GAME THEORY is a branch of mathematics used to analyze competitive
situations whose outcomes depend not only on one's own choices, and perhaps chance, but also on the choices made by other parties, or players. Since the outcome of a game is dependent on what all players do, each player tries to anticipate the probable choices of other players in order to determine his own best choice. How these interdependent strategic calculations may reasonably be made is the subject of the theory. odern game theory was created practically at one stro!e by the publication in "#$$ of %heory of &ames and 'conomic (ehavior by the mathematician )ohn von *eumann and the economist +s!ar orgenstern. %his boo! stimulated rapid, worldwide development of the mathematical theory and its applications to economics, politics, military science, operations research, business, law, sports, biology, and other fields. ,t has had a major influence, widening and refining common discourse on strategic thin!ing. %he theory has several major divisions, the following being the most important- .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.444 Introduction branch of applied mathematics fashioned to analyze certain situations in which there is an interplay between parties that may have similar, opposed, or mi5ed interests. ,n a typical game, decision0ma!ing players, who each have their own goals, try to outsmart one another by anticipating each other's decisions6 the game is finally resolved as a conse7uence of the players' decisions. 8 solution to a game prescribes the decisions the players should ma!e and describes the game's appropriate outcome. &ame theory serves as a guide for players and as a tool for predicting the outcome of a game. 8lthough game theory may be used to analyze ordinary parlour games, its range of application is much wider. ,n fact, game theory was originally designed by the Hungarian0born 8merican mathematician )ohn von *eumann and his colleague +s!ar orgenstern, a &erman0born 8merican economist, to solve problems in economics. ,n their boo! %he %heory of &ames and 'conomic (ehavior, published in "#$$, von *eumann and orgenstern asserted that the mathematics developed for the physical sciences, which describes the wor!ings of a disinterested nature, was a poor model for economics. %hey observed that economics is much li!e a game in which the players anticipate one another's moves and that it therefore re7uires a new !ind of mathematics, which they appropriately named game theory. &ame theory may be applied in situations in which decision ma!ers must ta!e into account the reasoning of other decision ma!ers. (y stressing strategic aspects9aspects controlled by the participants rather than by pure chance9the method both supplements and goes beyond the classical theory of probability. ,t has been used, for e5ample, to determine the formation of political coalitions or business conglomerates, the optimum price at which to sell products or services, the power of a voter or a bloc of voters, the selection of a jury, the best site for a manufacturing plant, and even the behaviour of certain species in the struggle for survival.,t would be surprising if any one theory could address such a wide range of games, and, in fact, there is no single game theory. 8 number of theories e5ist, each applicable to a different !ind of situation and each with its own !ind of solution :or solutions;.%his article discusses some of the simpler games and theories as well as the basic principles involved in game theory. 8dditional techni7ues and concepts that may be used in solving decision problems are treated in optimization. <or information pertaining to the classical theory of probability, see the articles mathematics, history of6 and probability theory..opyright / "##$01221 'ncyclop3dia (ritannica, ,nc Classification of games &ames are grouped into several categories according to certain significant features, the most obvious of which is the number of players involved. 8 game can thus be designated as being one0person, two0person, or n0person :with n larger than two;, and the games in each category have their own distinct natures. 8 player need not be a single person, of course6 it may be a nation, a corporation, or a team consisting of many people with identical interests relative to the game. ,n games of perfect information, such as chess, each player !nows everything about the game at all times. =o!er, on the other hand, is an e5ample of a game of imperfect information because players do not !now the cards the other players are dealt. %he e5tent to which the goals of the players are opposed or coincide is another basis for classifying games. >ero0sum :or, more accurately, constant0sum; games are completely competitive. =o!er, for e5ample, is a zero0sum game because the combined wealth of the players remains constant6 if one player wins, another must lose because money is neither created nor destroyed. =layers in zero0sum games have completely conflicting interests. ,n nonzero0sum games, however, all the players can be winners :or losers;. ,n a labour0management dispute, for e5ample, the two parties have some conflicting interests yet both may benefit if a stri!e is avoided. *onzero0sum games can be further distinguished as being either cooperative or noncooperative. ,n cooperative games players may communicate and ma!e binding agreements in advance6 in noncooperative games they may not. 8n automobile salesman and a potential customer are engaged in a cooperative game, as is a business and its employees6 participants bidding independently at an auction are playing a noncooperative game. <inally, a game is said to be finite when each player has a finite number of decisions to ma!e and has only a finite number of alternatives for each decision. .hess, chec!ers, po!er, and most parlour games are finite. ,nfinite games, in which there are either an infinite number of alternatives or an infinite number of decisions, are much subtler and more complicated. %hey are discussed only briefly in this article. 8 game can usually be described in one of three ways- in e5tensive, normal, or characteristic0function form. ost parlour games, which progress step by step, a move at a time, are described in e5tensive form6 that is, they are represented by a tree. 'ach step or position is represented by a verte5 of the tree, and the branches connecting the vertices represent the players' alternatives or moves. %he normal form is used primarily to describe two0person games. ,n this form a game is represented by a matri5 in which each column is associated with a strategy of one player and each row with a strategy of the second player6 the matri5 entry in a particular row and column is the outcome of the game if the two associated strategies are used. %he normal form is important theoretically and can also be used in practice if the number of available strategies is small. %he characteristic0function form, which is generally used only for games with more than two players, indicates the minimum value each coalition of players :including single0player coalitions; can obtain when playing against a coalition made up of all the remaining players. One-person games ,n the one0person game, the simplest game of all, there is only one decision ma!er. (ecause he has no opponents to thwart him, the player need only list the options available to him and then choose one. ,f chance is involved, the game might seem to be more complicated, but in principle the decision is still relatively simple. 8 man deciding whether to carry an umbrella, for e5ample, weighs the ris!s involved and ma!es his choice. He may ma!e the wrong decision, but he need not be worried about being outsmarted by other players6 that is, he need not ta!e into account the decisions of others. +ne0person games, therefore, hold little interest for game theoreticians. .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.4 Two-person zero-sum games Games of perfect information %he simplest game of any real theoretical interest is the finite two0person zero0 sum game of perfect information. '5amples of such games include chess, chec!ers, and the )apanese game go. ,n "#"1 'rnst >ermelo proved that such games are strictly determined6 this means that rational players ma!ing use of all available information can deduce a strategy that is clearly optimal and so the outcome of such games is preordained. ,n chess, for e5ample, e5actly one of three possibilities must be true- :"; white has a winning strategy :one that wins against any strategy of blac!;6 :1; blac! has an analogous winning strategy6 or :?; white and blac! each have a strategy that guarantees them a win or a draw. :=roper play by both white and blac! leads to a draw.; (ecause a sufficiently rapid computer could analyze such games completely, they are of only minor theoretical interest. Games of imperfect information %he simplest two0person zero0sum games of imperfect information are those that have saddle points. :8ll two0person zero0sum games of perfect information have saddle points.; Such games have a predetermined outcome :assuming rational play;, and each player can, by choosing the right strategy, obtain an amount at least e7ual to this outcome no matter what the other player does. %his predetermined outcome is called the value of the game. 8n e5ample of such a game is described in normal form below. %wo campaigning political parties, 8 and (, must each decide how to handle a controversial issue in a certain town. %hey can either support the issue, oppose it, or evade it. 'ach party must ma!e its decision without !nowing what its rival will do. 'very pair of decisions determines the percentage of the vote that each party receives in the town, and each party wants to ma5imize its own percentage of the vote. %he entries in the matri5 represent party 8's percentage :the remaining percentage goes to party (;6 if, for e5ample, 8 supports the issue and ( evades it, 8 gets @2 percent :and (, 12 percent; of the vote.8's decision seems difficult at first because it depends upon ('s strategy. 8 does best to oppose if ( supports, evade if ( opposes, and support if ( evades. 8 must therefore consider ('s decision before ma!ing its own. *o matter what 8 does, ( gains the largest percentage of votes by opposing the issue. +nce 8 recognizes this, its strategy should clearly be to evade and settle for ?2 percent of the vote. %his ?2 percentAB2 percent division of the vote is the game's saddle point. 8 more systematic way of finding the saddle point is to determine the ma5imin and minima5 values. Csing this method, 8 first determines the minimum percentage of votes it can obtain for each of its strategies and then finds the ma5imum of these three minimum values. %he minimum percentages 8 will get if it supports, opposes, or evades are, respectively, 12, 1D, and ?26 the largest of these, ?2, is the ma5imin value. Similarly, for each strategy it chooses, ( determines the ma5imum percentage of votes 8 can win :and thus the minimum that ( can win;. ,n this case if ( supports, opposes, or evades, the ma5imum 8 gets is @2, ?2, or @2, respectively. ( obtains its highest percentage by minimizing 8's ma5imum percentage of the vote. %he smallest of 8's ma5imum values is ?2, and ?2 is therefore ('s minima5 value. (ecause both the minima5 and the ma5imin values are ?2, ?2 is a saddle point. %he two parties might as well announce their strategies in advance6 neither gains from the !nowledge. .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc. Two-person nonzero-sum games ost of the early wor! in game theory involved completely competitive two0 person zero0sum games because they are the simplest to treat mathematically. (oth players in such games have a clear purpose :to outplay their opponent;, and there is general agreement about what constitutes a solution. ost games that arise in practice, however, are nonzero0sum ones6 the players have both common and opposed interests. <or e5ample, a buyer and a seller are engaged in a nonzero0sum game :the buyer wants a low price and the seller a high one, but both want to ma!e a deal;, as are two hostile nations :they may disagree about numerous issues, but both gain if they avoid going to war;.8ll zero0sum games are the same in one respect- the players are acting at cross0purposes. *onzero0 sum games differ radically from them :and from each other;, and many obvious properties of zero0sum games are no longer valid. ,n zero0sum games, for e5ample, players cannot gain :they may or may not lose, but they cannot gain; if they are deprived of some of their strategies. ,n nonzero0sum games, however, players may well gain if some of their options are no longer available. %his might not seem logical at first. +ne would thin! that if it benefited a player not to use certain strategies, the player would simply avoid those strategies and choose a more advantageous one, but this is not always possible. <or e5ample, in a region with high unemployment a wor!er may be willing to accept a low salary rather than lose a job, but, if a minimum wage law ma!es that option illegal, the wor!er may be forced to re7uire a higher salary than he might otherwise have accepted. +ne of the factors that most reveals the difference between zero0sum and nonzero0sum games is the effect of communication on the game. ,n zero0sum games it never helps a player to give an adversary information, and it never harms a player to learn an opponent's strategy in advance. %hese rules do not necessarily hold true for nonzero0sum games, however. 8 player may want his opponent to be well0informed. ,n a labourEmanagement dispute, for e5ample, if the labour union is prepared for a stri!e, it behooves it to inform management and thereby possibly achieve its goal without a long, costly conflict. ,n this e5ample, management is not harmed by the advance information :it, too, benefits by avoiding the costly stri!e;, but in other nonzero0sum games a player can be at a disadvantage if he !nows his opponent's strategy. 8 blac!mailer, for e5ample, benefits only if he informs his victim that he will harm the victim unless his terms are met. ,f he does not give this information to the intended victim, the blac!mailer can still do damage but he has no reason to. %hus, !nowledge of the blac!mailer's strategy wor!s to the victim's disadvantage. Cooperatie ersus noncooperatie games .ommunication is irrelevant in zero0sum games because there is no possibility of the players cooperating :their interests are e5actly opposite;. ,n nonzero0sum games the ability to communicate, the degree of communication, and even the order in which players communicate can have a profound influence on the outcome of the game. &ames in which players are allowed to communicate and ma!e binding agreements are called cooperative, and games in which players are not allowed to communicate are called noncooperative. ,n the e5ample shown below, it is to player ('s advantage if the game is cooperative and to player 8's advantage if the game is noncooperative. :*ote that in nonzero0sum games each matri5 entry consists of two numbers. (ecause the combined wealth of the players is not constant, it is not possible to deduce one player's payoff from the payoff of the other6 both players' payoffs must be given. &enerally, the first number in each entry represents the payoff to the player whose strategies are listed in a column Fhere, player (G, and the second number represents the payoff to the player whose strategies are listed in a row Fhere, player 8G.;Hithout communication, each player should apply the sure0thing principle of ma5imizing his minimum payoff6 that is, each should determine the minimum amount he could e5pect to get no matter what his opponent does. 8 would determine that he would do best to choose strategy , no matter what ( does :if ( chooses i, 8 gets 1 rather than "6 if ( chooses ii, 8 gets 0"22 rather than 0D22;. ( would similarly determine that he does best to choose i no matter what 8 does. Csing these two strategies, 8 would gain 1 and ( would gain ". ,n a cooperative game, however, ( can threaten to play ii unless 8 agrees to play ,,. ,f 8 agrees, his gain is reduced to " while ('s gain rises to ?6 if 8 does not agree and ( carries out his threat, ( neither gains nor loses, but 8 loses "22.+ften, both players may gain from the ability to communicate. %wo pilots trying to avoid a midair collision clearly benefit if they are allowed to communicate, and the degree of communication allowed between them may even determine whether or not they crash. &enerally, the more two players' interests coincide, the more important and advantageous communication becomes. %he solution to a completely cooperative game in which players share one common goal involves coordinating the players' decisions effectively. ,t is relatively straightforward, as is the solution to completely competitive, or zero0 sum, games. <or games in which players have both common and conflicting interests9in other words for most nonzero0sum games, whether cooperative or noncooperative9the solution is much harder to define persuasively ..opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.4 Two-person nonzero-sum games Mi!ed strategies ,n two0person zero0sum games it is often prudent for players to use mi5ed strategies in order to hide their intentions. %he 8merican social scientist Henry Hamburger showed that mi5ed strategies are useful in nonzero0sum games as well. He described the dilemma facing a community in which the number of high0speed automobile accidents was rising and the cost of strict enforcement of speed limits was prohibitively high. He assumed that all of the costs and benefits to drivers and to society associated with a speed limit could be calculated, and he then tried to determine how well the speed limit could be enforced. 8fter considering such factors as the cost of strict enforcement, the ris! of accidents to the driver and their cost to the community, the penalty to the driver when caught, and the time the driver saved by violating speed laws, he constructed the following payoff matri5. ,t can be seen that the cost to the community of strict enforcement is high whatever the driver does. %hus, a driver who tends to speed will deduce that strict enforcement is not li!ely and will violate the law6 the community deduces that it does better not to enforce the law strictly even if it is aware of the motorist's violation. ,f, however, the community rejects both pure strategies :complete enforcement and complete neglect; and announces that it will enforce the law precisely "2 percent of the time9and actually does so9the matri5 is transformed into the one shown below. 8 driver who violates the law now loses "2, which should induce him to obey the law :and brea! even;. %he cost to the community of this partial enforcement is thus only 1, which is lower than the cost of ignoring the law and allowing drivers to violate it. %he community clearly benefits by using the mi5ed strategy. "-person games %heoretically, n0person games in which the players are not allowed to communicate are not essentially different from two0person games. ,f cooperation is allowed, however, there is an opportunity for some players to join forces and act as a single unit, significantly changing the nature of the game. ,n such games the fate of each player is much more dependent upon the decisions of others than in one0 and two0person games6 conse7uently, the concept of a best strategy is even harder to define persuasively. .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.44 8dditional reading%he seminal wor! in game theory is )ohn Ion *eumann and +s!ar orgenstern, %heory of &ames and 'conomic (ehavior, ?rd ed. :"#D?, reprinted "#@2;. &ame theory as a whole is covered in the following boo!s, listed in order of increasing difficulty- 8natol Japoport, <ights, &ames, and Kebates :"#L2, reprinted "#LB;6 orton K. Kavis, &ame %heory- 8 *ontechnical ,ntroduction, rev. ed. :"#@?;6 J. Kuncan Muce and Howard Jaiffa, &ames and Kecisions- ,ntroduction and .ritical Survey :"#DB, reprinted "#LB;6 and &uillermo +wen, &ame %heory, 1nd ed. :"#@1;. 8pplications of game theory are presented in *esmith .. 8n!eny, =o!er Strategy- Hinning with &ame %heory :"#@", reprinted "#@1;6 Jobert 85elrod, %he 'volution of .ooperation :"#@$;, concerned with evolution and ecology6 Steven ). (rams, &ame %heory and =olitics :"#BD;6 and Henry Hamburger, &ames as odels of Social =henomena :"#B#;. Cseful essays and journal articles include Jobert ). 8umann and ichael aschler, %he (argaining Set for .ooperative &ames, in . Kresher, M.S. Shapley, and 8.H. %uc!er :eds.;, 8dvances in &ame %heory :"#L$;, pp. $$?E$BL, the basis of the 8umannEaschler solution concept6 Kaniel Nahneman and 8mos %vers!y, %he =sychology of =references, Scientific 8merican, 1$L:";-"L2E"B? :)anuary "#@1;, a discussion of the validity of the assumptions that underlie utility theory6 M.S. Shapley, 8 Ialue for *0=erson &ames, in H.H. Nuhn and 8.H. %uc!er :eds.;, .ontributions to the %heory of &ames, vol. 1 :"#D?;, pp. ?2BE?"B, the foundation of the Shapley value6 )ohn aynard Smith, %he 'volution of (ehavior, Scientific 8merican, 1?#:?;-"BLE"#1 :September "#B@;6 and =hilip K. Straffin, )r., %he (andwagon .urve, 8merican )ournal of =olitical Science, 1":$;-L#DEB2# :*ovember "#BB;, which e5plains bandwagon effect on the basis of the Shapley value.orton K. Kavis .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.44 *0person games 8pplications Ioting games ,n situations involving voting there is a well0defined procedure for decision ma!ing by a well0defined set of players, and any outcome is dependent upon these decisions. oreover, it often behooves a voter to anticipate the votes of others before he votes himself. ,t is not surprising, therefore, that many of the applications of n0person game theory are concerned with voting.+ne of the most basic problems game theory has tried to answer is how to choose a good voting procedure. ,t might seem that finding a suitable voting system is simple, but in fact there are many difficulties, as can be seen in the following e5ample. 8 voting body has three alternatives9 8, (, and .9from which it must choose one. 8 decision rule pic!s one of these three alternatives on the basis of how the electorate voted. Ioter preference is described by the notation :8(.; O !, which means that ! voters prefer 8 to ( and ( to ..+ne decision rule often used in practice is to choose the plurality favourite9the alternative receiving the largest number of votes. <ollowing this rule, if there are L2 voters, :8(.; O 1", :(.8; O 12, :.(8; O "#, and each voter bac!s his favourite candidate, then 8 would win. (ut almost two0thirds of the electorate consider 8 the worst choice and, in fact, 8 would lose a two0way election against either ( or . by almost 1 to ".%o avoid this parado5, a two0stage election can be instituted in which the least0preferred alternative is eliminated in the first round, and the winner is then chosen from the remaining two alternatives in the second round. ,n this type of election, if :(8.; O :.8(; O 1" and :8(.; O "@, then 8 would be eliminated in the first round and ( would defeat . ?# to 1" in the second round. ,n a two0way election between 8 and (, however, 8 would win by almost two to one. ,t is clear that two0stage elections may also provide parado5ical results. ,n fact, it has been proven that any voting system with at least three alternatives will under certain conditions lead to an outcome that most people would consider inappropriate.+ne of the basic difficulties in devising a satisfactory voting system is that group preferences are what mathematicians call intransitive. %his means that a group may prefer 8 to (, ( to ., and . to 8, as is the case when precisely one0third of the electorate has each of the following preferences- :8(.;, :(.8;, and :.8(;. 8nother difficulty with voting systems is that voters may vote strategically rather than sincerely. ,t may seem that in a plurality voting system the most popular first choice would win, but it often serves a voter's purpose to bac! his second choice. ,f :8(.; O 1", :(8.; O 12, and :.(8; O "#, the "# :.(8; voters do well to switch from . to (6 that way, they get their second, rather than their third, choice. %his is commonly !nown as voting for the lesser evil.%he Shapley value Hhen members of a voting body control different numbers of votes, it is natural to as! how the voting powers of these members compare. &ame theoreticians have therefore devoted much thought to the problem of how to calculate the power of an individual or a coalition. ,t is intuitively clear that the power of a senator differs from that of the president, but it is another thing to assign actual 7uantitative values to these powers. 8t first, it might seem that power is proportional to the number of votes cast, but the following e5ample demonstrates that this cannot be right. ,f 8, (, and . control three, two, and two votes, respectively, and decisions are made by majority vote, then, clearly, everyone has the same power despite 8's e5tra vote.%he 8merican mathematician Mloyd S. Shapley devised a measure of the power of a coalition based upon certain assumptions :e.g., the power of a coalition depends upon the characteristic function only;. ,n voting games, which are sometimes called simple games, every coalition has the value " :if it has enough votes to win; or 2 :if it does not;. %he sum of the powers of all the players is ". ,f a player has 2 power, his vote has no influence on the outcome of the vote6 and if a player has a power of ", the outcome depends on his vote only. %he !ey to calculating voting power is determining the fre7uency with which a player is the swing voter. ,n other words, it is assumed that the members of a body vote for a measure in every possible permutation :or order;. %he fraction of all permutations in which someone is the swing voter9that is, where the measure had insufficient votes to pass before the swing vote but enough to pass after it9is defined to be the Shapley value, or power, of that voter.<or e5ample, in an electorate with five members98, (, ., K, and '9with one, two, three, four, and five votes, respectively, decision is by majority rule. Kespite the disparity in the number of votes each member controls, 8 and ( each have a Shapley value of "A"D, . and K each have a value of BA?2, and ' has a value of 1AD. %his reflects the fact that any winning coalition containing 8 and not containing ( will still win if 8 and ( both change their votes9they each have e5actly the same power. 8 similar statement can be made concerning . and K. %he Shapley value has been used in a number of ways, some of them 7uite surprising. ,t has been used to calculate the relative power of the permanent and nonpermanent members in the Cnited *ations Security .ouncil :the five permanent members have #@ percent of the power;, the power of a C.S. congressman as compared to that of a senator or the president, and the power of members of various city councils. Csing a variation of this concept, )ohn (anzhaf, an 8merican attorney, successfully challenged a weighted system of voting in *assau .ounty, *ew Por!, in which si5 municipalities had, respectively, nine, nine, seven, three, one, and one members on the (oard of Supervisors. (anzhaf proved that the three municipalities with the lowest weights were effectively disenfranchised because two of the top three weights guaranteed "L out of ?2 votes, a majority. %he Shapley value shows that if a voting body has one powerful member with "D votes and ?2 members with one vote each, the power of the strong member is "DA?", which is considerably more than the third of the votes that that member controls. ,n general, one large bloc of votes amidst a sea of small blocs is usually disproportionately strong. :<or this reason, populous states are stronger in the C.S. electoral college than their votes would indicate, even though the fact that every state, whatever the size of its population, has two senators seems to give an edge to the smaller states.; .onversely, when a voting body consists of two large, appro5imately e7ual blocs and one small bloc, it is the small bloc that has the disproportionate power. :,f there are a total of three voters and two have "22 votes and one has one vote, all have the same power if the majority rules.; ,t is because of these two phenomena that political conventions are unstable. Hhen two e7ually powerful factions are vying for power, the uncommitted parties on the sidelines have a great deal of leverage6 but as soon as one of the rival factions shows signs of succeeding, its power increases 7uic!ly at the e5pense of its main competitor and those on the sidelines, leading to the well0!nown bandwagon effect. .uriously, the Shapley value, which was designed to measure the power of a coalition, has also been applied for an entirely different purpose, as is demonstrated in the following e5ample. <our airplanes share a runway, and, because they vary in size, the length of runway each airplane re7uires varies as well. ,f the value of a coalition of airplanes is defined to be the cost of the shortest runway re7uired to serve the airplanes in that coalition, the Shapley value of each airplane represents its e7uitable share of the cost of the runway. %his method of calculating one party's share of total cost has been adopted by accountants in practice and reported upon in accounting journals. Iarious principles and methods of game theory have similarly been applied to fields other than those for which they were originally developed. 8lthough a relatively new area of study, game theory plays a significant role in such diverse subjects as management science, behavioral science, political science, economics, information theory, control theory, and pure mathematics. ,t has proved especially useful in analyzing the effect of information, solving problems of cost or resource allocation, calculating relative political or voting power, determining e7uilibrium points, and e5plaining patterns of observed behaviour. .opyright / "##$01221 'ncyclop3dia (ritannica, ,nc.44