk and trust (PASSAT) and 2011 IEEE 3rd international conference on social computing (SocialCom), 2011 IEEE 3rd international conference on. IEEE, New York, pp 766–773

Zou J, Meir R, Parkes D (2015) Strategic voting behavior in Doodle polls. In: Proceedings of the 18th ACM conference on computer-supported cooperative work and social computing (CSCW), pp 464–472

## Author information

### Authors and Affiliations

### Corresponding author

## Additional information

### Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Appendix

### Appendix

### Proof: Theorem 1

Let us assume that for some *l* holds \(pw_{l+1}\succ pw_l\). Consider a partial joint profile, \({\mathcal {R}}_{-i}\), where \({\displaystyle {score}}(pw_l,{\mathcal {R}}_{-i})={\displaystyle {score}}(pw_{l+1},{\mathcal {R}}_{-i})=\eta \) and all other voters have a score of at most \(\eta -m\). \({\mathcal {R}}_{-i}\) is a possible (partial) joint profile given the set \(PW_i\). Now, if the voter *i* submits \(P_i\), then \(pw_l\) will become the winner. If \(P_i'\) is submitted, then \(pw_{l+1}\) will win the elections. This contradicts the definition of dominance: \(P_i'\) does not dominate \(P_i\).

Let us now assume that for some *l* holds \(|[pw_l\ ;P_i'\ ;pw_{l+1}]|< |[pw_l\ ;P_i\ ;pw_{l+1}]|\). Similar to the previous case, construct a possible (partial) joint profile \({\mathcal {R}}_{-i}\) so that \({\displaystyle {score}}(pw_{l+1},{\mathcal {R}}_{-i})-{\displaystyle {score}}(pw_l,{\mathcal {R}}_{-i})=|[pw_l\ ;P_i\ ;pw_{l+1}]|-\eta \). Where \(\eta =1\) if \(pw_l\) beats \(pw_{l+1}\) in tie-breaking and \(\eta =2\) otherwise. Furthermore, \({\mathcal {R}}_{-i}\) can be such that the score of all other candidates is at least *m* points less than \({\displaystyle {score}}(pw_l,{\mathcal {R}}_{-i})\). As before, if \(P_i\) is submitted by the voter *i*, then \(pw_l\) wins the elections, and if \(P_i'\) is submitted, then \(pw_{l+1}\) wins. Again, this contradicts \(P_i'\) dominating \(P_i\).

Lastly, assuming that \(|[pw_l\ ;P_i'\ ;pw_{l+1}]|\ge |[pw_l\ ;P_i\ ;pw_{l+1}]|\) for all \(l\in [1,\ldots ,k-1]\) holds, but there is *no*\(l\in [1,\ldots ,k-1]\) so that \(|[pw_l\ ;P_i'\ ;pw_{l+1}]|\gneq |[pw_l\ ;P\ ;pw_{l+1}]|\). In this case for any \({\mathcal {R}}=({\mathcal {R}}_{-i},P_i)\) holds that \({{\mathcal {F}}}({\mathcal {R}})={{\mathcal {F}}}({\mathcal {R}}_{-i},P_i')\). Hence \(P_i'\) can not dominate \(P_i\).

We conclude that all three conditions are necessary for \(P_i'\) to dominate \(P_i\). Furthermore, a simple reinspection of the proof quickly leads to the sufficiency of the conditions. \(\square \)

### Interactive Local Dominance Response Analysis

Theorem 1 has given a higher level structure to the set of possible manipulations in our model. It has allowed the reader to build intuition, comprehend our algorithm construction and understand their application examples. Now, it is possible to provide the detailed theoretical treatment, and in this section we will provide complete definitions supporting the algorithmic design. Proofs of all the theorems and

### Theorem 2

Assume that ”Common Givens” w.l.o.g. conditions hold. Algorithm 1 works in polynomial time in the number of voters and candidates, and finds a \(P_i'\) that satisfies Condition-1 and Condition-2, if such a preference profile exists.

Before we prove Theorem 2, i.e. the correctness of Algorithm 1, we provide a set of lemmas that are needed for supporting the proof. All lemmas adopt the “Common Givens” w.l.o.g. assumptions mentioned in Sect. 4.1. Let us examine voter \(v_i\)’s profile. The preferred order of possible winners according to \(v_i\) is: \(P_i=[pw_1,pw_2,\ldots ,pw_l\)]. When \(v_i\) is queried for her preference between \(c_j\) and \(c_k\), her response is: \(c_j\succ c_k\). We would like to build a new profile \(P_i'\) where voter \(v_i\)’s response to the same query is: \(c_k\succ c_j\). We need \(P_i'\) to satisfy conditions Condition-1 and Condition-2, i.e. \(P_i'\) should be a local dominant profile over \(P_i\) and also have the minimal swap distance to \(P_i\) out of all possible profiles.

The only way to create a profile \(P_i'\), that is local dominant and has a minimal swap distance, is if in profile \(P_i\):

\(c_j\) is above \(pw_1\) and \(c_k\) is below \(pw_l\): \(P_i\ \ :\ \ \dots \succ c_j\succ \dots \succ pw_1\succ \dots \succ pw_l\succ \dots \succ c_k\succ \dots \)

\(c_j\) is between \(pw_1\) and \(pw_l\) and \(c_k\) is below \(pw_l\): \(P_i\ \ :\ \ \dots \succ pw_1\succ \dots \succ c_j\succ \dots \succ pw_l\succ \dots \succ c_k\succ \dots \)

\(c_j\) is above \(pw_1\) and \(c_k\) is between \(pw_1\) and \(pw_l\): \(P_i\ \ :\ \ \dots \succ c_j\succ \dots \succ pw_1\succ \dots \succ c_k\succ \dots \succ pw_l\succ \dots \)

As an example of the latter case, if \(P_i=[c_j,\ldots ,pw_1,\ldots ,c_k,...pw_l]\) then switching between \(c_j\) and \(c_k\) by adding \(c_j\) to the sequence that is below \(pw_1\) and above \(pw_l\): \(P_i'= [pw_1,\ldots ,c_k,c_j ,\ldots , pw_l]\) results in a profile \(P_i'\) that is a local dominant with a minimal swap distance, i.e. satisfies both conditions Condition-1 and Condition-2.

Let us consider the alternatives. If in profile \(P_i\) both \(c_j\) and \(c_k\) are either:

below \(pw_1\) and above \(pw_l\): \(P_i\ \ :\ \ \dots \succ pw_1\succ \dots \succ c_j\succ \dots \succ c_k\succ \dots \succ pw_l\succ \dots \)

below \(pw_l\): \(P_i\ \ :\ \ \dots \succ pw_1\succ \dots \succ pw_l\succ \dots \succ c_j\succ \dots \succ c_k\succ \dots \)

above \(pw_1\): \(P_i\ \ :\ \ \dots \succ c_j\succ \dots \succ c_k\succ \dots \succ pw_1\succ \dots \succ pw_l\succ \dots \)

Then for \(P_i'\) to be a local dominant profile over \(P_i\), the total distance between \(pw_1\) and \(pw_l\) should increase with respect to the total distance between \(pw_1\) and \(pw_l\) in profile \(P_i\). Therefore, in these cases one must not only switch between \(c_j\) and \(c_k\) but must also insert at least one candidate between \(pw_1\) and \(pw_l\) so that the total distance is increased. However, inserting a candidate between \(pw_1\) and \(pw_l\) results in a profile \(P_i'\) that is local dominant but does not have a minimal swap distance.

Formally, the above descriptions can be expressed as:

### Lemma 1

Assume that there is \(P_i'\ne P_i\) that satisfies Condition-1 and Condition-2. Then either of the following holds:

\(P_i(c_j,pw_1)\) and \(P_i(pw_l,c_k)\);

\(P_i(c_j,pw_1)\) and \(c_k\in [pw_1\ ;P_i\ ;pw_l]\);

\(P_i(pw_l,c_k)\) and \(c_j\in [pw_1\ ;P_i\ ;pw_l]\).

### Proof: Lemma 1

Let us assume the contrary, i.e. that, in addition to Condition-1 and Condition-2, either of the following holds :

\(c_j,c_k\in [pw_1\ ;P_i\ ;pw_l]\)

\(c_j,c_k\in (+\infty \ ;P_i\ ;pw_l]\)

\(c_j,c_k\in [pw_l\ ;P_i\ ;-\infty )\)

Because Condition-1 holds for \(P_i'\), i.e. \(P_i'\) locally dominates \(P_i\), it follows from Theorem 1 that

Hence, there is a candidate \(c\in C\) so that either \(P_i(c,pw_1)\) and \(P_i'(pw_1,c)\), or \(P_i(pw_l, c)\) and \(P_i'(c,pw_l)\). Due to the symmetry of these two cases, let us assume without loss of generality that the former case holds, i.e. \(P_i(c,pw_1)\) and \(P_i'(pw_1,c)\). Let us assume that *c* is the highest candidate for which this condition holds with respect to \(P_i'\). Formally:

Let \(c'\) be the candidate immediately above *c* w.r.t \(P_i'\), i.e. \(P_i'(c',c)\) and the segment \((c'\ ;P_i'\ ;c)=\emptyset \). Let us show that the candidate pair \((c',c)\ne (c_k,c_j)\), in each of the contrary sub-cases:

If \(c_j,c_k\in [pw_1\ ;P_i\ ;pw_l]\), then \(c\ne c_j\) and \(c\ne c_k\) since \(c\in (+\infty \ ;P_i\ ;pw_1]\).

If \(c_j,c_k\in (+\infty \ ;P_i\ ;pw_1]\), then \(c\ne c_j\), otherwise we obtain contradiction to the Eq. 1, because \(P_i'(c_k,c_j)\).

If \(c_j,c_k\in [pw_l\ ;P_i\ ;-\infty )\), then \(c\ne c_j\) and \(c\ne c_k\) since \(c\in (+\infty \ ;P_i\ ;pw_1]\).

Furthermore, *c* and \(c'\) are such that \(P_i(c,c')\). Otherwise we again obtain contradiction to Eq. 1, since by the choice of *c* and \(c'\) holds that \(P_i(c,pw_1)\) and \(\left| (pw_1\ ;P_i\ ;c']\right| \ge 1\) (i.e. \(c'\) is either \(pw_1\) or below it).

Let us then consider \(P_i''\) obtained from \(P_i'\) by swapping *c* and \(c'\). It is easy to see that \({d_{swap}}(P_i,P_i'')\lneq {d_{swap}}(P_i,P_i')\), yet \(P_i''\models CL(Q_i\cup {(c_k,c_j)})\). This contradicts the assumption that Condition-2 holds for \(P_i'\). \(\square \)

As before, let us assume that in \(P_i\), \(c_j\succ c_k\). In \(P_i'\) the order of these two candidates is switched so that \(c_k\succ c_j\). Let us denote the set of all profiles that have a minimal swap distance from \(P_i\) as \(\mu (P_i)\). In order for \(P_i' \in \mu (P_i)\), i.e, in order for \(P_i'\) to have a minimal swap distance from \(P_i\), \(c_k\) and \(c_j\) need to be ordered directly one after the other, with no other candidates separating them. Formally:

### Lemma 2

Let \((c_j,c_k)\) be the query, and let there be *c* so that \(P_i'(c_k,c)\) and \(P_i'(c,c_j)\), i.e. \((c_k; P_i'; c_j)\ne \emptyset \), then \(P_i'\not \in \mu (P_i)\).

### Proof: Lemma 2

Let us have a closer look at the closed interval \([c_k; P_i'; c_j]\). There is a pair of candidates \((c,c')\in [c_k; P_i';c_j]\), so that \(P_i(c',c)\) and \((c; P_i', c')=\emptyset \). Because \((c_k; P_i'; c_j)\ne \emptyset \), it holds that \((c,c')\ne (c_k,c_j)\). Let \(P_i''\) be a preference order obtained from \(P_i'\) by swapping *c* and \(c'\). It is easy to see that \(P_i''\models Q_i\) and \({d_{swap}}(P_i,P_i'')\lneq {d_{swap}}(P_i,P_i')\). I.e. \(P_i'\not \in \mu (P_i)\). \(\square \)

Besides the proximity of \(c_j,c_j\), we can also show that certain sets of elements remain in their original order. In particular, the following lemma show that two sub-sets of elements, those with the closest consistent \(P_i'\in \mu (P_i)\) places either above \(c_k\) or below \(c_j\), inherit their relative order from \(P_i\).

### Lemma 3

Let \(P_i'\in \mu (P_i)\), then the following two equations hold

### Proof: Lemma 3

Let us assume that the Eq. 2 does not hold. Then, there are two candidates, \(c,c'\) so that \((c';P_i';c)=\emptyset \), \(P_i'(c',c)\) and \(P_i(c,c')\). Furthermore, it holds that \(P_i'(c,c_k)\). Let us define a new preference order \(P_i''\) obtained from \(P_i'\) by swapping *c* and \(c'\). Then \({d_{swap}}(P_i,P_i'')<{d_{swap}}(P_i,P_i')\) and \(P_i''\models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \), i.e. \(P_i'\not \in \mu (P_i)\), contradicting the lemma’s premise.

We obtain the same kind of contradiction by assuming that Eq. 3 does not hold. Hence the Lemma’s conclusion: both Eqs. 2 and 3 must hold. \(\square \)

Furthermore, if we consider two candidates that the original preference order \(P_i\) places outside the span between \(c_j\) and \(c_k\), then they demarcate an upper and a lower candidate intervals that maintain both their order *and* composition in \(P_i'\).

### Lemma 4

Let \(P_i'\in \mu (P_i)\), and let \(c_j',c_k'\in C\) so that \(P_i(c_j',c_j)\) and \(P_i(c_k,c_k')\). Then the following equations hold

### Proof: Lemma 4

Let us assume that the Eq. 4 does not hold, in spite of the lemma’s premise being true. That is, there exists a candidate \(c_j'\) so that \(P_i(c_j',c_j)\) and \((\infty \ ;P_i\ ;c_j']\ne (\infty \ ;P_i'\ ;c_j']\).

Three possible sub-cases exist in this context:

- 1.
\(\exists c\in C\ s.t.\ P_i(c_j',c)\wedge P_i'(c,c_j')\)

- 2.
\(\exists c\in C\ s.t.\ P_i(c,c_j')\wedge P_i'(c_j',c)\)

- 3.
Neither of the above holds.

If \(\exists c\in C\ s.t.\ P_i(c_j',c)\wedge P_i'(c,c_j')\), then it is easy to see that a pair of candidates \((c,c')\) exists so that \((c\ ;P_i'\ ;c')=\emptyset \), \(P_i'(c,c')\), and either \(P_i(c',c_j')\) or \(c'=c_j'\). Let us then obtain a preference order \(P_i''\) from \(P_i'\) by swapping *c* and \(c'\). It holds that \({d_{swap}}(P_i,P_i'')<{d_{swap}}(P_i,P_i')\) and, in addition, \(P_i' \models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \). Hence, we contradict the lemma’s premise that \(P_i'\in \mu (P_i)\).

The sub-case where it holds that \(\exists c\in C\ s.t.\ P_i(c,c_j')\wedge P_i'(c_j',c)\) is similar to the above.

Let us now investigate the third sub-case. It occurs if there is no element that has switched from being above (below) \(c_j'\) in \(P_i\) to being below (above) \(c_j'\) in \(P_i'\). In particular the following two sets are equal (as sets):

If \((\infty \ ;P_i\ ;c_j')=\emptyset \), then the assumption of Eq. 4 not being true can not hold. If, however, \((\infty \ ;P_i\ ;c_j')\ne \emptyset \), then \(P_i\downarrow _{B}\ne P_i'\downarrow _{B}\). That is, there is a pair of candidates \(c,c'\in B\) so that \((c'\ ;P_i'\ ;c)=\emptyset \), \(P_i(c,c')\) and \(P_i'(c',c)\). Defining an alternative order \(P_i''\) obtained from \(P_i'\) by swapping *c* and \(c'\), we once again obtain a contradiction to the premise \(P_i'\in \mu (P_i)\).

We conclude that Eq. 4 must hold. Symmetric proof establishes Eq. 5. \(\square \)

Now, as Lemma 2 showed, \(c_j\) and \(c_k\) are placed next to each other, when changing the preference order from \(P_i\) to \(P_i'\). However, to achieve this some other elements may need to be separated. The following lemma shows that this does not occur without need. That is, if two elements were placed next to each other in \(P_i\), but not in \(P_i'\), then they were separated to accommodate the placement of \(c_k\) and \(c_j\) between them.

### Lemma 5

Let \(P_i'\in \mu (P_i)\), and let \(a,b\in C\) be two candidates so that \(P_i\downarrow _{\{a,b\}}=P_i'\downarrow _{\{a,b\}}\), \((a\ ;P_i\ ;b)=\emptyset \), and \((a\ ;P_i'\ ;b)\ne \emptyset \). Then \(a\in (\infty \ ;P_i'\ ;c_k]\) and \(b\in [c_j\ ;P_i'\ ;-\infty )\).

### Proof: Lemma 5

Let us assume that the Lemma’s conclusion does not hold. In particular this would mean that \(c_j,c_k\not \in (a\ ;P_i'\ ;b)\). On the other hand, \((a\ ;P_i'\ ;b)\ne \emptyset \), so there is a candidate \(c\in (a\ ;P_i'\ ;b)\). Because *a* and *b* are next to each other in the preference ordering \(P_i\), i.e. \((a\ ;P_i\ ;b)=\emptyset \), it holds that either \(P_i(c,a)\) or \(P_i(b,c)\). Which, in turn, implies that \(P_i\downarrow _{[a\ ;P_i'\ ;b]}\ne P_i'\downarrow _{[a\ ;P_i'\ ;b]}\). Therefore, there is \(c'\in [a\ ;P_i'\ ;b]\) so that \(c'\ne c\) and \((c'\ ;P_i'\ ;c)=\emptyset \), i.e. *c* and \(c'\) are next to each other in the ordering \(P_i'\). Furthermore, it must hold that these two elements were switched betwen \(P_i\) and \(P_i'\), that is \(P_i(c,c')\) and \(P_i'(c',c)\). Let us define a new preference order \(P_i''\) by swapping *c* and \(c'\) in \(P_i'\). It would hold that \({d_{swap}}(P_i'',P_i)\lneq {d_{swap}}(P_i',P_i)\), while \(P_i''\models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \), thus contradicting the premise that \(P_i'\in \mu (P_i)\). \(\square \)

One final observation that we will need to prove Theorem 2 has to do with the general change in the relative position of elements committed by \(Q_i\) to a particular order w.r.t \(c_k\) or \(c_j\). Lemma 6 shows that among all elements above (correspondingly, below) \(c_k\) (correspondingly, \(c_j\)) only those committed to be ordered after \(c_j\) (correspondingly, before \(c_k\)) will change their relative position when moving from preference order \(P_i\) to \(P_i'\). All other elements will maintain their order.

### Lemma 6

Let us assume that \(P_i'\in \mu (P_i)\). Let *c* be some candidate so that \(P_i'(c,c_k)\) and \((c,c_k)\not \in Q_i\). Then the following equality holds:

Symmetrically, let *c* be some candidate so that \(P_i'(c_j,c)\) and \((c_j,c)\not \in Q_i\). Then:

### Proof: Lemma 6

First, notice that if \(c\not \in (c_j\ ;P_i\ ;c_k)\), then the lemma is a direct conclusion if Lemma 4. In more detail, if \(P_i(c,c_j)\), then only the premise of the first equation holds. Furthermore, since \(P_i\models Q_i\), \((\infty \ ;P_i\ ;c]\cap [c_j\ ;Q_i\ ;-\infty )=\emptyset \). Thus, the lemma’s conclusion requires that \((\infty \ ;P_i'\ ;c]=(\infty \ ;P_i\ ;c]\), which holds due to Lemma 4. Symmetrically, if \(P_i(c_k,c)\), then the premise of the second equation is true, and the conclusion similarly holds according to Lemma 4. Therefore, in the remainder of this proof, we will assume that \(c\in (c_j\ ;P_i\ ;c_k)\).

Now, let us assume that there is in fact a candidate \(c\in C\) that satisfies the first premise of the lemma, but violates its conclusion. Denote by *X* the following set:

Let *x* denote the least preferred candidate of *X* w.r.t the preference order \(P_i'\), i.e., for any \(x\ne c\in X\) holds that \(P_i'(c,x)\).

From Lemma 3 we know that \(P_i'\downarrow _{(+\infty \ ;P_i'\ ;x]}=P_i\downarrow _{(+\infty \ ;P_i'\ ;x]}\). Which also means that \((+\infty \ ;P_i'\ ;x]\subseteq (+\infty \ ;P_i\ ;x]\). Furthermore, since \(P_i'\models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \), for all \(c\in [c_j\ ;Q_i\ ;-\infty )\) holds that \(P_i'(c_j,c)\) or \(c=c_j\). Since \(P_i'(x,c_j)\), by the transitivity of \(P_i'\) it is also true that \(P_i'(x,c)\) for all \(c\in [c_j\ ;Q_i\ ;-\infty )\). Hence, we obtain \((+\infty \ ;P_i'\ ;x]\cap [c_j\ ;Q_i\ ;-\infty )=\emptyset \). In addition, since we have assumed that the lemma’s conclusion does not hold, we obtain the following strong subsumption:

This means, in particular, that there is a candidate \(y\in C\) so that \(P_i(y,x)\), \(P_i'(x,y)\), and \((c_j,y)\not \in Q_i\). Taking into account Lemma 3, \(P_i\) and \(P_i'\) have the following overall structures:

\(P_i\ \ :\ \ \dots \succ c_j\succ \dots \succ y\succ \dots \succ x\succ \dots \succ c_k\succ \dots \)

\(P_i'\ \ :\ \ \dots \succ x\succ \dots \succ c_k\succ c_j\succ \dots \succ y\succ \dots \)

Let us denote *A* the number of candidates between *x* and \(c_k\) with respect to \(P_i'\), i.e. \(A=|(x\ ;P_i'\ ;c_k]|\), and, correspondingly \(B=|[c_j\ ;P_i'\ ;y)|\).

Consider now alternative preference orderings *R* and \(R'\), obtained from \(P_i'\) by either moving *x* below *y* or, alternatively, moving *y* just below *x*. That is, *P* and \(P'\) have the following structures:

\(R\ \ :\ \ \dots \succ c_k\succ c_j\succ \dots \succ y\succ x\succ \dots \)

\(R'\ \ :\ \ \dots \succ y\succ x\succ \dots \succ c_k\succ c_j\succ \dots \)

Furthermore, \(P_i'\downarrow _{C{\setminus }\{x,y\}}=R\downarrow _{C{\setminus }\{x,y\}}=R'\downarrow _{C{\setminus }\{x,y\}}\). Let us now denote by \(D={d_{swap}}(P_i,P_i')\), and consider \({d_{swap}}(P_i,R)\) and \({d_{swap}}(P_i,R')\).

It holds that \(P_i\downarrow _{[c_j\ ;P_i'\ ;y)}=R\downarrow _{[c_j\ ;P_i'\ ;y)}\), while \(P_i'(x,c)\) for any \(c\in [c_j\ ;P_i'\ ;y)\), hence *P* is closer to \(P_i\) by *B* element swaps. At the same time \(P_i\downarrow _{(x\ ;P_i'\ ;c_k]}=P_i'\downarrow _{(x\ ;P_i'\ ;c_k]}\), yet *R*(*c*, *x*) for all \(c\in (x\ ;P_i'\ ;c_k]\). Similarly the order of *x* and *y* is also ”restored”, i.e. it holds that *R*(*y*, *x*), \(P_i(y,x)\) and \(P_i'(x,y)\). As a result we have \({d_{swap}}(P_i,P)=D-B+A-1\). Similarly \({d_{swap}}(P_i,R')=D-A+B-1\). Since either \(-A+B-1<0\) or \(-B+A-1<0\), we have that either *R* or \(R'\) is closer to \(P_i\) than \(P_i'\). Because no pair of candidates \(x,y,c_j,c_k\) is restricted by \(Q_i\), we also have that both \(R\models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \) and \(R'\models CL\left( Q_i\cup \{(c_k,c_j)\}\right) \), therefore violating the assumption of \(P_i'\in \mu (P_i)\). \(\square \)

These lemmas are the setting for the proof of Theorem 2—the correctness of our algorithm.

### Proof: Theorem 2

First, let us do away with the question of computational complexity of the Algorithm 1, as the simpler portion of the algorithm’s analysis.

Prior to the main loop of the algorithm, a preliminary feasibility of manipulation is run in Line 2 , based on Lemma 1. The Lemma includes a finite number of membership checks, each of which runs in time linear in the number of candidates. It does, however, presume that the set of possible winners can be obtained efficiently. Since we use the definition of the PW set from (Konczak and Lang 2005, Lu and Boutilier 2013, Naamani-Dery et al. 2014), it can be found efficiently in the number of voters and candidates. Therefore, the overall preliminary check of Line 2 is polynomial in both voter and candidate set sizes.

Once the pre-check is complete the main loop of the algorithm is repeated for every candidate in the worst case. If we show that each loop is polynomial in the size of the problem as well, the overall algorithm’s complexity will be obtained.

Lines 7-10 operate on ordered subsets of the candidate set, and each such operation takes at most 2|*C*| basic steps to complete. The arguments of these operations are also polynomial-time constructed. One, \(P_i\), is given explicitly as input, and taking a sub-interval of it is linear in |*C*|. The other is obtained, e.g., by a spanning tree traversal of \(Q_i\). However, \(Q_i\) set is at most quadratic in the number of candidates. Hence, the calculates of sets \(X_{good},X_{bad}, Y_{good}\) and \(Y_{bad}\) take time, polynomial in the candidate set size. Line 11 is a sanity check, since the aforementioned subsets of *C* were obtained from an ordered sequence of candidates, hence linear in |*C*|. Similarly, construction of the new preference order \(P_i'\) takes linear time. Calculating the distance \({d_{swap}}(P_i,P_i')\) (line 13) takes at most \(|C|^2\) steps, as it is equivalent to running the bubble-sort algorithm. The last non-trivial step, line 17, depends on how efficiently we can confirm the Local Dominance property of \(P_i'\) with respect to \(P_i\). This confirmation, however, can be performed by using the three conditions of Theorem 1. As we have already accounted for the calculation of the PW set, each condition of the Theorem takes time linear in |*C*|. We conclude that the main loop runs at most in \(O(|C|^3)\). Hence, the overall run time of the algorithm is polynomial in the sizes of *V* and *C* sets.

Now, given that we know that the Algorithm 1 operates in polynomial time, let us prove that it operates correctly.

Let \(P_i'\in \mu (P_i)\), and let us analyze its structure.

Let us assume for the moment, that there are two elements \(a_1,b_2\in C\) that satisfy the conditions of Lemma 6, i.e., \(P_i'(a_1,c_k), P_i'(c_j,b_2)\) and \((a_1,c_k)\not \in Q_i\), \((c_j,b_2)\not \in Q_i\). Furthermore, let us assume that \(a_1\) is the minimum (\(b_2\) is the maximum) such element with respect to \(P_i'\). Combining this assumption with Lemma 2, \(P_i'\) can be broken down into the following structure \(P_i'=(F_j,G_k,c_k,c_j,G_j,F_k)\), where the intervals \(F_j,F_k,G_j,G_k\) are characterized as follows:

\(F_j=(\infty \ ;P_i'\ ;a_1]\),

\(F_k = [b_2\ ;P_i'\ ;-\infty )\),

\(G_k\subset (\infty \ ;Q_i\ ;c_k)\)

\(G_j\subset (c_j\ ;Q_i\ ;-\infty )\)

Let us now denote \(a_2,b_1\in C\), so that \(P_i(a_1,a_2), P_i(b_1,b_2)\) and \((a_1\ ;P_i\ ;a_2)=\emptyset \), \((b_1\ ;P_i\ ;b_2)=\emptyset \). Such \(a_2\) and \(b_1\) exist, since \(P_i(a_1,c_k)\) and \(P_i(c_j,b_2)\) due to Lemma 3. Furthermore, it also entails that \(P_i(a_2,c_k)\) and \(P_i(b_1,c_j)\).

The following three subcases are possible.

Either \((a_1\ ;P_i'\ ;a_2)\ne \emptyset \) or \((b_1\ ;P_i'\ ;b_2)\ne \emptyset \).

**Sub-Case A**\(a_2=b_2\)**Sub-Case B**\(P_i'(a_2,b_2)\)

**Sub-Case C**Both \((a_1\ ;P_i'\ ;a_2)=\emptyset \) and \((b_1\ ;P_i'\ ;b_2)=\emptyset \).

W.l.g., let us first assume that \((a_1\ ;P_i'\ ;a_2)\ne \emptyset \). Then, according to Lemma 5, \(P_i'(c_j,a_2)\). Combining this with Lemma 3, we obtain that \(P_i(c_j,a_2)\). Thus, we also conclude that \(a_2\in [c_j\ ;P_i\ ;c_k]\).

*Sub-Case A* If in addition, \(a_2=b_2\), then the following holds according to Lemma 6 and seting \(z=a_2\) in Algorithm 1:

By its definition \(G_j\subset (c_j\ ;Q_i\ ;-\infty )\). Furthermore, \(G_j\subset [c_j\ ;P_i'\ ;a_2]\). Thus, by Lemma 3, \(G_j\subset [c_j\ ;P_i\ ;a_2]\). Since \(a_2\not \in G_j\), we conclude that \(G_j\subset (\infty \ ;P_i\ ;a_2)\). Hence, by setting \(z=a_2\) in Algorithm 1, we have:

Similarly, \(G_k\subset (\infty \ ;Q_i\ ;c_k)\) by definition. Furthermore, \(G_k\subset [a_1\ ;P_i'\ ;c_k]\). Thus, by Lemma 3, \(G_k\subset [a_1\ ;P_i\ ;c_k]\). Since \(a_1\not \in G_k\) and \((a_1\ ;P_i\ ;a_2)=\emptyset \), we can conclude that \(G_k\subset [a_2\ ;P_i\ ;-\infty )\). Letting \(z=a_2\) in Algorithm 1, we have:

Thus we have \(P_i'=(X_{good},Y_{bad},c_k,c_j,X_{bad},Y_{good})\) during a run of the Algorithm 1, where \(z=a_2\). That is, this sub-case of \(P_i'\in \mu (P_i)\) will be recovered by the Algorithm 1.

*Sub-Case B* Let us now consider the situation where, rather than \(a_2=b_2\), we have \(P_i'(a_2,b_2)\). Similar to the case where \(a_2=b_2\), we will have that \(a_2\in [c_j\ ;P_i\ ;c_k]\) and that \(X_{good}=F_j\), \(Y_{bad}=G_k\) when Algorithm 1 constructs a hypothetical manipulative preference profile with \(z=a_2\). It remains to show that \(X_{bad}\) and \(Y_{good}\) combine into the segment \((c_j\ ;P_i'\ ;-\infty )=(G_j,F_k)\), and then conclude that, even if \(P_i'(a_2,b_2)\), the preference profile \(P_i'\) will be discovered by Algorithm 1 for \(z=a_2\). To this end, let us have a closer look at segments \([a_2\ ;P_i'\ ;b_2]\) and \([a_2\ ;P_i\ ;b_2]\).

Let there be \(x\in [a_2\ ;P_i\ ;b_2]{\setminus }[a_2\ ;P_i'\ ;b_2]\). If \(P_i'(c_j,x)\), then, according to Lemma 3, we obtain a contradiction that \(x\in [a_2\ ;P_i'\ ;b_2]\). Hence \(P_i'(x,c_k)\). On the other hand, it must be that \(P_i(a_1,x)\), since \(P_i(a_1,a_2)\) and \(x\in [a_2\ ;P_i\ ;b_2]\). Hence, \(x\in G_k\subset (\infty \ ;Q_i\ ;c_k)\). Notice also that, due to Lemma 3, we have \([a_2\ ;P_i'\ ;b_2]{\setminus }[a_2\ ;P_i\ ;b_2]=\emptyset \). As a result, \([a_2\ ;P_i'\ ;-\infty )=[a_2\ ;P_i\ ;-\infty ){\setminus } (\infty \ ;Q_i\ ;c_k)=Y_{bad}\), where \(Y_{bad}\) is computed for \(z=a_2\).

Finally, notice that it is impossible to have \(P_i'(b_2,a_2)\), and that the reasoning is symmetric for the case where \((b_1\ ;P_i'\ ;b_2)\ne \emptyset \). Hence, if either \((a_1\ ;P_i'\ ;a_2)\ne \emptyset \) or \((b_1\ ;P_i'\ ;b_2)\ne \emptyset \), then \(P_i'\) is discovered by Algorithm 1.

*Sub-Case C* Let us now have a closer look at a \(P_i'\) where \((a_1\ ;P_i'\ ;a_2)=(b_1\ ;P_i'\ ;b_2)=\emptyset \).

Denote \(d_1,d_2\) a pair of candidates that satisfy conditions^{Footnote 2} of Lemma 6, and, in addition, that \(P_i(a_1,d_1)\) and \((a_1; P_i; d_1)\) is minimal.

Then the reasoning of Sub-Case B above can be repeated, replacing \(b_2\) by \(d_2\) in its arguments. We conclude that \(P_i'\) with \((a_1; P_i'; a_2)=\emptyset \) and \((b_1;P_i';b_2)=\emptyset \) will also be discovered by Algorithm 1. In other worlds Algorithm 1 will discover all elements of \(\mu (P_i)\). As the algorithm selects a locally dominant order \(P_i'\) among all those that it finds, the final outcome will satisfy both condition Condition-1 and Condition-2. \(\square \)

## Rights and permissions

## About this article

### Cite this article

Dery, L., Obraztsova, S., Rabinovich, Z. *et al.* Lie on the Fly: Strategic Voting in an Iterative Preference Elicitation Process.
*Group Decis Negot* **28, **1077–1107 (2019). https://doi.org/10.1007/s10726-019-09637-2

Published:

Issue Date:

DOI: https://doi.org/10.1007/s10726-019-09637-2

### Keywords

- Iterative voting
- Preference elicitation
- Group decisions
- Crowdsourcing