scieee Science in your language
[en] (orig)

Candidate quality in a Downsian Model with a continuous policy space

Abstract

This paper characterizes a mixed strategy Nash equilibrium in a one-dimensional Downsian model of two-candidate elections with a continuous policy space, where candidates are office motivated and one candidate enjoys a non-policy advantage over the other candidate. We assume that voters have quadratic preferences over policies and that their ideal points are drawn from a uniform distribution over the unit interval. In our equilibrium the advantaged candidate chooses the expected median voter with probability one and the disadvantaged candidate uses a mixed strategy that is symmetric around it. We show that this equilibrium exists if the number of voters is large enough relative to the size of the advantage.

Read accessible full text

Candidate quality in a Downsian Model with a continuous policy space

Author: Aragonès, Enriqueta; Xefteris, Dimitrios
Publisher: Dipòsit Digital de Documents de la UAB
Year: 2011
Source: https://ddd.uab.cat/pub/worpap/2011/hdl_2072_152034/85911.pdf
Candida e quali y in a Downsian Model wi h a
Con inuous Policy Space.1
En ique a A agonès2
Ins i u d’Anàlisi Econòmica, CSIC
Dimi ios Xe e is3
Uni e si y o Cyp us
Janua y 10, 2011
1A agones acknowledges inancial suppo by he Gene ali a de Ca alunya G an numbe 2009 SGR 1126, he Spanish
Minis y o Educa ion g an numbe ECO2009-08820 and he Ba celona G adua e School o Economics.
2Ins i u d’Anàlisi Econòmica, CSIC, Campus UAB, 08193 Bella e a (Spain). Email: [email p o ec ed]
3Depa men o Economics, Facul y o Economics and Managemen , Uni e si y o Cyp us. P.O. Box 20537, CY-1678
Nicosia (Cyp us). Email: xe e is.dimi [email protected]
Abs ac
This pape cha ac e izes a mixed s a egy Nash equilib ium in a one-dimensional Downsian model o wo-candida e
elec ions wi h a con inuous policy space, whe e candida es a e office mo i a ed and one candida e enjoys a non-
policy ad an age o e he o he candida e. We assume ha o e s ha e quad a ic p e e ences o e policies and ha
hei ideal poin s a e d awn om a uni o m dis ibu ion o e he uni in e al. In ou equilib ium he ad an aged
candida e chooses he expec ed median o e wi h p obabili y one and he disad an aged candida e uses a mixed
s a egy ha is symme ic a ound i . We show ha his equilib ium exis s i he numbe o o e s is la ge enough
ela i e o he size o he ad an age.
Key wo ds: spa ial compe i ion; mixed s a egies; candida e quali y
1In oduc ion
Candida e quali y is conside ed o be a c i ical a iable in elec o al compe i ion. I affec s he
decisions o poli icians ega ding whe he o un o office, campaign und- aising, o e beha io ,
elec ion ou comes, and, ul ima ely, policy ou comes. Quali y diffe ences be ween wo candida es
can a ise o many easons, including cha isma, office-holding expe ience, incumbency, ad e ising,
scandal, and any o he non-policy dimension ha may affec he ela i e a ac i eness o wo
candida es. In Poli ical Science candida e quali y is also deno ed by “ alence dimension” and i s
impo ance has been widely demons a ed o e se e al decades o ca e ul empi ical esea ch.1
All else cons an , high quali y candida es will a e be e han low quali y candida es. Fu he -
mo e, quali y diffe ences p oduce signi ican changes in he na u e o poli ical compe i ion. The
equilib ium p ope ies o spa ial compe i ion be ween wo candida es who diffe in quali y ha e
been analyzed heo e ically. Recen pape s by Ansolabehe e and Snyde (2000), A agones and
Pal ey (2002), G oseclose (2001) and Hummel (2010) epo a numbe o heo e ical esul s abou
he equilib ium p ope ies o spa ial compe i ion be ween wo candida es who diffe in quali y.2
These pape s use a amewo k o s udying he effec o candida e quali y on poli ical com-
pe i ion, based on he s anda d Downsian model compe i ion be ween wo candida es wi h an
impo an wis : any o e will s ic ly p e e he “highe quali y” candida e o he “lowe qual-
i y” candida e i he candida es loca e so ha he o e is indiffe en be ween he wo candida es
on he policy dimension.
G oseclose (2001) shows ha , o office mo i a ed candida es, exis ence o pu e s a egy equi-
lib ium is especially p oblema ic o small- o-in e media e alues o he quali y ad an age. When
he policy space is an in e al o he eal line and o e s’ p e e ences a e Euclidean he payoff
unc ions o he candida es a e discon inuous. This discon inui y implies ha he bes esponse
o he disad an aged candida e is no well de ined. In his se up A agones and Pal ey (2002)
desc ibed he equilib ium s a egies o a pa icula case: when he policy space is a ini e g id o
1See, e.g., S okes (1963), Kiewie (1983), and Kiewie and Zheng (1993).
2The e a e also some ea lie heo e ical pape s ha s udied ela ed kinds o asymme y, such as incumbency o pa isanship, e.g.,
Adams (1998), Be nha d and Ingbe man (1985), and Lond egan and Rome (1993).
1
poin s on he [01] in e al and when he ad an age is ela i ely small. Hummel (2010) s udies he
same en i onmen wi h he only diffe ence ha he conside s ha he ad an age migh be highe
han A agones and Pal ey (2002). In his case he median o e migh p e e o o e o he can-
dida e wi h supe io alence e en i he p e e s he policy o he o he candida e. Hummel (2010)
cha ac e izes he op imal ac ions o bo h candida es in a pa icula equilib ium o he game bu ,
unlike A agones and Pal ey (2002), he does no ully cha ac e ize any equilib ium o he game.
The cha ac e iza ion o he equilib ium s a egies o a con inuous policy space has no been
s udied so a and i is he main objec i e o his pape . The model we analyze ollows he s anda d
Downsian model wi h a con inuous policy space and he ollowing modi ica ions: we assume ha
o e s ha e quad a ic p e e ences ins ead o Euclidean and we assume ha he belie s o candida es’
on he dis ibu ion o he median o e ’s ideal poin a e unimodal (no necessa ily uni o m). By
changing he o e s’ p e e ences om linea o quad a ic he payoff unc ions o he candida es
become con inuous and he bes esponse o he disad an aged candida e is well de ined. Wi hin
his se up we a e able o ind condi ions o exis ence o mixed s a egy equilib ium and we a e
also able o desc ibe he equilib ium s a egies.
We ind a amily o unimodal dis ibu ions o he median o e ’s ideal poin ha gua an ees
exis ence o an equilib ium in which he ad an age candida e chooses a pu e s a egy ha con-
cen a es all he p obabili y in he expec ed loca ion o he median o e , while he disad an aged
candida e chooses a mixed s a egy ha alloca es p obabili y symme ically a ound he expec ed
loca ion o he median o e .
We ind necessa y and sufficien condi ions o his equilib ium o exis . These condi ions impose
es ic ions only on he candida es’ belie s on he o e s’ dis ibu ion o p e e ences. We ind ha
his equilib ium exis s only i he le el o unce ain y abou he loca ion o he median o e is
low enough ela i e o he size o he ad an age, ha is, when candida es belie e ha he median
o e ’s ideal poin is a ound 1/2 wi h high enough p obabili y.
Thecondi ionswe ind o exis ence do no es ic a all he size o he quali y ad an age, ha
is, he diffe ence in quali y be ween he wo candida es, as i was he case in G oseclose (2001).
2
Thus we show ha his mixed s a egy equilib ium may exis e en o small alues o he quali y
ad an age, which is when pu e s a egy equilib ium ails o exis .
As in simila models, we ind ha in equilib ium he ad an aged candida e ob ains a la ge
p obabili y o winning han he disad an aged candida e. We also ind ha as he alue o he
ad an age becomes la ge he p obabili y o wining o he ad an aged candida e inc eases, he
equilib ium s a egies o he wo candida es a e mo e diffe en ia ed, and he condi ions o exis ence
o equilib ium a e elaxed, ha is, he numbe o o e s (o he p obabili y wi h which he expec ed
median o e is a ound 12) needed o an equilib ium o exis is smalle . Finally, as he alue
o he ad an age becomes smalle , ha is, as he diffe ence be ween he wo candida es anishes,
he op imal s a egy o he disad an aged candida e mo es close o he ad an age candida e’s.
Tha is bo h playe s’ equilib ium s a egies con e ge o he expec ed median o e as candida e
’s ad an age sh inks o ze o.
The es o hepape p oceedsas ollows. Thenex sec ion desc ibes he o mal model. Sec ion
3 p esen s he de i a ion o he equilib ium s a egies and analyzes i s p ope ies. Finally, sec ion
4 con ains some concluding ema ks.
2TheModel
The policy space is he [01] in e al. The e a e  o e s, whe e is an odd and ini e numbe .
Each o e has a u ili y unc ion, wi h wo componen s, a policy componen , and a candida e
image componen . The policy componen is cha ac e ized by an ideal poin in he policy space,
wi h u ili y o al e na i es in he policy space a quad a ic unc ion o he dis ance be ween he ideal
poin and he loca ion o he policy. The image componen is cap u ed by an addi i e cons an o
he u ili y a o e ge s i he highe quali y candida e wins he elec ion.
The e a e wo candida es, and , who a e e e ed o as he ad an aged candida e and he
disad an aged candida e, espec i ely. Each candida e’s objec i e is o maximize his p obabili y
o winning he elec ion. We assume ha candida es belie e ha he ideal poin o each o e is
3

an i.i.d. d aw om a uni o m dis ibu ion in [01]
Thegame akesplacein wos ages. In he i s s age, candida es simul aneously choose posi-
ions in [01]. In he second s age, each o e o es o he candida e whose elec ion would gi e
him he highes u ili y. In case o indiffe ence, a o e is assumed o o e o each candida e wi h
p obabili y equal o 12.
Le deno e he policy posi ion chosen by candida e ,andle deno e he policy posi ion
chosen by candida e  Then, he u ili y ha a o e wi h ideal poin ob ains i wins he
elec ion is gi en by ()=−(−)2and his u ili y i candida e wins is gi en by ()=
−(−)2whe e 0deno es he size o candida e ’s ad an age.
Since he beha io o he o e s is unambiguous in his model, we de ine an equilib ium o
hegameonlyin e mso heloca ions a egieso he wocandida esin he i s ound. A
pu e s a egy equilib ium is a pai o candida e loca ions ( )such ha bo h candida es a e
maximizing he p obabili y o winning, gi en he choices o he o he candida e. A mixed s a egy
equilib ium is a pai o p obabili y dis ibu ions ()o e [01] such ha he e is no mixed
s a egy o  ha gua an ees highe p obabili y o winning han ,gi enand he e is no
mixed s a egy o  ha gua an ees highe p obabili y o winning han ,gi en.
No ice ha in his se up, i  hen all o e s wi h −(−)2−(−)p e e o o e
o candida e . The e o e, we ha e ha all o e s wi h an ideal poin +
2+
2(−)=b( )
p e e o o e o candida e  Since he ideal poin o each o e s is d awn om a uni o m
dis ibu ion, his implies ha he p obabili y ha a o e o es o he ad an aged candida e is
gi en by ( )=min{b( )1}and he p obabili y ha a o e o es o he disad an aged
candida e is gi en by ( )=max{01−b( )}
Simila ly i we ha e ha all o e s wi h an ideal poin +
2+
2(−)=b( )p e e
o o e o candida e  This implies ha he p obabili y ha a o e o es o he ad an aged
candida e is gi en by ( )=min{11−b( )}and he p obabili y ha a o e o es o he
disad an aged candida e is gi en by ( )=max{0b( )}
4
Since we assume ha he e a e  o e s, he p obabili y wi h which he ad an aged candida e
wins he elec ion is gi en by he p obabili y ha he ad an aged candida e ob ains he o es o a
leas a majo i y o he o e s. Because each o e will o e o candida e wi h p obabili y ( )
he p obabili y wi h which candida e is elec ed may be compu ed by he sum o he Be noulli
dis ibu ions co esponding o a leas a majo i y o successes o e  ials, ha is,
( )=

P
=+1
2¡
¢( )(1 −( ))−
Simila ly we could also show ha he p obabili y wi h which he disad an age candida e wins
he elec ion is gi en by ( )=

P
=+1
2¡
¢( )(1−( ))−=

P
=+1
2¡
¢(1−( ))( )−=
1−( )
Obse e ha ( )and ( )a e con inuous unc ions o ∈[01] and ∈[01]and he e o e
( )and ( )a e con inuous unc ions o ∈[01] and ∈[01] as well.
Finally, i = we ha e ha ( )=1and ( )=0, ha is, i bo h candida es choose
he same loca ion hen he ad an aged candida e wins wi h p obabili y one, because in his case
all o e s would p e e o o e o him.
No ice ha he payoff unc ions o he candida es in ou se up coincide wi h he c.d. . o a Be a
dis ibu ion wi h pa ame e s ==+1
2Such a dis ibu ion is unimodal and symme ic a ound
1
2This obse a ion allows us o offe an al e na i e in e p e a ion o ou model. Suppose ha
we ha e any numbe o o e s, e en a con inuum, and a unique median o e . Suppose ha he
candida es’ belie s abou he dis ibu ion o he median o e ’s ideal poin a e ep esen ed by his
Be a dis ibu ion. In his case he candida es’ payoff unc ions would be ep esen ed by he same
amily o c.d. . pa ame ized by he pa ame e o he Be a dis ibu ion ins ead o he numbe o
o e s o he o iginal se up. In his case ( )would ep esen he ideal poin o he o e ha
is indiffe en be ween he wo candida es. Thus in bo h cases we ha e ha he candida es’ belie s
abou he loca ion o he median o e ’s ideal poin a e mo e concen a ed a ound 1
2whene e he
numbe o o e s inc eases o , wha is he same, when he pa ame e o he Be a dis ibu ion
inc eases.
5
The payoff unc ions o he candida es in ou se up also coincide wi h hose o he Condo ce
ju y membe s (see, o example, Ki s ein and Wangenheim, 2010). This coincidence will p o e o
be help ul o ou analysis.
3 Equilib ium S a egies
When =0, nei he candida e has an ad an age, and we a e in he s anda d Downsian wo ld,
whe e in equilib ium he wo candida es loca e a 1
2and each wins wi h p obabili y 1
2. In gene al,
when 0, he e does no exis a pu e s a egy equilib ium. Diffe en e sions o his esul ha e
been s a ed and p o en in G oseclose (1999) and Be ge , Munge , and Po hoff(1999). The in u-
i ion is simple. I he disad an aged candida e’s loca ion is pe ec ly p edic able, he ad an aged
candida e can copy ha s a egy and win o su e. The e o e, a leas he disad an aged candida e
mus be mixing. The esul is ue unless is sufficien ly la ge ha canloca ea hemedian
and gua an ee a payoffo 1.
P oposi ion 1 I ≥1
4 he e is a pu e s a egy equilib ium in which wins wi h p obabili y
one.
(All p oo s may be ound in he appendix.)
In ou case, i 1
4, hen he e will be no pu e s a egy equilib ium. The aim o his pape is
o show ha i ∈¡01
4¢ he e exis s a mixed s a egy Nash equilib ium3in which he ad an aged
candida e chooses a pu e s a egy and he disad an aged candida e chooses a mixed s a egy. In
pa icula we show ha in his equilib ium he ad an aged candida e chooses a pu e s a egy
co esponding o he ideal poin o he expec ed median o e , =1
2while he disad an aged
candida e mixes be ween he wo policy loca ions =1
2−√and =1
2+√each wi h equal
p obabili y. We ind ha his equilib ium exis s as long as he numbe o o e s is la ge enough
ela i e o he size o he ad an age .Wealso ind he minimal numbe o o e s ha gua an ees
3I candida es ha e p i a e in o ma ion wi h con inuous ypes, hen his mixed equilib ium can be “pu i ied.” Tha is, he e will
exis an equilib ium in pu e s a egies, whe e he equilib ium loca ions o candida es will a y wi h his p i a e in o ma ion (A agones
and Pal ey 2005).
6
exis ence o his Nash equilib ium as a unc ion o he size o he ad an age.
We s a by demons a ing ha he s a egy p oposed o candida e ,e=(=1
2−√wi h
p obabili y 1
2and =1
2+√wi h p obabili y 1
2) is an op imal esponse o candida e choosing
e=1
2We p o e ha his holds ue o all alues o 
P oposi ion 2 Fo all 0and o all 01
4we ha e ha e=(=1
2−√wi h
p obabili y 50% and =1
2+√wi h p obabili y 50%) is a bes esponse o e=1
2
Nex we ha e o show ha he s a egy p oposed o candida e ,e=1
2is a bes esponse
o candida e choosing e=(=1
2−√wi h p obabili y 1
2and =1
2+√wi h p obabili y
1
2)No ice ha when candida e is choosing s a egy e=(
1
2−√w.p. 1
2;1
2+√w.p. 1
2)
candida e ’s p obabili y o elec ion is gi en by:
( e)=1
2( 12−√)+1
2( 12+√)
whe e
( 12−√)=

P
=+1
2¡
¢( 12−√)(1 −( 12−√))−
and
( 12+√)=

P
=+1
2¡
¢( 12+√)(1 −( 12+√))−
Mo eo e , obse e ha he unc ion ( )inc eases wi h  o all 1−p−2+2+1
emains cons an (( )=1) o ∈[1 −p−2+2+1p2+]and dec eases wi h  o
all p2+
[Figu e 1]
Tha is, ( 12−√)is inc easing in ∈[01−q1
4+2+√)cons an in ∈[1 −
q1
4+2+√ q1
4+2−√]and dec easing in ∈(q1
4+2−√ 1] and ( 12+√)is
inc easing in ∈[01−q1
4+2−√)cons an in ∈[1 −q1
4+2−√ q1
4+2+√]and
7
a pu e s a egy equilib ium in which he ad an aged candida e ob ains he o es om all o e s.
The e o e, wins wi h p obabili y one. ¨
P oo o p oposi ion2:
Gi en ha =1
2we sea ch o a alue o  ha maximizes he payoff unc ion o he disad-
an aged candida e, ha is, he ollowing exp ession.
(1
2)=

P
=+1
2¡
¢(1
2)(1 −(1
2))−
Ki s ein and Wangenheim (2010) show ha ()
 =¡−1
−1
2¢[( )(1 −( ))]−1
20 o
≥1
2.Thusweha e ha (1
2)is s ic ly inc easing in (1
2).
The e o e, in o de o ind a alue o  ha maximizes (1
2)i is enough o ind he alues
12 ha maximize (1
2)=max©0b(1
2)ª=max{01
4+
2−
1−2}; and he alues 12
ha maximize (1
2)=max{01−b( )}=maxn03
4−
2+
1−2o
No ice ha 12−√=a gmax1
4+
2−
1−2,and12−√∈£01
2¤when ∈(01
4]. Simila ly,
when 12we ind ha 12+√=a gmax3
4−
2+
1−2and 12+√∈£1
21¤when ∈(01
4]
The e o e, he p oposed mixed s a egy o candida e is a bes esponse o =12 o any
0¨
P oo o p oposi ion3:
I =1candida e ’s p obabili y o winning is gi en by
1( e)=1
2( 12−√)+1
2( 12+√)
We ha e seen ha 1( e)is inc easing in ∈[0q1
4+2−√)and ha 1( e)is
dec easing in ∈(1 −q1
4+2−√ 1] and, he e o e, he bes esponse o mus belong in
∙q14+2−√ 1−q14+2−√¸
14

Obse e ha o all ∈(01
4)we ha e ha 12−√q14+2−√121−
q14+2−√12+√and ha i =1candida e ’s p obabili y o elec ion 1( e)=
1
2( 12−√)+1
2( 12+√)can also be w i en as
1( e)=12(1 −+12−√
2+
2(−12+√))+12(+12+√
2−
2(−12−√))
o ∈[q14+2−√ 1−q14+2−√]
Thus, 1()
 =−
4(−12+√)2+
4(−12−√)20i and only i ³−12+√´2³−12−√´2
This implies ha 1( e)is dec easing o ∈[q14+2−√ 12) and i is inc easing o
∈(121−q14+2−√].
[Figu e 2]
The e o e 1( e)is inc easing o ∈[0q14+2−√],dec easing o ∈[q14+2−√ 12
)
inc easing o ∈(121−q14+2−√]and dec easing o ∈[1−q14+2−√ 1].This
implies ha when =1 he op imal esponses o candida e a e ei he =q14+2−√o
=1−q14+2−√bu no =1
2¨
P oo o p oposi ion4:
Le ’s show ha ()
0 o ∈[q14+2−√ 1
2)A simila analysis would p o e ha
()
0 o ∈(1
21−q14+2−√]
Since ( e)=1
2( 12−√)+1
2( 12+√)we ha e ha
()
 =1
2
(12−√)
 +1
2
(12+√)

whe e
( 12−√)=

P
=+1
2¡
¢( 12−√)(1 −( 12−√))−
15
and
( 12+√)=

P
=+1
2¡
¢( 12+√)(1 −( 12+√))−
We can compu e he de i a i e o ( 12−√)wi h espec o ()using he esul s in Ki s ein
and Wangenheim (2010) and ob ain (12−√)
 =¡−1
−1
2¢[( 12−√)(1−( 12−√))]−1
2
Thus we ha e ha he o al de i a i e o ( 12−√)wi h espec o by composing i wi h
i s pa ial de i a i e, ha is, (12−√)
 =(12−√)

(12−√)
 which can be w i en as
(12−√)
 =¡−1
−1
2¢[( 12−√)(1 −( 12−√))]−1
2(12−√)

and simila ly we ha e ha
(12+√)
 =¡−1
−1
2¢[( 12+√)(1 −( 12+√))]−1
2(12+√)

The e o e,
()
 =
2¡−1
−1
2¢[[( 12−√)(1 −( 12−√))]−1
2(12−√)
 +[( 12+√)(1 −
( 12+√))]−1
2(12+√)
 ]
and we need o p o e ha o la ge enough alues o we ha e ha
[( 12−√)(1−( 12−√))]−1
2(12−√)
 +[( 12+√)(1−( 12+√))]−1
2(12+√)
 ]
0
whene e ∈[q14+2−√ 1
2)
This holds i and only i
∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2(12−√)
 +(12+√)
 0
Fi s o all we will show ha ∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2
dec eases wi h and i ends o
ze o as  ends o in ini e.
No ice ha o ∈[q14+2−√ 1
2)we ha e ha ( 12−√)( 12+√)1
2
16
which implies ha ( 12−√)³1−( 12−√)´( 12+√)³1−( 12+√)´
always holds, since i does no depend on .
Thus ∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2
1and lim→∞ ∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2
=0
Sinceweha e ha (12+√)
 0we also ha e ha ∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2(12−√)
 +
(12+√)
 0will hold o la ge alues o 
Simila ly we could show ha o ∈(121−q14+2−√]we ha e ()
0which
comple es he p oo o he p oposi ion. ¨
P oo o p oposi ion5:
F om he las p oposi ion we know ha ()
 0 o ∈[q14+2−√ 1
2)i and only
i
∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸−1
2(12−√)
 +(12+√)
 0
which can also be w i en as
2
ln−
(12+√)

(12−√)
 
ln(12−√)(1−(12−√))
(12+√)(1−(12+√))+1
becauseweha e ha (12−√)
 0andwealsoha e ha (12−√)(1−(12−√))
(12+√)(1−(12+√))1implies
ln ∙(12−√)(1−(12−√))
(12+√)(1−(12+√))¸0
Fo =12 o be a local maximum we need o ha e ()
 ≥0 o =12−whe e 0
and →0
Thus we ha e o compu e lim→122
ln−
(12+√)

(12−√)
 
ln(12−√)(1−(12−√))
(12+√)(1−(12+√))+1
17
No ice ha when app oaches 1
2we ha e ha −
(12+√)

(12−√)
 →1and (12−√)(1−(12−√))
(12+√)(1−(12+√))→
1 hus we may apply l’Hopi al’s ule and we ob ain ha

lim→12
ln−
(12+√)

(12−√)
 
ln(12−√)(1−(12−√))
(12+√)(1−(12+√))= lim→12
−
(12−√)

(12+√)
 

−
(12+√)

(12−√)





(12+√)(1−(12+√))
(12−√)(1−(12−√))
(12−√)(1−(12−√))
(12+√)(1−(12+√))

Obse e ha since o ∈(12−√ 12+√)we ha e ha
( 1
2−√)=1−+1
2−√
2−
2(1
2−√−)and ( 1
2+√)=+1
2+√
2+
2(1
2+√−)
hen
( 1
2−√)
 =−1
2−
2(1
2−√−)2and (1
2+√)
 =1
2+
2(1
2+√−)2
which implies ha
µ−
(12−√)

(12+√)
 ¶=
1+ 
(1
2−√−)2
1+ 
(1
2+√−)2→1
and
−(12+√)(1−(12+√))
(12−√)
 
 =
2
(1
2+√−)3−2
(1
2−√−)3+22
(1
2+√−)2(1
2−√−)21
(1
2+√−)−1
(1
2−√−)

1+ 
(1
2−√−)2

2→2
√
Simila ly we ind ha
(12+√)(1−(12+√))
(12−√)(1−(12−√))→
1
4−
1
4−=1
and
(12−√)(1−(12−√))
(12+√)(1−(12+√))
 =
=(1−2(12−√))(12−√)

(12+√)(1−(12+√))−(12+√)(1−2(12+√))(12+√)
 (12−√)(1−(12−√))
(12+√)2(1−(12+√))2→4√
1
4−
becauseweha e ha
18
(12−√)
 →−1; (12+√)
 →+1
and
1−2( 12−√)→−2√;1−2( 12+√)→−2√
The e o e, we ob ain ha
lim→122
ln−
(12+√)

(12−√)
 
ln(12−√)(1−(12−√))
(12+√)(1−(12+√))+1=2¡1−4
8¢+1= 1
4
Fu he mo e, we compu e he sign o

ln

−
(12+√)

(12−√)




ln(12−√)(1−(12−√))
(12+√)(1−(12+√))
 o all ∈(q1
4+2−√ 1
2)
and o all ∈(01
4)using Ma hema ica and we ge ha i is posi i e (see igu e 7).
This implies ha i 
ln−
(12+√)

(12−√)
 
ln(12−√)(1−(12−√))
(12+√)(1−(12+√))+1 o →1
2 hen he inequali y should hold
o all ∈(q1
4+2−√ 1
2)as well.
Thus, =1
2is a global maximum i and only i ≥1
4¨
19

Figu e 1: The p obabili y ha a o e o es o he ad an aged candida e, p(x,y), as a
unc ion o he ad an aged candida e’s policy choice (x) gi en a policy choice o he
disad an aged candida e (y).
x
p(x,y)
½ ( y+ d/y )
½ ( (1-y)+ d/(1-y) )
y
Figu e 2: The p obabili y ha candida e A, P1(x,D), as a unc ion o he ad an aged
candida e’s policy choice (x) gi en he bes esponse o he disad an aged candida e
(D) when n=1 and d=0.05.
Figu e 3: The p obabili y ha candida e A wins, Pn(x,D), as a unc ion o he
ad an agedcandida e’s policy choice (x) gi en he bes esponse o he disad an aged
candida e (D) when n=5 and d=0.1, hus n>1/4d
Figu e 4: The p obabili y ha candida e A wins, Pn(x,D), as a unc ion o he
ad an aged candida e’s policy choice (x) gi en he bes esponse o he disad an aged
candida e (D) when n=5 and d=0.05 hus n=1/4d.