130 Chapter 4. Mixed Strategy Equilibrium
St. Continue
Stop I '--i, 1 -- £ - 1 - E, u
.---•... --_._---,- ,,'--._.0-
I
Continue
__.�=_�_,
____Q�.Q.
_ __' Figure 130.1 he gme n Execse 130.2.equilibrium, as does any smmetric game in which each player has initely many acions, by he following result (he proof of which requires relaively advanced mahemaical tools).
• PROPOSITION
130.1
(Existence of smmeric mixed strategy Nash equilibrium in smmeric nite games)Evy strategic game with vNM prferences in which each
player has the same inite set of actions has a symmetric mixed strategy Nash equilibrium.
!EXERCISE
130.2
(Approaing cars) Members of a single populaion of car drivers are randoly matched in pairs when hey simultaneously approach intersecions rom diferent directions. n each interaction, each driver can eiher stop or con inue. he drivers' preferences are represented by he expected value of he payof funcions given in Figure130.1;
he parameter E, wih Q< E <
I, relects he fact hat each driver dislikes being he only one to stop. Find he symmetric Nash ebrium (equilibria?) of the game (ind boh the equilibrium strategies and he equilibrium payoffs). .Now suppose hat drivers are (re)educated to feel guilty about choosing
Con
tinue,
wih he consequence hat heir payofs when choosingContinue
fall by0
> O. hat is, he enry(2, 1- E )
in Figure130.1
is replaced by(2 - 0, 1 - E ) ,
he enry(1- E,2 )
is replaced by(1- E, 2 - 0),
and he enry(0, 0)
is replaced by(-0, -0).
Show hat all drivers are
better of
in he symmetric equilibrium ofs
game han hey are in he smmeric equilibrium of he original game. Why is he society better of if everyone feels guilty about being aggressive? (he equilibrium ofs
game, lke hat of he equilibrium of he game of epert diagnosis n Secion4.6,
may aracively be nterpreted as represening a steady state in which some mem bers of he populaion always choose one acion and oher members always hoose he oher acion.)
j
EXERCISE130.3
(Bargaing) Pairs of players rom a single populaion bargan over he ivision of a pie of sze10.
he members of a pair smultaneously make demands; he possible demands are he nonegaiveven
integers up to10. f
he demands sum to10,
then each player receives her demand; if he demands sum to lessn 10,
hen each player receives her demand plus half of he pie hat re mains after boh demands have been saisied; if the demands sum to moren
10,
hen neiher player receives any payof. Find all he symmeric mixed srategy Nash equilibria n which each player assigns positive probability to at most two demands. (Many siuaions in which each player assigns positive probability to two actions, saya'
and a", can be ruled out as equilibria because when one player uses such a strategy, some acion yields he oher player a payof higher han does one or boh of he acions a' anda"
.)4.8 illustation: reporting a crime
131 4.8 Illustration: reporting a crime
A crime is observed by a group of
n
people. Each person would like the I' b infomed but prefers hat someone else make the phone call Specii llPO Ice to e h at ea person attaches he value chv
to he police being informed d b . ca y, supposeif h an ears he
cost
c
s e makes he phone call, wherev
>c
>0
hen he situai ' d b h y e followmg srategic game wih vNM preferences. . . on IS mo eledPlayers
hen
people .Actions
Each player's set of actions is{Call, Don't call}.
Preferences
.Ea
�
player's pr�
ferences are represented by he expected value of a payof funcion hat assIgns0
to he proile in wich no one calls _to any proile in which she cals, and
v
to any proile in which at le�
s�
on�
person calls, but she does not.�
.g�
e is a variant of he one in Exercise33.1,
withk
=1.
It hasn
pure Nash eq�bna, m each of which exactly one person calls.(f
hat person swhes to not calling, her payoff falls fromv - c
to0;
if any other person switches to callng, her payofls ro� v
tov
. . c. � f
he members of he group difer in some respect, hen the�
e as�
erlc.eqillbna may be compelling as steady states. For example, the sooal norm m which the oldest person in the group makes the phone call is stable.
f
he. members of he group eiher do not difer signiicanly or are not aware of any dffere
�
ces among he'
elves-if hey are drawn rom a single homoge neo� p?
pul�
ion�hen here 5 no way.for them to coordinate, and a symmetric eqilibrlm, m which every player uses he same srategy, is more compelng.he game has no, symmeric pure Nash equilibrium.
(f
everyone calls, hen any person is better of SWitching to not calling.f
no one calls, then any person is beter of Switcing to calling.)However, it has a symmeric mixed srategy equilibrium in which each person calls wih posiive probability less than one.
n
any such equilibrim, each per son's epected payof to callng is equal to her epected payof to not calling. Each person's payof to calling isv - c,
and her payof to not calling is0
if no one else calls andv
if at least one oher person calls, so he equilibrium condiion isor or
v - c
=0
. Pr{ no one else calls} +v
. Pr{ at least one oher person calls},v - c
=v· (1-
Pr{no one else calls}),c/v
= Pr{no one else calls}.(131.1)
Denote by
p
he probability wih which each person calls. he probabiliy hat no one else calls is he probability hat every one of he ohern - 1
people does not call, namely(1- p)n-l.
Thus he equilibrium condition isc/v
=(1- p)n-l,
orp
=1 - (c/v)l/(n-l).
132 Chapter 4. Mixed Strategy Equilibrium
is number
p
is between0
and I, so we conclude hat he game has a ique smmeric �ed srategy equilibrim, in which each person cals wih probabilityl_'(c/v)l/(n-l).
hat is, here is a steady state in which whenever a person is in a group ofn
people facing he situation modeled by he game, she calls wihprobabiliy
1- (c/v)l/(n-l). .
How does his equilibrim change as he size of he group increases? We see hat as
n
increases, he probabilityp
hat any given person calls decreases. (Asn
increases,l/(n -1)
decreases, so hat(c/v)lI(n-l)
inreases.) What about he probability hatat least
one person cals? Fix any playeri.
hen he event "no one calls" is he same as he event"i
does not call and no oneother than i
calls". husPr{no one calls} = Pr{i does not call}· Pr{no one else calls}.
(132.1)
Now, he probability hat any given person calls decreases as
n
increases, or equiv alently he probabiy hat she does not call increases asn
ncreases. Furher, rom he ebrium condition(131.1),
Pr{no one else calls} is equal toc/v, independent
of n.
We conclude hat he probability hat no one callsincreases
asn
increases. hat is, the larger he group, heless
likely he police ar:: iformed of he crime!The condiion deining a ixed srategy equilibrium is responsible for
s
re sult. For any given person to be indfferent between calling and not callng,s
condiion requires hat he probability hat no one else calls be independent of he size of he group. hus each person's probability of not calling is larger in a larger group, and hence, by he laws of probability relected in(132.1),
he probability hat no one calls is larger in a larger group.he result hat he larger he group, he less likely any given person calls is not surprising. he result hat he larger he group, he less lkely at least one person calls is a more subtle implicaion of he noion of equilibrium. In a larger group no individual is any less concened hat the police should be caled, but in a steady state he behavior of he group drives down he chance hat he police are noied of he rime.
�
EXERCISE132.2
(Reporing a crme when he winesses are heterogeneous) Con sider a variant of he model stuied ins
secion n whichnl
winesses ncur he costcl
to report he crme, andn2
winesses incur he costC2,
where0 < Cl < V,
o
< C2 < v,
andnl
+n2
=n.
Show hatf cl
andC2
are suficiently close, hen he game has a mixed strategy Nash eqibri� in which every winess's sr�tegy assigns posiive probabiliies to boh reporing and not reporting.}
EXERCISE132.3
(Contributing to a public good) Consider an extension of he anal ysis in is secion to he game in Exercise33.1
fork :: 2.
(In is case a player may contribute even houh he good is not provided; he player's payoff ins
case is-c,)
Denote byQn-l,m(P)
he probability hat exactlym
of a group ofn -1
players conribute when each player conributes wih probabilityp.
What condiion must be satisied byQn-l,k-l(P)
in a symeric mixed srategy equilib im (in wich each player conributes wih he same probability)? (When does4.8 illustration: reporting a crime
133 a player'S conribuion make a diference to he outcome?) For he case v : 1
n
=4, k
=2,
andc
=�
,nd he eqibria eplicitly. (You need to use he fact tha;
Q3,1(P)
=3p(1-p)2,
and do a bit of algebra.)REPORTING A CRIME: SOCIAL PSYCHOLOGY AND GAME THEORY
y-eight people winessed he brutal murder of Caherine
("iy")
Genovese over a period of half an hour in New York City in March194.
During his period no one sigicanly responded to her sreams for help; no one even called h�
police. Jonaliss, psychiarists, soCiologists, and ohers subsequently sruggled to nderstand he winesses' inacion. Some ascribed t to apahy engendered by life in a large city: "Indfference to one's neighbor ands
roubles is a conditioned relex of fe in New York as it is in other big ciies" (Rosenhal194, 81-82).
he event paricularly interested social psychologists. It led them to try to un derstand he circmstances under which a bystander would help someone in trou ble. Experiments quickly suggested that, conrary to the popular theory, people even hose living in large ciies-are not in general apathetic to ohers' plights. An epermental subject who is he lone winess of a person n disress is very likely to
y
to help. But as he size of he group of winesses inreases, here is a decline not only in he probability hat any iven one of hem ofers assistance, but also n he probability hat at least one of hem ofers assistance. Soial psychologists hypohesize hat hree factors explain hese experimental indings. First," difu sion of responsibiity": he larger he group, the lower he psychological cost of not helpng. Second, "audience ibition": he larger he group, he greater he ebarrassment sufered by a helper in case he events
out to be one in wih help is inappropriate (because, for example, t is not in fact an emergency). ird,"social inluence": a person infers he appropriateness of helping rom ohers' be havior, so hat in a large group everyone else's lack of intervenion leads any given person to
k
ntervenion is less likely to be appropriate.In terms of he model in Secion 4.8, hese hree factors raise he expected cost and/or reduce he expected beneit of a person's intervening. hey all seem plausible. However, hey are not needed to explain he phenomenon: our game heoreic analysis shows hat even
f
he cost and beneit areindpendent
of group size, a decrease in he probability hat at least one person ntervenes is an implica ion of equlibrim. is game-heoreic analysis has an advantage over he socio psyhological one: it derives he conclusion rom he same principles hat underlie all he oher models studied so far (oigopoly, aucions, voing, and elections, for example), raher han posing special features of he specic environment in wich a group of bystanders may come to he aid of a person in disress.he criical element missing rom he socio-psychological analysiS is he notion of an
equilibrium.
Wheher any given person intervenes depends on he probabiliy she assigns to some oher person's ntervening. n an equilibrium each person134 Chapter 4. Mixed Stategy Equilibrium
must be ndfferent beween intervng nd not ntervng, and as we have seen
is condition leads inexorably to he conclusion at an nrease in group size
reduces he probability hat at least one person intervenes.
4.9 The formation of players' beliefs
n
a Nash eqilibrium, each player chooses a srategy hat maxizes h � r expected
payof, knowing the oher players' strategies. So far we � ave not co : ldered how
players may 'acquire such information. nformally, he Id � a under � ymg he pr � -,
vious analysis is hat he players have leaned each oher s strategIes ro � herr
experience playng he game.
nhe idealized situation to wi � he � al ! S � S cor
responds, for eah player in he game here is a large populaion of m � l � du � s
who may tke he role of hat player; n any play of he game, one pariClpant IS
dran randoly rom each population.
nis situaion, a new ndividual who
joins a populaion hat is in a steady state (i.e. is using a Nash equilibrium srategy
prole) cn learn he oher players' strategies by obsev � g heir acions over �� y
plays of he gme. As long as he nover in players
5. s �n enough, � 1S�g
players' encounters wih neophytes (who may use noneq�bnum str � tegIes) will
be sficiently rare hat heir beliefs about he steady state ll not be disturbed, so
hat a new player'S problem is simply to lean he oher players' acions. .
s
analysis leaves open he quesion of what might happen
fnew players
simltaneously join more hn one population in suicient numbers hat hey have
a siicant chance of facng opponents who are themselves new. n paricular,
can we expect a steady state to be reached when no one has experience playng he
game?
4.9.1
Eliminating dominated actions
n
some games he players may reasonably be expected to choose heir Nash equi
librium acions rom an ntrospecive analysis of he game. At an exreme, each
player'S best acion may be ndependent of he oher players' actions, as in he
Prisoner's Dilemma (Example 14.1).
nsuch a game no player needs to worry about
e oher players' actions.
na less exreme case, some player's best acion may
depend on he oher players' acions, but he acios he oher players ll hoose
may be clear because each of hese players has an acion hat srictly donates
all ohers. For example, n he game in Figure 135.1, player 2's action R sricly
donates L, so hat no mater what player 2 s player 1 ll do, she should
choose R. Consequently, player
I,who can deduce by s argument hat player 2
will choose R, may reason hat she should choose B. hat is, even nexperienced
players may be led to he nique Nash equilibrium ( B, R) in his game.
his line of argument may be extended. For example, in he game n Fig
ure 135.2, player l's acion
Tis stricly dominated, so player 1 may reason hat
4.9 The formation of players' beliefs
L R
T / 1,2 ! 0,3 I
B [§JILIJ
Figure 135.1 A game n which player 2 has a sricly dominant acion and player 1 does not.
L
R
T ! 0,2 I 0,0 I
M [,T[C2--j
B L!!;�J
135
Figure 135.2 A game
,n which player 1 may reason hat she should choose B because player 2 wll reason hat player 1 will not choose T, so that player 2
ll
choose Rplayer 2 wll deduce hat player 1 ll not choose
T.Consequently player 1 may
deduce hat player 2 will choose R, making B a better acion for her than M.
he set of acion proiles hat reman at he end of such a reasong process
contns all Nash eqbia; for many games (nlike hese examples) he set con
tains many oher acion proiles as well.
nfact, in many games no acion pro
les are einated, because no player has a srictly domnated acion. Neverhe
less, n some classes of games he process is powerful; its loical consequences are
explored in Chapter 12.
4.9.2