# A note on monotonic power indices, smaller coalitions, and new members

- First Online:

DOI: 10.1007/s11238-015-9530-z

- Cite this article as:
- Karos, D. Theory Decis (2016) 81: 89. doi:10.1007/s11238-015-9530-z

- 511 Downloads

## Abstract

Brams’ paradox of new members and Shenoy’s paradox of smaller coalitions are, in a sense, equivalent. They are both implied by the monotonicity of a power index: while the first is exhibited on every simple game that is not strong, the latter can be observed on every simple game in which players are not almost symmetric. For the Shapley–Shubik index, this symmetry condition is not only necessary but also sufficient to avoid the paradox of smaller coalitions.

### Keywords

Paradox of smaller coalitions Paradox of new members Monotonic power indices## 1 Introduction

The problem of measuring power in simple games appeared to be solved after the discoveries of the Shapley–Shubik index (Shapley 1953; Shapley and Shubik 1954) and the Banzhaf–Coleman index (Banzhaf 1965; Coleman 1971). Of course, these are not the only power measures (see for instance Fara et al 2014, for more on power measures), but they can be considered as the most important ones. However, both indices, when applied to games with different player sets, have properties that are not intuitive. These issues were first described by Brams (1975) and Brams and Affuso (1976) who showed, using the extension of the EU Council in 1973, that adding new players to a game might increase the power of some old players. Brams (1975) called this discovery the paradox of new members.

Shenoy (1979) derived a similar result when applying power indices not only to the grand coalition but to each of its subcoalitions. Intuitively, players should have more power in smaller coalitions. However, it turned out that in some games there are smaller coalitions in which some players are worse off than in the original larger coalition. In this case, a power index would exhibit the paradox of smaller coalitions.

The latter paradox played a crucial role in the theory of coalition formation in simple games. Shenoy (1979) showed that if a power index does not exhibit the paradox of smaller coalitions on a simple game, then there is a minimal winning coalition which is core stable—that is, which will not be left by any group of players. Dimitrov and Haake (2008) relaxed this condition and showed that it is sufficient that any coalition has at least one subcoalition in which each player is not worse off. This seems like a fair generalization of Shenoy’s result; however, I will show that in the case of monotonic power indices, it is still quite restricting.

The initial work of Brams and Affuso (1976) derived results on how likely the paradox is to appear. I follow a different approach and investigate the connection between the paradox and the monotonicity of a power index. The relevant definitions can be found in Sect. 2; I also show there that the paradoxes of Shenoy and Brams are, in a sense, equivalent. Two versions of the paradox of new members are investigated in Sect. 3. In particular, it is shown that they are exhibited by any monotonic power index on (almost) every simple game that can be extended. The two versions of the paradox of smaller coalitions are under consideration in Sect. 4, where I show that they are exhibited if players are not almost symmetrical. The condition derived there is both sufficient and necessary for the Shapley–Shubik index. I close the article by briefly discussing the consequences of the incompatibility of monotonicity and the paradoxes’ not being exhibited.

## 2 Four paradoxes and an axiom

### 2.1 Preliminaries

Let \(\Omega \) be a universe of players, and let \(\mathcal P_f\left( \Omega \right) \) be the collection of finite subsets of \(\Omega \), called *coalitions*.^{1} Throughout the paper, when I talk about a player set *N*, *N* is finite. A *simple game* is a function \(v:\mathcal P_f\left( \Omega \right) \rightarrow \left\{ 0,1\right\} \) with \(v\left( \emptyset \right) =0\), such that \(v\left( S\right) \le v\left( T\right) \) for all \(S,T\in \mathcal P_f\) with \(S\subseteq T\), and \(v\left( S\right) +v\left( T\right) \le 1\) for any two disjoint coalitions *S*, *T*; the latter property is called *properness*. If \(v(S)=0\) for all coalitions *S*, *v* is called the *zero game*.

A coalition *S* with \(v\left( S\right) =1\) is called *winning*. Let *S* be winning. If \( v\left( S\right) - v\left( S{\setminus }\left\{ i\right\} \right) =1\) then *i* is called *pivotal* (in *S* with respect to *v*). If *S* is such that each \(i\in S\) is pivotal, *S* is called *minimal winning*. The set of minimal winning coalitions is denoted by \(\mathcal M\left( v\right) \). The *carrier*\(c\left( v\right) \) of a simple game *v* is defined as \(c\left( v\right) =\bigcup _{S\in \mathcal M\left( v\right) }S\). Throughout the paper, I consider games with finite carrier. If \(\vert \mathcal M\left( v\right) \vert = 1\), then *v* is called *unanimous*. If \(v\left( S\right) - v\left( S{\setminus }\left\{ i\right\} \right) =1\) for all winning coalitions *S* then *i* is called a *veto player* (with respect to *v*). If \(i\notin c\left( v\right) \), then *i* is called a *null player* with respect to *v*.^{2} For a coalition *S*, the *subgame*\(v_S\) is defined as \(v_S\left( T\right) =v\left( S\cap T\right) \) for all coalitions *T*. In particular, \(c\left( v_S\right) \subseteq S\).

A *power index*\(\varphi \) maps a simple game *v* and player set *N* with \(c(v)\subseteq N\) into \(\mathbb R^N\). A power index \(\varphi \) satisfies the *null player axiom* if \(\varphi _i\left( v,N\right) =0\) for all \(i\notin c(v)\). I will use the null player axiom only for a few remarks, it will not be necessary for the main results. In particular, I will formulate all paradoxes and theorems for arbitrary player sets that include the carrier of the respective game.

### 2.2 New members and smaller coalitions

*v*be a simple game. An

*extension*of

*v*is a simple game

*u*such that \(\mathcal M(v)\subseteq \mathcal M(u)\) and \(c(v)\subsetneq c(u)\). A simple game

*v*is

*extendable*if there is an extension

*u*of

*v*. Expecting that the addition of new players to a game would not make old players more powerful, Brams (1975) formulated the following paradox.

*Strong paradox of new members*A power index \(\varphi \) exhibits the strong paradox of new members on a simple game

*v*if there is an extension*u*of*v*and a player \(i\in c(v)\) such that \(\varphi _i\left( u,N\right) >\varphi _i\left( v,N\right) \) for all player sets*N*with \(c(u)\subseteq N\).

*weak extension*of a simple game

*v*is a simple game

*u*such that \(c(v)\subsetneq c(u)\) and \(v(S)\ge u_{c(v)}(S)\) for all coalitions

*S*. This means that instead of only adding new winning coalitions, old winning coalitions might be removed. Clearly, every non-trivial extension is a weak extension with \(v(S)= u_{c(v)}(S)\).

*Weak paradox of new members*A power index \(\varphi \) exhibits the weak paradox of new members on a simple game

*v*if there is a weak extension*u*of*v*and a player \(i\in c(v)\) such that \(\varphi _i\left( u,N\right) >\varphi _i\left( v,N\right) \) for all player sets*N*with \(c(u)\subseteq N\).

Votes and quotas in the EU Council before the Treaty of Nice

Country | 1957\(^{\mathrm{a}}\) | 1973\(^{\mathrm{b}}\) | 1979\(^{\mathrm{c}}\) | 1985\(^{\mathrm{d}}\) | 1995\(^{\mathrm{e}}\) |
---|---|---|---|---|---|

France | 4 | 10 | 10 | 10 | 10 |

Germany | 4 | 10 | 10 | 10 | 10 |

Italy | 4 | 10 | 10 | 10 | 10 |

UK | – | 10 | 10 | 10 | 10 |

Spain | – | – | – | 8 | 8 |

Belgium | 2 | 5 | 5 | 5 | 5 |

The Netherlands | 2 | 5 | 5 | 5 | 5 |

Greece | – | – | 5 | 5 | 5 |

Portugal | – | – | – | 5 | 5 |

Sweden | – | – | – | – | 4 |

Austria | – | – | – | – | 4 |

Denmark | – | 3 | 3 | 3 | 3 |

Ireland | – | 3 | 3 | 3 | 3 |

Finland | – | – | – | – | 3 |

Luxembourg | 1 | 2 | 2 | 2 | 2 |

Total | 17 | 58 | 63 | 76 | 87 |

Quota | 12 | 41 | 45 | 54 | 62 |

*Weak paradox of smaller coalitions*A power index \(\varphi \) exhibits the weak paradox of smaller coalitions on a simple game

*v*if there are winning^{3}coalitions \(S,T\subseteq c(v)\), \(T\subseteq S\), and \(i\in c\left( v_T\right) \) such that \(\varphi _i\left( v_S,N\right) >\varphi _i\left( v_T,N\right) \) for all player sets*N*with \(c(v_S)\subseteq N\).

*Strong paradox of smaller coalitions*A power index \(\varphi \) exhibits the strong paradox of smaller coalitions on a simple game

*v*if there is a winning coalition \(S\subseteq c(v)\) such that for all \(T\subseteq S\) there is \(i\in c\left( v_T\right) \) with \(\varphi _i\left( v_S,N\right) >\varphi _i\left( v_T,N\right) \) for all player sets*N*with \(c(v_S)\subseteq N\).

Hence, if each (non-minimal) winning coalition has at least one winning sub-coalition where each member gets at least the same payoff, a power index does not exhibit this stronger paradox. The following easy observation closes this subsection.

**Proposition 1**

A power index \(\varphi \) exhibits the strong paradox of new members on a simple game *v* if and only if there is a simple game *u* such that \(\varphi \) exhibits the weak paradox of smaller coalitions on *u*.

*Proof*

Let \(\varphi \) exhibit the weak paradox of smaller coalitions on the simple game *u*. Then, there are winning coalitions \(S,T\subseteq c(u)\) and a player \(i\in c\left( u_T\right) \) such that \(T\subseteq S\) and \(\varphi _i\left( u_S,N\right) >\varphi _i\left( u_T,N\right) \) for all player sets *N* with \(c(u_S)\subseteq N\). As \(u_S\) is an extension of \(u_T\), the strong paradox of new members follows.

On the other hand, if \(\varphi \) exhibits the strong paradox of new members on *v*, there are an extension *u* of *v* and a player \(i\in c(v)\) such that \(\varphi _i\left( u,N\right) >\varphi _i\left( v,N\right) \) for all player sets *N* with \(c(u)\subseteq N\). As \(v=u_{c\left( v\right) }\), the weak paradox of smaller coalitions follows. \(\square \)

### 2.3 Monotonic power indices

^{4}

*Monotonicity*- A power index \(\varphi \) is called
*monotonic*^{5}if for all simple games*u*,*v*and all*N*where \(c(u)\cup c(v)\subseteq N\) it holds that \(\varphi _i\left( v,N\right) > \varphi _i\left( u,N\right) \) for all \(i\in N\) such that$$\begin{aligned}&v\left( S\right) - v\left( S{\setminus }\left\{ i\right\} \right) \ge u\left( S\right) - u\left( S{\setminus }\left\{ i\right\} \right) \quad \text {for all } S\subseteq N,\\ \text {and}\quad&v\left( S\right) - v\left( S{\setminus }\left\{ i\right\} \right) > u\left( S\right) - u\left( S{\setminus }\left\{ i\right\} \right) \quad \text {for some } S\subseteq N. \end{aligned}$$

It can easily be seen that if *i* is a null player in *u* but not in *v* then monotonicity implies that *i* has strictly greater power in *v* than in *u*; in particular, if the null player axiom is, additionally, satisfied then *i*’s power in *v* is strictly positive.

## 3 The paradox of new members

Before any results on the exhibition of the strong paradox of new players are derived, I shall answer the question: what games are extendable at all? To formulate the next lemma, I need the following definition. A simple game *v* is called *strong* if \(v\left( S\right) +v\left( c(v){\setminus } S\right) =1\) for all \(S\subseteq c(v)\). That means, it is not only required that the complement of a winning coalition [in *c*(*v*)] is losing, but also that the complement of a losing coalition [in *c*(*v*)] is winning.

**Lemma 1**

A simple game *v* is extendable if and only if it is not strong.

*Proof*

Let *v* be a strong game. Then, there are exactly \(2^{\vert c(v)\vert -1}\) winning coalitions contained in *c*(*v*). Suppose there is an extension *u* of *v*. Then, *v* has exactly \(2^{\vert c(u)\vert -1}\) winning coalitions that are contained in *c*(*u*). As *u* is an extension of *v*, it must have at least \(2^{\vert c(u)\vert -1}+1\) winning coalitions in *c*(*u*). Hence, *u* cannot be proper, i.e., it cannot be an extension of *v*.

On the other hand, let *v* not be strong. Then, there is a nonempty coalition *S* such that \(v\left( S\right) =v\left( c(v){\setminus } S\right) =0\). Let \(k\notin c(v)\) and *u* be the (unique) simple game with minimal winning coalitions \(\mathcal M\left( v\right) \cup \left\{ S\cup \left\{ k\right\} \right\} \). Then, *u* is a non-trivial extension of *v*. \(\square \)

The good news is that extendable games can easily be identified. From a theoretical perspective, these games can now easily be handled and much is already known about them. The following theorem, however, shows that if a simple game is extendable then any monotonic power index exhibits the strong paradox of new members on it.

**Theorem 1**

Let \(\varphi \) be a monotonic power index. Then, \(\varphi \) does not exhibit the strong paradox of new members on a simple game *v* if and only if *v* is strong.

*Proof*

*v*is strong, it is not extendable and thus \(\varphi \) cannot exhibit the paradox. So, let

*v*not be strong. Then, there is a nonempty coalition \(S\subseteq c(v)\) such that \(v\left( S\right) =v\left( c(v){\setminus } S\right) =0\). Let \(k\notin c(v)\) and let

*u*be the simple game with minimal winning coalitions \(\mathcal M\left( v\right) \cup \left\{ S\cup \left\{ k\right\} \right\} \). Obviously, \(c(u)=c(v)\cup \left\{ k\right\} \). Let

*N*be a player set containing

*c*(

*u*), let \(i\in S\), and let \(T\subseteq N\). Suppose that

*T*is such that

*i*is pivotal in

*T*with respect to

*v*. As \(k\notin c(v)\), this means that

*i*is also pivotal in \(T{\setminus }\left\{ k\right\} \) with respect to

*v*. In particular, \(u\left( T\right) \ge u\left( T{\setminus }\left\{ k\right\} \right) = v\left( T{\setminus }\left\{ k\right\} \right) =1\). Further, the only minimal winning coalition with respect to

*u*that contains

*k*is \(S\cup \left\{ k\right\} \), and therefore it contains

*i*as well. This means that

*k*cannot be pivotal (with respect to

*u*) in any coalition that does not contain

*i*. Hence, \(u\left( T{\setminus }\left\{ i\right\} \right) =u\left( T{\setminus }\left\{ i,k\right\} \right) =v\left( T{\setminus }\left\{ i,k\right\} \right) =0\). Therefore,

*i*is pivotal with respect to

*v*. If

*T*is such that

*i*is not pivotal in

*T*with respect to

*v*, then \(v\left( T\right) - v\left( T{\setminus }\left\{ i\right\} \right) =0\) and Eq. (1) holds as well. Further,

The following corollary is an easy consequence of the foregoing proof. It states that for every simple game *v* that is not strong and for every player, whether or not he is a null player with respect to *v*, an extension of *v* can be found that makes this player better off.

**Corollary 1**

Let \(\varphi \) be a monotonic power index and let *v* be a simple game that is not strong. Then, for each \(i\in \Omega \), there is an extension *u* of *v* such that \(\varphi _i\left( u,N\right) >\varphi _i\left( v,N\right) \) for each *N* with \(c(u)\cup \left\{ i\right\} \subseteq N\).

*Proof*

Let *v* be a simple game that is not strong. If \(i\in c(v)\), there is \(S\subseteq c(v)\) such that \(v(S)=v(c(v){\setminus } S)=0\) and \(i\in S\). Using the construction in the proof of Theorem 1, an extension *u* of *v* can be obtained with \(\varphi _i(u,N)>\varphi _i(v,N)\) for all *N* with \(c(v)\subseteq N\).

If \(i\notin c(v)\), there is \(S\subseteq c(v)\) such that \(v(S)=v(c(v){\setminus } S)=0\). Let *u* be the simple game with minimal winning coalitions \(\lbrace S\cup \lbrace i\rbrace \rbrace \cup \mathcal M(v)\). Then, by monotonicity, \(\varphi _i(u,N)>\varphi _i(v,N)\) for each *N* with \(c(u)\cup \left\{ i\right\} \subseteq N\), as *i* is not pivotal in any coalition with respect to *v* but is pivotal in \(S\cup \lbrace i\rbrace \) with respect to *u*.

The remainder of this section will deal with weak extensions. Note that every game can be weakly extended; I have already mentioned how it can be done for weighted voting games in Sect. 2. And, again, monotonicity will imply the paradox on almost all games. A *dictatorship* is a simple game *v* with \(c(v)=\lbrace i\rbrace \) for a player \(i\in \Omega \). In such a game, a decision can be made by *i* only, independently of the player set that *v* is applied to. Note that if there is a veto player in a strong game *v* then *v* is a dictatorship.

**Theorem 2**

Let \(\varphi \) be a monotonic power index. Then, \(\varphi \) exhibits the weak paradox of new members on all simple games that are not dictatorships.

*Proof*

*v*be a game that is not a dictatorship and let \(N'\) be a player set with \(c(v)\subseteq N'\). I first show that there exist a game \(v'\) and a player \(i\in c(v')\) such that \(v'\) is not strong, \(v'(S)\le v(S)\) for all coalitions

*S*, and \(\varphi _i(v,N')\le \varphi _i(v',N')\). If

*v*is not strong, choose \(v'=v\). If

*v*is strong, there is \(i\in c(v)\) that is not contained in all minimal winning coalitions (otherwise,

*v*would be a dictatorship). Let \(S\in \mathcal M(v)\) such that \(i\notin S\) and let \(v'\) be the simple game that is defined byThen, \(v'(T)-v'(T{\setminus }\lbrace i\rbrace )\ge v(T)-v(T{\setminus }\lbrace i\rbrace )\) for all \(T\subseteq N'\) and

By Corollary 1, there is an extension *u* of \(v'\) such that \(\varphi _i(u,N)>\varphi _i(v',N)\ge \varphi _i(v,N)\) for all *N* with \(c(u)\subseteq N\). Since, in particular, *u* is a weak extension of *v*, the claim is proved. \(\square \)

The following corollary is now clear and the proof is omitted.

**Corollary 2**

Let \(\varphi \) be a monotonic power index and let *v* be a simple game which is not a dictatorship. Then, for each \(i\in \Omega \), there is a weak extension *u* of *v* such that \(\varphi _i\left( u,N\right) >\varphi _i\left( v,N\right) \) for each *N* with \(c(u)\subseteq N\).

## 4 The paradox of smaller coalitions

The strong paradox of new members and the weak paradox of smaller coalitions are equivalent in the sense that if one is exhibited by a power index on an arbitrary simple game then there is another game on which the other one is exhibited. Nonetheless, they are not exhibited on the same set of games. In the next subsection, a condition on simple games is derived that is necessary for the non-exhibition of the strong paradox of smaller coalitions by a monotonic power index. In Sect. 4.2, I will show that this condition is also sufficient in case of the Shapley–Shubik index.

### 4.1 Avoiding the paradox of smaller coalitions

The key problem of monotonic power indices is that veto players have more power in larger coalitions than in smaller ones. The next lemma illustrates this observation. Recall that \(\mathcal M(v)\) is the collection of minimal winning coalitions with respect to *v*. For a player *i*, let \(\mathcal M_i(v)\) be the collection of those minimal winning coalitions that contain *i*. Clearly, \(\mathcal M_i(v)\ne \emptyset \) if and only if \(i\in c(v)\).

**Lemma 2**

Let \(\varphi \) be a monotonic power index and let *v* be a simple game such that \(\varphi \) does not exhibit the strong paradox of smaller coalitions on *v*. Then, for each winning coalition *S*, either \(v_S\) is an unanimous game or there is no veto player with respect to \(v_S\).

*Proof*

Let *S* be a winning coalition with players \(i,j\in S\) such that *i* is pivotal in *S* and *j* is neither null player nor pivotal. Let *N* be a player set with \(c\left( v_S\right) \subseteq N\) and let \(T\subseteq N\) be such that \(v_{S{\setminus }\left\{ j\right\} }(T) - v_{S{\setminus }\left\{ j\right\} }(T{\setminus }\lbrace i\rbrace )=1\). Then, \(v_S(T)=1\) and \(v_S(T{\setminus }\lbrace i\rbrace )\le v(S{\setminus }\lbrace i\rbrace )=0\). On the other hand, let \(T\subseteq N\) such that \(T\in \mathcal M_j\left( v_S\right) \). (Such a *T* exists as *j* is not a null player in *S*.) Then, *i* is pivotal in *T* with respect to \(v_S\) but \(v_{S{\setminus }\left\{ j\right\} }\left( T\right) =v_S\left( T{\setminus }\lbrace j\rbrace \right) =0\). Hence, \(\varphi _i\left( v_S,N\right) >\varphi _i\left( v_{S{\setminus }\lbrace j\rbrace },N\right) \) by strong monotonicity—that is, \(\varphi \) exhibits the strong paradox of smaller coalitions on *v*. \(\square \)

Lemma 2 reveals that a monotonic power index will exhibit the paradox on a number of simple games. This has been elaborated for instance in Karos (2014). The next theorem characterizes the class of simple games that have the property described in Lemma 2.

**Theorem 3**

Let *v* be a simple game. Then, *v* satisfies the necessary condition of Lemma 2 if and only if \(\vert \mathcal M_j\left( v_S\right) \vert =\vert \mathcal M_k\left( v_S\right) \vert \) for all winning coalitions \(S\subseteq c(v)\) and all \(j,k\in c\left( v_S\right) \).

*Proof*

First, let \(\vert \mathcal M_i\left( v_S\right) \vert =\vert \mathcal M_j\left( v_S\right) \vert \) for all winning coalitions *S* and all \(i,j\in c\left( v_S\right) \). Let *T* be a winning coalition. If \(\mathcal M_i\left( v_T\right) =\mathcal M\left( v_T\right) \) for all \(i\in c\left( v_T\right) \), then \(v_T\) is unanimous. If \(\vert \mathcal M_i\left( v_T\right) \vert <\vert \mathcal M\left( v_T\right) \vert \) for all \(i\in c\left( v_T\right) \), then there are no veto players with respect to \(v_T\). Hence, *v* satisfies the condition in Lemma 2.

Now, let *v* be a simple game such that for each winning coalition *S*, the game \(v_S\) is either unanimous or does not contain any veto players. The claim is proved by induction on the cardinality of *c*(*v*), *n*. For \(n=1\) and \(n=2\), it is obviously true. So, let *v* be such that \(n=3\). Then, *v* is either the three player unanimity game or the symmetric game in which each coalition of at least two players is winning. In both cases, the theorem is true.

*v*is the unanimous game, the claim is clearly true. So let

*v*not be unanimous. In this case, there cannot be any veto players with respect to

*v*by our initial condition. In particular, \(v_{c(v){\setminus }\left\{ i\right\} }\) is not the zero game for any \(i\in c\left( v\right) \). For a player \(i\in c(v)\) let

*l*that “help”

*i*in the sense that after one of them is removed from the game,

*i*becomes a null player. The set \(E_i\) contains those players

*l*on whose presence

*i*does not depend: If

*l*is removed,

*i*still has some (maybe less) power. Note that

*c*(

*v*) can be partitioned into \(D_i\), \(E_i\) and \(\lbrace i\rbrace \). In particular, \(E_i\ne \emptyset \) as otherwise \(c(v)\in \mathcal {M}_i(v)\) in contradiction to

*v*’s not being unanimous. I first show that \(\vert \mathcal M_j\left( v\right) \vert =\vert \mathcal M_k\left( v\right) \vert \) for all \(j,k\in c(v)\). Consider two cases.

- 1.
Let \(k\in D_j\). In this case \(\mathcal M_j(v)\subseteq \mathcal M_k(v)\) or, equivalently, \(\mathcal M(v_{c(v){\setminus }\lbrace k\rbrace })\subseteq \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace })\). I will show that these two sets are equal. For this purpose, note the following observation. Let \(i\in E_j\) and \(l\in D_j\). Then, \(\mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\subseteq \mathcal M_l(v_{c(v){\setminus }\lbrace i\rbrace })\) and the induction hypothesis delivers that \(\vert \mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\vert = \vert \mathcal M_l(v_{c(v){\setminus }\lbrace i\rbrace })\vert \). Hence, \(\mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })=\mathcal M_l(v_{c(v){\setminus }\lbrace i\rbrace })\) for all \(i\in E_j\) and all \(l\in D_j\). Assume now that \(\mathcal M_j(v)\subsetneq \mathcal M_k(v)\)—that is, there is a coalition \(S\in \mathcal M_k(v_{c(v){\setminus }\lbrace j\rbrace })\). Then, \(E_j\subseteq S\). For, if there were \(i\in E_j{\setminus } S\), then \(S\in \mathcal M_k(v_{c(v){\setminus }\lbrace i\rbrace })=\mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\) (as seen before), which would mean that

*S*were a minimal winning coalition containing*j*, in contradiction to \(S\in \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace })\). Recall that \(E_j\ne \emptyset \), let \(i\in E_j\), and let \(T\in \mathcal M(v_{c(v){\setminus }\lbrace i\rbrace })\) (such a*T*exists, as \(v_{c(v){\setminus }\lbrace i\rbrace }\) is not the zero game). Suppose \(j\notin T\). Then, \(T\notin \mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })=\mathcal M_l(v_{c(v){\setminus }\lbrace i\rbrace })\) for any \(l\in D_j\). In particular, \(T\subseteq c(v){\setminus }( D_j\cup \lbrace i,j\rbrace )\subseteq E_j{\setminus }\lbrace i\rbrace \subsetneq S\), contradicting the minimality of*S*. Hence, \(j\in T\) for all \(T\in \mathcal M(v_{c(v){\setminus }\lbrace i\rbrace })\). This means that \(v_{c(v){\setminus }\lbrace i,j\rbrace }\) is the zero game and therefore,*i*is pivotal in \(v_{c(v){\setminus }\lbrace j\rbrace }\). By the initial condition, \(v_{c(v){\setminus }\lbrace j\rbrace }\) must be unanimous, and as \(S\in \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace })\) it must hold that \(\lbrace S\rbrace = \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace }) \supseteq \mathcal M(v_{c(v){\setminus }\lbrace k\rbrace })\). Hence, as \(v_{c(v){\setminus }\lbrace k\rbrace }\) is not the zero game, \(\lbrace S\rbrace =\mathcal M(v_{c(v){\setminus }\lbrace k\rbrace })\), in contradiction to \(S\in \mathcal M_k(v)\). Hence, there is no \(S\in \mathcal M_k(v_{c(v){\setminus }\lbrace j\rbrace })\) and therefore \(\mathcal M_k(v)=\mathcal M_j(v)\). - 2.Let \(k\in E_j\). I first show that there is \(i\in N\) such that \(\vert \mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\vert = \vert \mathcal M_k(v_{c(v){\setminus }\lbrace i\rbrace })\vert \). By the induction hypothesis, it is sufficient to show that there is \(i\in N\) such that either both
*j*and*k*are null players in \(v_{c(v){\setminus }\lbrace i\rbrace }\) or neither of them is. Equivalently, either \(i\in D_j\cap D_k\) or \(i\notin D_j\cup D_k\). Suppose that such*i*does not exist. In this case, \(c(v)=D_j\cup D_k\) and \(D_j\cap D_k=\emptyset \). From the first part of the proof, it follows that \(l\in D_j\) if and only if \(j\in D_l\). In particular, each \(l\in D_j\) is a null player in \(v_{c(v){\setminus }\lbrace j\rbrace }\). Hence, \(v(c(v){\setminus } (D_j\cup \lbrace j\rbrace ))=v(c(v){\setminus }\lbrace j\rbrace ) = 1\). But this means that \(v(D_k)=v(c(v){\setminus } D_j)=1\) and for the same reasons \(v(D_j)=1\). Hence, as \(D_j\) and \(D_k\) are disjoint,*v*is not proper—a contradiction. Therefore, there must be \(i\in N\) such that \(\vert \mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\vert = \vert \mathcal M_k(v_{c(v){\setminus }\lbrace i\rbrace })\vert \). For such*i*it holds thatConsequently,$$\begin{aligned}\nonumber \vert \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace })\vert= & {} \vert \mathcal M(v_{c(v){\setminus }\lbrace i,j\rbrace })\vert + \vert \mathcal M(v_{c(v){\setminus }\lbrace k,j\rbrace })\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace i,k,j\rbrace })\vert \\= & {} \vert \mathcal M(v_{c(v){\setminus }\lbrace i\rbrace })\vert - \vert \mathcal M_j(v_{c(v){\setminus }\lbrace i\rbrace })\vert + \vert \mathcal M(v_{c(v){\setminus }\lbrace k,j\rbrace })\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace i,k,j\rbrace })\vert \\= & {} \vert \mathcal M(v_{c(v){\setminus }\lbrace i\rbrace })\vert - \vert \mathcal M_k(v_{c(v){\setminus }\lbrace i\rbrace })\vert + \vert \mathcal M(v_{c(v){\setminus }\lbrace k,j\rbrace })\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace i,k,j\rbrace })\vert \\= & {} \vert \mathcal M(v_{c(v){\setminus }\lbrace i,k\rbrace })\vert + \vert \mathcal M(v_{c(v){\setminus }\lbrace j,k\rbrace })\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace i,j,k\rbrace })\vert \\= & {} \vert \mathcal M(v_{c(v){\setminus }\lbrace k\rbrace })\vert . \end{aligned}$$$$\begin{aligned} \vert \mathcal M_j(v)\vert= & {} \vert \mathcal M(v)\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace j\rbrace }) \vert \\= & {} \vert \mathcal M(v)\vert - \vert \mathcal M(v_{c(v){\setminus }\lbrace k\rbrace })\vert \\= & {} \vert \mathcal M_k(v)\vert . \end{aligned}$$

The condition on a simple game given in Theorem 3, namely that in every subgame every non-null player is contained in the same number of minimal winning coalitions, is quite strong; it is satisfied, for instance, by simple games in which all players are symmetric. Nevertheless, the latter is not a necessary condition as can be seen in the simple game with minimal winning coalitions \(\lbrace 1,2,3\rbrace ,\lbrace 3,4,5\rbrace \), and \(\lbrace 1,2,4,5\rbrace \).

### 4.2 The Shapley–Shubik index

*N*, I will therefore simply write \(\sigma (v)\) for \(\sigma (v,N)\). Also, \(\sigma \) is monotonic (Sagonti 1991) and therefore exhibits the strong paradox of smaller coalitions on all games not satisfying the condition in Theorem 3. But this condition is, in fact, both necessary and sufficient.

**Corollary 3**

Let *v* be a simple game. Then, the Shapley–Shubik index does not exhibit the strong paradox of smaller coalitions on *v* if and only if \(\vert \mathcal M_i(v_S)\vert =\vert \mathcal M_j(v_S)\vert \) for all winning coalitions *S* and all \(i,j\in c(v_S)\).

*Proof*

*S*and all \(i,j\in c(v_S)\). To show that this is sufficient, I show that \(\sigma _j(v)\le \max _{T\in \mathcal M_j}\frac{1}{\vert T\vert }\) for any such

*v*. In this case, for any coalition

*S*, a minimal winning coalition \(T\subseteq S\) of minimal cardinality delivers higher power for each player in

*T*than

*S*. The claim is shown by induction over \(n=\vert c(v)\vert \). It is clearly true if

*v*is unanimous, in particular, it is true if \(n\le 2\). Hence, let \(n\ge 3\) and let

*v*not be unanimous. Then, \(v_{c(v){\setminus }\lbrace k\rbrace }\) is not the zero game for any

*k*. Let

*T*be a minimal winning coalition of minimal cardinality and let \(j\in T\). Then,

## 5 Conclusion

I have shown that some natural properties one would expect from power indices contradict each other on many simple games. Nevertheless, these two properties are reasonable in different contexts: while monotonicity seems very natural for games with the same carrier, non-exhibition of the paradox of new players (or smaller coalitions) usually concerns games with different carriers.

There are authors arguing that the paradox of new members is not paradoxical (Felsenthal and Machover 1995) on the grounds that the power of a player is *a global property that involves the whole structure* of the game. Although this is clearly true, it makes the (equivalent) paradox of smaller coalitions even more peculiar: a player has less power in a smaller coalition despite the fact that this coalition is part of the same game and the only difference is that (other) players are left out.

It therefore seems that monotonic power indices (including the Shapley index and the Banzhaf index) are not the right instrument to measure power in a council or parliament where the formation of coalitions is likely. It rather makes a case for power measures that take into account players’ options both within and outside coalitions, such as the bargaining set (Davis and Maschler 1963), or more recently a class of solutions investigated in Karos (2013) where players’ power is based on bargaining solutions.

The reason why I introduce a universe of players instead of working on a fixed player set *N* lies in the nature of the problem: I am interested in the effects on power indices if the player set is changed.

Equivalently, *i* is a null player if \(v(S)=v\left( S{\setminus }\left\{ i\right\} \right) \) for each coalition *S*.

Kóczy (2009) shows that the null player axiom, together with a normalization, implies the strong paradox of new players for all simple games. This is not true here as null players are excluded from the set of players that can profit from an extension. In that sense, the paradox here is stronger than the one in Kóczy (2009).

## Acknowledgments

I would like to thank Steven Brams for his comments, as well as the referees and the editor in charge for their insightful suggestions.

## Copyright information

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.