On (2005) Timing and Durability of Cabinets

The Timing of Cabinet Reshuffles in Five Westminster Parliamentary Systems
Author(s): Christopher Kam and Indriđi Indriđason

Source: Legislative Studies Quarterly, Vol. 30, No. 3 (Aug., 2005), pp. 327-363
Published by: Comparative Legislative Research Center
Stable URL: http://www.jstor.org/stable/3598639 .
Accessed: 28/03/2011 14:52
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=clrc. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Comparative Legislative Research Center is collaborating with JSTOR to digitize, preserve and extend access
to Legislative Studies Quarterly.
http://www.jstor.org
CHRISTOPHER KAM
Universityof British Columbia
INDRIDIINDRIDASON
Universityof Iceland
The Timingof
CabinetReshuffles in Five
WestminsterParliamentarySystems
Despite theirpolitical prominence,cabinetreshuffleshave not attracteda great

deal of scholarlyattention.We provide a theory of cabinetreshufflesthat emphasizes
both systematicand time-varyingcauses. In particular,we arguethatprime ministers
employ cabinetreshuffles to retainpower in the face of both intrapartyand electoral
challengesto theirleadership.Weuse repeated-eventsdurationmodels to examinethe
timing of cabinet reshuffles in Australia, Canada, Ireland, New Zealand, and the
United Kingdom in the period 1960-2001, and find support for several of our
hypotheses.
Introduction
A cabinetreshuffleinvolves the promotion(or demotion)of

ministersora reallocation of portfoliosbytheprimeministerduringthe
parliamentary term.Reshufflesare prominentpoliticalevents,but,
relativeto coalitionformation, portfolioallocation,andcabinetsurvival,
theyarelittlestudied.1Thisoversightis curiousgiventhatjournalists
coverreshufflesintensively(forexample,TheEconomist1998a,1998b,
and1999)andacademics frequentlycomplainof reshufflesundermining
thecabinet'smanagerial capacity(see, forexample,Franks1987,248;
Heady1974; and Rose 1987).Thisstateof affairsis alsotheoretically
unsatisfying: it is not immediatelyobvious,for example,why prime
ministerswouldeverreshuffletheircabinetsif theonlyeffectwereto
undercut thecabinet'sadministrative capacity.Itmaybe thatreshuffles
are simply mattersof routineor tradition-parliamentary standard
operatingprocedures, as it were-conducted without much thoughtas
to the likelycosts andbenefits.Suchan explanationmustbe viewed
skeptically,however.It runsagainsta gooddealof scholarlyresearch
LEGISLATIVESTUDIES QUARTERLY,XXX, 3,August 2005 327

328 ChristopherKam and Indri6iIndri6ason
demonstratingthat parliamentarypolitics are manifestly strategic in

nature(for instance, Cox 1987, Huber 1996, Lupia and Strom 1995,
and Mullerand Strom 1999), and it sits uneasily alongsidejournalistic
and historicalaccountsof reshuffles,which tend to connect reshuffles
to cabinetbattles, policy changes, and electoraltactics.
Indeed,the political and strategiccharacterof cabinetreshuffles
is clearlyvisible in HaroldMacmillan'sdismissalof Selwyn Lloyd and
six other British cabinet ministers in 1963. From late 1961 onward,
Lloyd,theChancellor,exhibitedcoolnesstowardtheexpansionistpolicies
that Macmillan(1963, 67-68, 86) and-in Macmillan'sestimation-
the Conservativepartyfavored.Macmillanworriedthatwithouta clearly
expansionistpolicy,thepartywouldlose direction(69) anddissentwould
fester (46), leadingto, "anattackon the Leadershipof the Party"(58).
Angeredby Lloyd'sindifferenceto this politicalimperativeandworried
that"theenemies of the Leadership,alreadynumerous[had]undoubt-
edly been strengthened" (59) by the government's inaction and
unpopularity, Macmillanconcludedthat"we [his cabinet]shouldrequire
not only new measures but new men" (72). Lloyd was therefore
dismissed along with six otherministerson June 13, 1962. There is a
clearlinkin Macmillan'saccountbetweena recalcitrantminister,internal
tensionin theparliamentary party,the government'sunpopularityamong
voters, and Macmillan's decision to reshufflehis cabinet.
We find such historical accounts to be both instructive and
convincing, promptingus to consider reshuffles as strategic devices
thatprime ministers(PMs) use to fend off intrapartyrivals and to win
elections rather than as standard operating procedures or purely
meritocraticmechanisms designed to recruit talent and competence
into cabinet. Hence, we propose a framework in which cabinet
reshuffles are contingent on the PM's position within the governing
party(or coalition) and in the electorate:The more vulnerablethe PM
to eitheran internalleadershipchallengeor electoraldefeat,the greater
the PM's incentive will be to reshuffle.The PM's vulnerabilityis, we
argue,a functionof institutionalrules and prevailingpolitical circum-
stances. To the extent that party rules governing leadershipselection
and cabinet management lower the overhead costs to rivals of
confrontingthe PM and limit the PM's authorityover ministers,these
rules expose PMs to leadership challenges and so invite cabinet
reshuffles. Insofaras political circumstancesare concerned,the PM's
position can be seen to grow weakeror strongerin relationto his or her
electoral and parliamentarypopularity. We thereforeexpect PMs to
reshuffle their cabinets whenever the PMs become less popularwith
voters or backbenchmembersof the governingcoalition.
CabinetReshuffles 329
Our model generatesa series of hypotheses about the timing of

cabinetreshuffles in parliamentarysystems. We test these hypotheses
against data on the timing of cabinet shuffles in Australia, Canada,
Ireland,New Zealand, and the United Kingdom in the period 1960-
2001. We analyzethese datawith duration-modelingtechniquessimilar
to those that have been used to study cabinet survival, although we
modify the techniquesto take accountof the fact thatreshuffles,unlike
cabinet failures, are repeatable events. Our chief finding is that
reshuffles are more likely during periods when the PM's personal
approvalratingsamongvoterslag behindthe party'sor coalition'spopu-
larity ratings. This result nicely connects our intrapartyand electoral
explanationsof reshufflesas it suggeststhatPMsreshuffletheircabinets
wheneverthey themselvesappearas politicalliabilities,a conditionthat
invites both internalchallenges and electoraldefeat.
The article follows in five sections. Section 1 describeshow we
define and count cabinetreshuffles.Althoughwe might have deferred
this matterto the methods section of the article,we feel it is betterto
define the article'scentralconceptvery clearly at the outset, given the
novelty of our topic. Section 2 outlines our theoreticalframeworkand
hypotheses.Section3 describesourdataandmethods.Section4 presents
and elaboratesupon the statisticalresults. The fifth and final section
offers our conclusions.
1. Defining Cabinet Shuffles
The Extent of Cabinet Change
We follow White (2000) and define reshuffles as any change in

ministerial personnel or responsibilities that affects more than two
officeholders and at least two portfolios. Isolated personnel moves,
such as a retiring cabinet minister being replaced by the portfolio's
junior minister,are ruledout by this definition.This classificationis a
good one for three reasons. First, to the extent that changes in the
statusof any single ministeraredueto idiosyncraticcauses(forexample,
ill health, scandal, and the like), they are resistantto causal analysis.
Second, contemporaneouschangesin ministers'ranksor portfoliosare
probably not independent events, providing all the more reason to
considerthe reshuffle, that is, the set of ministerialmoves, as the unit
of analysis.An excerptfroma recentinterviewwith a Canadianminister
drives home the point.2
Q: Why does the PM not simply change one underperformer

at a time ratherthan making these broadermidtermchanges?
There is organizationalstabilityto think about,afterall.
A: Well, let me give you an example. In our first Parliament,

David Collenettegets into trouble-he called a judge or some-
thing about a case-and our traditionhere is that when some-
thing like that happens, the minister resigns, and David did.
Now, at the time, he held the defenseportfolio,a majorportfolio.
So, now this createstwo problems.First,do you pull a rookieup
from the backbenchand put them into a majorportfolio?Or do
you move a veteran into the defense portfolio and give the
rookie a minor portfolio?And don't forget that the PM has to
deal with all sorts of restrictions:he has to make sure that all
provinces have a minister,and then there's the relative impor-
tance of a particularportfolioto certainregions.
A variety of constraintsmake it difficult, then, for PMs to respondto

the loss of an individualministerin a limited andperfunctoryfashion.
Instead-and this leads to the third and final reason for employing
White's definition-we frequentlyobserve PMs choosing (or being
compelled) to set into motion a chain of promotionsand lateralmoves
thatreverberatethroughoutthe ministryand the party.At a minimum,
the fact that PMs make more expansive personnelmoves than strictly
necessary reveals that they are willing to trade off organizational
stabilityto satisfy otherobjectives.
The TemporalDimension
Alt (1975) adds a temporaldimensionto reshuffles,arguingthat

thepersonnelmoves andorganizationalchangesinvolvedin thereshuffle
need not occur simultaneously.Clearly,if the PM sackedthreecabinet
members one day and named replacementsa week later, one would
still say the activityconstituteda single reshuffle.Alt suggestsallowing
reshufflesto take place over a two-monthtime span.This periodis too
expansive for our purposesas our sample includesAustraliaandNew
Zealand(whichhave only three-yearconstitutionalinterelectionperiods
[CIEPs]);we use a one-monthwindow instead.
The Organizational Level
The titles of the variousministerialoffices vary by country,but it

is almostalwayspossibleto distinguishclearlybetweencabinetministers
proper,seniorministers-who sit outsidecabinetbut still exercisesome
independentauthorityandmay occasionallyattendcabinetmeetings-
andjuniorministers,who aremeredelegatesof theirrespectivecabinet
ministers.3Alt and White restricttheir attentionto the movement of
cabinetministers.Recent researchindicates,however, thatthe alloca-
tion andmovementof noncabinetministerswithin the governmentis a
significantelementof parliamentary politics(Mershon1999;Thies2001).
Accordingly,we also take accountof movementsof seniornoncabinet
ministers.Under our rule, the movement of a single cabinet minister
and two senior ministersoutside cabinetwould count as a reshuffle-
provided that these ministers straddledat least two portfolios and all
three moves occurredwithin one month.4
It is worthemphasizingthatwe takereshuffles(howeverextensive
theymightbe) to involvethe reorganization of an existingcabinet,notthe
wholesaleinstallationof a new cabinet.Thisassumptionis tacitlyacknowl-
edged in the literature-studentsof cabinetsurvival,for example,do not
countreshufflesas terminations-but it needs to be made explicithere.
We deemeda new cabinetto have come into existenceafterelectionsor
changesin the partisancompositionof the cabinet.Thuscabinetchanges
betweenelectionsthatdidnotalterthecabinet'spartisancompositionwere
classifiedas reshuffles(providingthese changesmet the aforementioned
criteria).We are sensitiveto the fact thata numberof argumentscan be
raisedregardingthisclassificationscheme.5Thatsaid,themannerin which
we classifynew andexistingcabinetscomportswith how coalitiontheo-
ristshave dealtwiththe matter(forexample,Kinget al. 1990,856; Laver
and Schofield 1990, 145; and Warwick1994, 27-28). The one area in
whichwe have departedfromcoalitiontheoristsis in the continuityof the
premiership.Whereas many coalition theorists take a change in the
premiershipto be indicativeof a new government,we do not-unless the
new PM is froma differentparty.ThusCallaghan'stakeoverfromWilson
in 1977 and Major's takeover from Thatcherin 1990 are counted as
reshufflesof existingcabinets,becausethesetakeoverswere also accom-
paniedby wider changesto the cabinet.Again, we realize thatthere is
roomfordebate,not simplyon thesespecificcodingdecisions,butalso on
thebroaderquestionof how analyticallyseparatereshufflesare(or should
be) fromthewholesaleformationandterminationof governments.Thisis
a debatebest left for futureresearch,however, and one best taken up
aftera broaderbase of knowledgeaboutreshuffleshas been established.
2. The Causes of Cabinet Reshuffles
Cabinet Ministers, PMs, and Reshuffles
Underpinningourmodel of reshufflesaretwo assumptionsabout

the natureof parliamentarypolitics. The first assumptionis that PMs
are primarilyconcernedwith maintainingpower, an assumptionmade
in the coalitionliteratureby Leubberts(1986). To maintainpower,PMs
must: a) win elections, and b) fend off intrapartyrivals to theirleader-
ship. The second assumptionis that cabinetministershave inherently
mixed motives with respect to the PM and the party (or their respec-
tive parties, should a coalition be in power). On one hand, ministers'
fortunesare tied to their party for the simple reason that to become a
cabinetministerone mustbelong to an electorallysuccessful party.On
the otherhand, cabinetministersharborprivatedesires, perhapsfor a
more importantcabinet post (such as finance or foreign affairs) or
even the premiershipitself. Thus, cabinetmembers,the PM included,
are at once colleagues and rivals, a situationthatmakes it difficult for
the PM to maintaincollectiveresponsibilityandensurethatthe ministers
remainfaithfulto the PM and the government'sdeclaredplatform.
As a consequence, PMs confront adverse selection and moral
hazardproblems.The adverseselection problemarises at the cabinet-
buildingstage.Themostambitiousandpoliticallytalentedpartymembers
are attractiveministerialcandidates-yet it is precisely those qualities
thatmakethesemembersthemostwilling andbestequippedto challenge
the PM. From the PM's perspective, then, the problem is selecting
cabinetmemberswho will remainloyal to him or her and to the party
once placed in chargeof a portfolio.The attendantdifficulty is thatthe
PM must make these selections without full informationbecause the
minister's loyalty and ambition(that is, the minister's type) has only
been imperfectlyobserved, if at all. Once the cabinet is constructed,
the moralhazardproblememerges.In parliamentarygovernment,PMs
delegatepolicy portfoliosto cabinetministers,who in turndevelop and
implement policy within their portfolios (Laver and Shepsle 1996).
Cabinetministers,however, have incentives to use their departments
to serve their own ambitions.So, for example, ministersmay initiate
unauthorizedprojectsto gain the favorof a pivotalsection of the parlia-
mentaryparty or electorate, or they may surreptitiouslyslow down
projectsin theirportfoliothat are closely associatedwith the PM, and
so on.6Nevertheless,once a ministeris installedin a particularportfolio,
the PM is limitedin his orherabilityto monitorandcontroltheminister's
actions directly,hence the moralhazardproblem.
PMs can use reshuffles to address these problems.7 Quite

obviously, reshuffles provide the PM with the opportunityto sack or
demote waywardministers.Reshuffles can also be used to handrivals
"poisonedchalices,"thatis, to placethemin controversialorunappealing
portfolios.At a somewhatmore theoreticallevel, reshuffles allow the
PM to addressthe adverse selection problem:as the PM updates on
ministers'types,he or she canreshuffleundesirabletypesoutof powerful
positions into weaker ones from which they are unlikely to be able to
challenge the PM's authority.What is perhapsless obvious is that if
ministersare affected by their colleagues' actions and if, in addition,
their scope of action in a portfolio depends to a degree on their
predecessor'sbehavior,thenreshuffleslead ministersto curbtheirself-
interestedbehavior(Indri6asonand Kam 2003).8 Thatis, reshuffles-
even if they consist only of ministers switching chairs at the cabinet
table-combat the moral hazardproblem. Reshuffles can thus serve
as cheap substitutesfor more costly directmonitoringof ministersby
the PM.
It is because reshuffles have these corrective effects (from the
PM's perspective)thatwe drawconnectionsbetween a parliamentary
system's institutionalfeaturesand the frequencyof cabinetreshuffles
in that system. Two types of institutionalfeaturesare relevant:those
that influence the cabinet ministers' incentives to challenge the PM,
andthose thatdirectlylimitthe PM's scope of action.Institutionalrules
thatmakeit easierforministersto mountchallengesto the PM'sauthority
effectively increase the extent of the moral hazardproblem that the
PM faces, and so requirethe PM to be more vigilant. With reshuffles
serving as cheap and effective means of reducing the PM's agency
loss to ministers,we expect the PM's increasedvigilance to take the
formof frequentreshuffles.Institutionalrulesthatlimit the PM's scope
of actionwill also leadto frequentreshuffling.A PM canpursuevarious
strategiesto fend off challenges, but when the set of strategiesavail-
able to a PM is constrainedin some manner,we expect the PM to rely
more heavily on those strategiesthat remain available. We consider
eachof these institutionalfeaturesandtheirrelationto cabinetreshuffles
in turn.
The Relationship between

Leadership Selection Rules and Reshuffles
The institutionalfactors that induce ministersto engage in self-
interestedbehaviorinfluence the PM's propensityfor reshufflingthe
cabinet. Of particularrelevance in this regardare the party'srules on
leadershipselection.9Rules governing leadershipselection influence

ministers'opportunitiesto advancetheircareers,somerulesencouraging
ministersto front direct challenges to the PM, others forcing them to
seek advancementpatiently underthe tutelage of the PM. Generally
speaking,the more vulnerablethe PM undera given set of leadership
selection rules, the greaterthe ministers'incentiveswill be to use their
portfolios in a self-interestedfashion, and the greaterthe benefit the
PM will receive from reshufflingthe cabinet.
Historically, party leaders, of cadre parties in particular,were
selected by informal processes. Senior party figures acting as
informateurswould sound out various elements of the partyand then
consensuallyselect a leader.This was, for example,the situationin the
British Conservativeparty until 1964. Mass and populist parties, in
contrast,have always soughtto democratizeleadershipselection (Katz
and Mair 1995). The two main ways of doing this are to a) allow the
members of the parliamentaryparty to vote on the leadership,or b)
place leadershipselection in the hands of broadconventionsof mem-
bers of Parliament,extraparliamentary representatives,and affiliated
groups(for instance,unions). The limiting case of this second method
is a one-member-one-vote(OMOV) system in which all registered
partymemberscanvote in the leadershipelection.Eachof thesesystems
leaves a PM more or less vulnerableto internalchallenges, with the
PM's vulnerabilitydecreasingas the size of the leadership"selectorate"
expands (Bueno de Mesquitaet al. 2002).
The explanationfor the relationshiplies in Michel's IronLaw of
Oligarchy:As an institutionexpandsits membershipanddemocratizes
its procedures,there is a tendency for power to become increasingly
concentratedin a small oligarchyat the institution'sapex. Thus a PM
atthe headof a convention-basedor OMOVpartyis in a securesituation
(Weller1994),beholdento a largebutpoorlyorganizedgroupof people
who, for logistical reasons, come together perhapsonly annually or
biennially.In addition,the PM is one of the few people in the partywho
has the resources(patronage,a largepolitical staff,partywidecontacts,
and so forth) required to manage the party's infrastructure and
membership.The security affordedto the PM by these arrangements
removes one of the main incentives for cabinetreshuffles.
Arrangementsthat leave leadershipselection in the handsof the
parliamentaryparty or a small circle of party insiders are altogether
more dangerousfrom the PM's perspective. First, the parliamentary
partyis not a large group,perhaps50 to 350 members;it can easily be
organizedby an internalrivaland,given the regularityandfrequencyof
caucusmeetings,thisorganizationcouldbe accomplishedat shortnotice.
Moreover,many backbenchmemberswill not have the PM's interests

at heart:some will see the government'spolicies as a threatto their
reelectionhopes; otherswill resentthe fact thatthey have been left out
of the cabinet.These motives are always present,of course, no matter
what the leadershipselection rules are. When an informalor caucus-
based system is in place, however,membersof the parliamentaryparty
not only have the motive to challengethe PM, but also the opportunity
and the capacity to do so. In sum, PMs selected by caucus-basedor
informalsystemswill frequentlyfindtheirauthoritychallengedby internal
rivals (likely theircabinetministers),andhence have strongincentives
to engage in repeatedreshuffles.10
Cabinet Management Rules, Coalition Government,

and the Adverse Selection Problem
Rules on leadershipselection primarilyaffect the moral hazard

problem of cabinet government. The adverse selection problem is
affected by other types of rules, in particular,rules limiting the PM's
managementof the cabinet. Consider the counterfactualscenario in
whichthePMhasthe abilityto pickperfectlyloyal andcapableministers.
In such circumstances,there is no reason, save random scandal, for
the PM to reshuffle the cabinet.Of course, ministersare not perfectly
loyal and capable,and,underthese more realisticconditions,PMs can
be expected to use reshuffles to get rid of some of the bad apples and
recruitinto cabinetmore-loyal(andperhapsmore-competent)ministers.
Not all parties allow PMs to do this, however. Laborparties in
Australia and New Zealand, for example, elect and present to their
PMs a slateof cabinetministers,leavingthePM only theroleof assigning
portfolios among the ministers.Such rules limit a PM's ability to deal
with the adverseselection problem:even if the PM knew a ministerto
be overly ambitiousand threateningor ideologically uncongenial,the
party's rules on cabinet management would prevent the PM from
refusing that ministera place in cabinet or unilaterallyremoving that
minister from the cabinet. Moreover, whereas PMs can always be
expected to consider past loyalty when selecting their ministers,it is
unlikely that the parliamentaryparty-composed as it is of members
intenton advancingtheirown careers-will place as greata weight on
a minister'spersonalloyaltyto the PM.Thusone wouldexpectministers
in partiesemployingthese sortsof rulesto be relativelymoreambitious
and threateningthan ministersin cabinets selected solely by the PM.
At any given point in time, then,PMs who areconstrainedin managing
theircabinetshavemuchstrongerincentivesforreshufflingtheircabinets
than do PMs who have complete authorityover the composition of

their cabinets. Of course, under these restrictiverules, reshuffles are
dominatedby lateralmoves, with sittingministersbeing reassignedto
new portfoliosratherthanbeing sackedandreplacedby backbenchers.
The effect of limiting the PM's ability to manage the cabinet is
contingent on the leadershipselection system in place. If the prime
ministercannotrefuseoverly ambitiousministersa place in the cabinet,
and if, in addition, he or she cannot unilaterally sack them from
the cabinet,then the ministersare subjectto greatermoralhazard.Add
to this situation an informal or caucus-based leadership selection
system and one furnishesthese overly ambitiousministerswith easy
means to challenge the PM. This scenario exacerbates the moral
hazardproblem still furtherand creates even strongerincentives for
the PM to reshuffle the cabinet. In contrast, a convention-basedor
OMOV leadershipselection system presentseven the most ambitious
ministerswith significant organizationalobstacles to moving against
the PM. Thus, although the adverse selection problem still persists
under a convention-basedleadershipselection system, the extent of
the moral hazardproblem and the concomitantneed for reshuffles is
somewhat reduced.
It is not always the PM's party that limits his or her ability to
select or drop cabinet ministers. PMs of coalition governmentsmay
also be constrainedby coalition partners'political demandsor by the
formal-or, more often, informal (Strom and Miller 2000, 268)-
stipulationsof a coalitionagreement.Coalitionpartnerscommonlyinsist
on controllinga particularset of portfolios and may even requestthat
othercoalitionpartiesrefrainfromplacingspecific individualsin certain
portfolios.1 In theory,these types of constraintsexacerbatethe adverse
selection problemthat the PM faces, just as ideological and strategic
differencesbetween coalition partnersmay magnify the moralhazard
problem. The PM may not, however, have a free hand to reshuffle;
coalitionpartnersmay veto certainministerialmoves, andthe coalition
agreementmay place the managementof a set of portfolios beyond
the PM's authority.At any point in time, then, coalition cabinets are
less likely to be reshuffledthan single-partycabinets, if only because
coalition partnersmay wield a veto over reshuffles.
These argumentsare structuralor institutionalin nature;they
connect PMs' propensityto reshuffle their cabinets to fixed institu-
tional characteristics.This line of reasoning allows us to construct
hypotheses aboutthe probabilityof a reshuffle occurringat any given
timein certaininstitutionalenvironments.Weofferfoursuchhypotheses,
each prefacedby an obligatory"At any given point in time":
Hypothesis1: Reshufflesaremorelikelyif the PM is selected

informallyor by the parliamentarypartythan if the PM is
selectedin a conventionorby one-member-one-vote(OMOV).
Hypothesis2. Reshufflesaremorelikelyif the PM'sabilityto

managethecabinetis limited.
Hypothesis 3: The effects of cabinet management and

leadershipselectionrulesareinteractive.Therefore,reshuffles
aremostlikelyif institutional
rulesbotha) limitthePM'sability
to selectanddropministers,andb) empowertheparliamentary
partyto selectleaders.
Hypothesis4. Reshufflesare less likely if the cabinetis a

coalitioncabinet.
Changing Political Conditions and Reshuffles
So far we have connected the PM's propensityto reshuffle to a

set of fixed institutional rules. These time-invariant institutional
hypotheses are importantinsofaras they provideus with ex ante fore-
casts of the timing of cabinet reshuffles. The political environmentis
not invariant,however,and PMs may undertakereshufflesin response
to or in anticipationof changesin the politicalenvironment.Thistype of
logic is prominentin the cabinet survivalliterature(see, for example,
Alt and King 1994, Lupiaand Strom 1995, Martin2000, andWarwick
1994),andone mightthinkof reshufflesanalogously,thatis, as responses
to changes in the political environment. Adopting a time-varying
approachto cabinetreshufflesdoes not entailabandoningthe theoretical
frameworkthatwe have outlined.On the contrary,the assumptionthat
PMs are motivated primarilyby a desire to maintainpower remains
central to our argument. We see PMs using reshuffles to manage
electoralpolitics andrespondto parliamentarycrises thatthreatentheir
leadership.
Reshuffles, Electoral Cycles, and Political Popularity

PMs in the countriesstudiedhere (and, indeed, in several other
parliamentarysystems) are empowered to dissolve Parliamentand
initiate elections. Research indicates that PMs use this authority
strategically,calling elections when the PM is well positioned to win
them (for exmaple, Lupia and Strom 1995 and Str0m and Swindle
2002). It is reasonableto hypothesize(asjournalistsoften do) thatPMs

use reshuffles similarly,reshufflingtheir cabinets in the run-upto an
election in orderto put a new gloss on theirgovernments.12 IrishPrime
MinisterGarretFitzGerald,for example,reshuffledhis cabinetto tryto
distanceministersfromthe unpopularpolicies they had been forcedto
implementin theirerstwhileportfolios(FitzGerald1991,621). Comments
from a senior Canadiancabinetministerecho this point:
[of a reshuffle].Do
Q: Youmentionedthatthere'santicipation
you havea senseof whena shuffleis about
to occur?
A: Abouttwo to threeyearsin, by aboutthe midpointof the

term,therewill be a substantial
shuffle.Beforethen,of course,
they'llbe a fewministerswhohaveto bemovedbecausethere's
troublein theirportfolios.
Q: Now somebodytoldme thatthesemid-to late-termshuffles

arenot madefor overtlyelectoralreasons,butratherbecause
thePMO[thePrimeMinister'sOffice]will soundoutministers
to see who might be retiring,and then shuffle to ease the
transition.Is thataccurate?
A: Well,they'rewrong.Politicalpartiesarein the businessof

communication, andthey haveto makesurethatthey'vegot
in
people place that can defendthe government's recordand
who canleadthe chargeon new issues.
This cabinet minister suggests that reshuffles are part of an electoral

programand that they become more likely as the parliamentaryterm
wears on.
Of course, public opinion and political events do not always
coincide with this electoraltimetable-scandals breakout, economies
stall, and unpopularpolicies are sometimes implemented.As citizens
become aware of the effects of these events, they attribute them
(correctlyor not) to some subset of political actors and then translate
theseattributionsintovote intentions(GomezandWilson2001).Declining
political popularitysignals that the public has reachedthe attribution
stage of this process, that it has blamedthe governmentfor prevailing
conditionsandis now poised to vote againstit. Low levels of popularity
portend electoral defeat and make the parliamentaryparty nervous
and receptiveto leadershiprivals. Reshuffles are one means thatPMs
have of respondingto these conditions.If, for example, the decline in
political popularityis due to scandalousministerialbehavior,then the

offending ministers can be sacked to create an image of probity and
responsibility. If, in contrast, the decline in popularity is due to an
unpopularpolicy, then the responsibleministercan be reassignedto a
less public, less controversialportfoliowhile a ministerwho is a "safe
pair of hands" assumes the task of selling or quietly scrapping the
policy. A formerCanadianministerdescribedthe politicalthinkingand
dynamicsbehind these sorts of moves:
I had the PM's trust, and so I was very much the PM's "hit man" on crisis manage-
ment. He'd say to me, "You go and run that department."So, for example, in 1976
when we had a separatistgovernmentcome to power in Quebec, he createda ministry
of federal-provincialaffairs and put me in charge because we needed a strategy to
counterseparatismdirectly,to deal with the issues, and develop policy. Then ... let's
see, in '72 I had Health and Welfarebecause social policy was a prioritybecause we
needed the NDP's [New Democratic Party] support. In '78, '79, I had federal-
provincial relations because of the separatistand constitutional issues. Moving to
Justice was largely continuing these constitutional things. Later I went to EMR
[Energy,Mines, and Resources] because we had an energy crisis, and then in 1982 I
went to Finance because of stagflation.Allan McEachanhad been there, and he had
the PM's confidence, but he had troublewith the 1981 budget. He was facing a very
difficult environment,and the PM felt that he needed a new face in the area, that he
needed to createa new political environment.You see a shuffle is not like a change in
government,you have the same PM, the same policies ... it is more having to do with
implementation.That and a sense among the public that the governmentis tired and
unpopular.You need to change faces, or more often change chairs.
Do PMsreshufflein responseto declininglevels of partypopularity

or declininglevels of personalpopularity?The two quantitiesarenot at
all the same thing: the party can always find a new PM to replace an
unpopularincumbent;a PM cannot find a new party.Consequently,a
dropin the PM's personalpopularityhas both electoraland intraparty
implications.Declines in the PM'spersonalpopularitynot only threaten
the party'schances at reelection (Clarke,Ho, and Stewart2000), they
also invite leadershipchallenges.We expect, then,thatwhile reshuffles
are generally more likely when the governing party's (or coalition's)
popularitydeclines, they are especially likely when the PM's personal
popularitydeclines. The second partof this hypothesisis perhapsmore
precisely statedin a conditionalform:at any given level of partypopu-
larity(thatis, when we controlfor partypopularity),reshufflesbecome
more likely as the PM becomes more personallyunpopular.
Reshuffles and Coalition Politics
The relationshipbetween coalition partnersoccurs at the inter-

section of electoral and parliamentarypolitics. Research (Lupia and
Strom 1995; Martin2000) shows thatcoalition governmentsare more
likely to collapse when opinion polls show that one of the incumbent
coalition partnersstandsto gain more ministerialportfolios from new
elections or a new roundof coalitionbargaining.Reshufflesprovidean
opportunityfor coalitionpartnersto strikenew deals (thatis, to attaina
redistributionof portfolios)withouthavingto terminatetheirgovernment.
Naturally,one expectsPMs to turnshiftsin popularitybetweencoalition
partnersto their advantage,althoughwhat is to their advantageis not
immediatelyobvious.If, for example,thejuniorcoalitionpartnermakes
popularitygains, then the PM may reshuffle in a fashion that recog-
nizes the junior partner'simprovedposition and that, in consequence,
securesthe government'scontinuationandmaintainsthe PM in office.
If, on the otherhand, it is the PM's party (almost certainlythe senior
coalition partner)thatrises in the polls, then the PM may reshufflethe
cabinet to his or her own party's advantage.Whateverthe case, we
would expect the likelihood of a reshuffle to rise whenever there are
shifts in popularitybetween coalitionpartners.
Reshuffles and Parliamentary Politics
Parliamentaryevents may also induce PMs to reshuffle their

cabinets.Dissent and defections, for example, indicatethatbackbench
membersof the governing coalition are dissatisfiedwith the cabinet's
policies. Overt breakdownsin party discipline also open the door to
leadershipchallenges, the rebels representing,in effect, a fifth column
within the party or coalition that challengerscan use to their advan-
tage. Insofaras reshufflesmove unpopularministersout of cabinetor
outof controversialportfolios,andinsofaras theyundercutrivals'efforts
to organize discontentagainst the PM, we expect reshuffles to occur
morefrequentlywhen levels of dissentwithinthe PM'spartyarehigh.13
This time-varying account of cabinet reshuffles provides five
additionalhypothesesaboutthe risk of cabinetreshuffles:
Hypothesis 5: Reshuffles become more likely as the

parliamentarytermprogressesand the election nears.
Hypothesis 6: Reshuffles become more likely as the governing

party's(or parties') popularitydeclines.
Hypothesis7. Reshufflesbecome more likely as the PM's

personalpopularitydeclines.
Hypothesis8: Reshufflesbecomemorelikelywheneverthere
arechangesin therelativepopularityof coalitionpartners.
Hypothesis9. Reshufflesbecomemorelikelyas parliamentary

dissentwithinthe PM'spartyincreases.
The Cost of Reshuffles
Ourargumentis not intendedto give the impressionthatreshuffles

are costless or thatPMs do not encounterdiminishingmarginalreturns
from constant reshuffling. On the contrary,too much reshuffling is
likely to generatenegative returnsof two sorts. First, it may lead to a
loss of control over the bureaucracybecause ministers are never in
place long enough to develop the administrativeand policy expertise
needed to adequately oversee their departments.This is the classic
academicdiagnosis of cabinetreshuffles. Second, constantreshuffling
may createan impressionof governmentalinstabilityamongobservers,
including party members, foreign investors, and voters, and this
impressionmay damage the PM and the government.We operateon
the premise that these costs are roughly constantacross countriesand
parties,whereasthe benefits of reshufflingvarywith the relevantinsti-
tutionalrules and political circumstances.In otherwords, undersome
conditions(describedpreviously),morefrequentreshufflesareoptimal
given the fixed costs involved.
3. Data and Methods

Data
Our data come from five countries: Britain, Ireland, Canada,

Australia,andNew Zealand.The databegin in 1958, end in 2002, and
encompass2,647 cabinetmonthsbrokeninto246 spells(periodsbetween
reshuffles).Of these 246 spells, 172 end in a reshuffleandthe remainder
are censored.14 The decision to limit our sampleto these five countries
reflects a conscious choice on our partto avoid conceptualstretching.
These Westminster-styleparliamentarysystemsoperateon similar(and
comparativelyrigid) constitutionalassumptionsabout collective and
individualministerialresponsibility.Giventhe difficultyin codingwhat
are often implicit constitutionalconventions, we opted to tackle the
problemby limiting our analysis to these similarsystems. The study's
concentratedscope is also reflectiveof the fact thatwe triedto improve

the qualityof our databefore increasingthe quantity.For example,we
startedourtime series in each countrywith the first governmentof the
1960s because these governmentstend to be the first for which public
opinion statistics are available on a consistent basis. The datasetalso
contains informationon many hard-to-getvariables, such as prime
ministerial approvalratings and the number of rolls calls on which
governing parties experienced internaldissent. As a result, we were
able to do a fairly good job of testing the alternativehypotheses for
reshufflesthatwe outlined-at leastup to thepointthatmulticollinearity
set in. We do not claim that our choices on these mattersare perfectly
corrector beyond reproach,but we do feel they are defensible.
Insofaras partyrules are concerned,the datacontaina good deal
of variance.Canadianpartiesselect and remove their leaders in large
conventions that take place every several years. Australianand New
Zealandparties,in contrast,employ very flexible selectionrules:at any
meetingof the parliamentaryparties,a membermay introducea motion
of nonconfidencein the leader;a simple majoritysecures or deposes
the leadership.The two Britishpartieshave revampedtheirrulesseveral
times over the past 40 years. In 1981, Labouralteredits rules so thata
party conference of members of Parliament, unions, and
extraparliamentary membersratherthanthe parliamentarypartyalone
selected the leader(ButlerandButler 1994). UnderBlair,the partyhas
moved even closer to an OMOV system. The British Conservatives
have also widened their leadershipselectorateover time. In 1965, the
old,informalsystemof leadershipselectiongaveway to a formalballoting
of the parliamentaryparty,and this system has, in turn,recentlygiven
way to a conference-OMOVhybridthatis similarto Labour'ssystem.
The two main Irishparties,FiannaFail and Fine Gael, have also seen
theirleadershipselectionrulesevolve overtime, frominformalsystems
to caucus-basedsystems very similarto those employed by Australian
and New Zealand parties. Finally, the Australianand New Zealand
Labourpartiesimpose formallimitationson the PM's managementof
the cabinet. In these two parties, the PM's power over the cabinet is
limitedto distributingportfoliosamonga groupof ministerschosen by
the parliamentaryparty.15
Methods
We tested ourhypotheseswith a variantof the Cox regression,or

model. Ourdependentvariableis, therefore,the
proportional-hazards,
hazardrateof a cabinetreshuffle,thatis, the rateat which cabinetsare
reshuffledat a given instantin time, conditionalon not yet havingbeen

reshuffled. The hazardrate of cabinet reshuffles is a latent variable;
what we observe (hence what we often talk of) is a cabinet's survival
time, that is, the time until cabinets of a given type are reshuffled.
Proportional-hazards models arefamiliarstatisticaltools in the analysis
of cabinetsurvival(for example,King et al. 1990 andWarwick1994),
buttheirapplicationto cabinetreshufflesis not straightforward because
reshuffles,unlike cabinetfailures,are repeatableevents: a cabinetcan
collapseonly once, butit canbe reshuffledseveraltimes. The repetition
of eventspushedus to analyzeourdatawith a conditional-risksmodel.16
A conditional-risksmodel makes three modificationsto the standard
proportional-hazardsmodel. First, the risk set is constructed (and
likelihoods calculated) on the assumption that later events are
encounteredonly afterearlierevents have occurred.17 Hence, a cabinet
is at risk of a second reshuffle only afterhaving been reshuffledonce
already.Second, standarderrorsand baseline hazardsare adjustedto
account for event dependence in the data.18In our application,this
adjustmentinvolves clusteringthe standarderrorsby governmentand
stratifyingthe model so that differentbaseline hazardfunctions (but
identicalcoefficients)areestimatedfor initialandsubsequentreshuffles.
Third,time is measuredas inter-eventtime. In otherwords,time returns
to 0 aftereach reshuffle. This is not the only way to delineateanalysis
time in a conditional-risksmodel, but doing so not only resultsin statis-
tically well-behavedmodels, it betteraccordswith the assumptionswe
have made aboutthe costs of reshuffling;if reshuffles involve a fixed
cost to the PM, then the PM's decision to reshuffleshould reflect how
long it as been since the last reshuffle.
4. Results
Table 1 presentsour results.In additionto countrydummies,the

models containthe following controlvariables:the numberof ministers
in the government,the cabinet's majoritystatus, the time in months
until the end of the country'sCIEP,a Januarydummy,and the number
of priorreshuffles.If we assumethatministersrunroughlyequalrisks
of succumbingto ill health,scandal,and so on, then the more ministers
thereare in the ministry,the more vacancies should open up in a given
period of time. In consequence, PMs will have to reshuffle larger
ministriesmore frequentlythansmallerministries,if only to offset the
effects of attrition.Workon cabinetfailure(see Warwick1994) shows
that majority cabinets are more stable than minority cabinets. The
majoritystatusdummycontrolsforthepossibilitythatthisstabilityextends
to reshuffles. The time until the end of the CIEP controls for the fact
that Australiaand New Zealand have 36-month CIEPs, whereas the
other countrieshave 60-month CIEPs. The Januarydummy controls
for seasonal cycles in the data as PMs (purportedly)tend to schedule
major reshuffles during breaks (for example, over Christmas)in the
parliamentary calendar.19Thenumberof priorreshufflesis an additional
means to control for event dependencein the data.20
As for variables of substantive interest, we included dummy
variables identifying parties that use informal and caucus-based
leadershipselection systems and those that impose limits on cabinet
management.With all Canadiancases using convention-basedleader-
ship selection systems and with the theoreticaldistinctionthat we are
drawing between selection systems controlled by a few people and
those controlled by many, it makes sense to use this category as the
baseline. Note that the Australianand New Zealand Labourparties,
the partiesthat impose limits on the PM's managementof the cabinet,
also employcaucus-basedleadershipsystems.Thusthe dummyvariable
denoting limited prime ministerialcontrol of the cabinet also denotes
caucus-basedleadershipselection (thatis, it is the interactionof limits
and caucus leadership).This patternof covariationpreventsus from
estimatingthe effect of limits on cabinetmanagementfree fromcaucus-
based leadership.This is a weakness in our dataset,but as we have not
yet come across a party that places limits on cabinet management
without using caucus-basedleadershipselection, we do not see how
this problemcan be rectified.We also includeda dummyfor coalition
cabinets. The results are presentedas hazardratios ratherthan coeffi-
cients, thus they show the hazard rate proportionalto the baseline
category (convention-basedleadershipselection systems).
Institutional Determinants of Reshuffles
Models 1 and 2 test our institutionalhypotheses. There is little

evidence in Model 1 thatpartyrulesgoverningleadershipselection and
cabinet managementaffect the timing of reshuffles. Only the control
variablesare significant.Model 2 offers a more flexible specification,
allowingthe coefficientsfor caucus-basedleadershipselectionandlim-
ited cabinetmanagementto vary accordingto whetherthe reshuffleis
an initial or subsequentreshuffle.21Once this is done, we begin to see
institutionaleffects. The basic patternis of caucus-basedleadership
selection increasingthe hazardof an initial reshuffle but suppressing
the hazardof a subsequentreshuffle.Limitationson the PM's cabinet
managementability work against this trend, lowering the hazard of
TABLE 1
Conditional-Risks Models of Cabinet Reshuf
Covariates Model 1 Model 2 Model 3 M
Majority Status .66 .72 .61*

(-1.52) (-1.12) (-1.94)
Coalition Government .99 1.06 1.18
(-.04) (.19) (.43)
Caucus Leadership 1.09 2.17' 1.50
(.26) (1.74) (.78)
Caucus Leadership .32*** .26***
x Subsequent Reshuffle (-2.88) (-3.34) (
Limited Control & Caucus Leadership 1.32 .76 .84
(.75) (-.64) (-.34)
Limited Control & Caucus Leadership 2.67** 2.73**
x Subsequent Reshuffle (2.16) (1.99)
Informal Leadership Selection .83 .75 .32
(-.34) (-.56) (-1.65)
Party Popularity,t .98
(-1.57)
PM's Approval Rating,,
(
Senior Partner Popularity
- Junior Partner Popularity,_
E By-election Losses .87*
t-
(-1.94)
i
Ito Dissenting Divisions
TABLE 1
(continued)
Covariates Model 1 Model 2 Model 3 M
N Cabinet & Senior 1.03* 1.03 1.01

Noncabinet Ministers (1.91) (1.60) (.57) (
Months Left in CIEP 1.00 1.00 .98*
(-.37) (-.24) (-1.74) (
N Prior Reshuffles .88* .88 .88
(-1.69) (-1.47) (-1.33) (-1
January Dummy 1.56* 1.57* 1.58*
(1.83) (1.87) (1.75) (
United Kingdom .77 .91 2.49
(-.49) (-.18) (1.33) (-1
Australia .88 .87 1.12
(-.28) (-.31) (.18) (-1
New Zealand .77 .77 .88
(-.69) (-.70) (-.23) (
Ireland .50* .42** .53
(-1.82) (-2.18) (-1.22) (
Log Likelihood -630.2 -625.3 -478.5 -2
N Units (Spells) 237 237 188
N Failures (Reshuffles) 164 164 135
N Observations (Months) 2,596 2,596 1,838 1
Schoenfield Test of PH (X) 7.21 15.36 8.54
Note: Cell entries are hazard ratios computed using the Efron method of resolving ties. Figures in
standarderrors clustered by government. *p < .10; **p< .05; ***p< .01.
initialreshufflesand increasingthe hazardof laterreshuffles.But once

we calculate the net effects of the interactions,it is clear that cabinet
managementlimitations do not entirely offset the effects of caucus-
based leadership selection. The hazardratio for initial reshuffles of
cabinets in which caucus-basedleadershipsystems are used is 2.17
and declines to .70 (that is, to 70% of the ratio for convention-based
systems) for subsequentreshuffles.When cabinetmanagementlimita-
tions are also in effect, the hazardratio for initial reshuffles remains
high at 1.65 and declines only to 1.43 for subsequentreshuffles. The
overall effects come throughin Figures 1A and 1B, which graphthe
time until a single-partymajoritycabinet is reshuffled,contingenton
the leadershipselection and cabinetmanagementrules in effect. When
the cabinet is headed by a convention-selectedPM, the median time
until an initial reshuffle is 17 months, with another8 months until a
subsequent reshuffle. For cabinets headed by PMs selected by the
parliamentaryparty,these mediantimes are 11and 12 monthsfor initial
and subsequentreshuffles,respectively.When, in addition,these PMs
havelimitationsplacedon theirmanagementof the cabinet,thesemedian
times become 12 and 6 months,respectively.What standsout in these
results,then,is how limitationson the PM's abilityto managethe cabinet
greatly reduce the time between reshuffles.
Time-Varying Determinants of Reshuffles
Models 3 through6 test our time-varyinghypotheses.We began

by introducingtwo measuresof governmentpopularityto the model
(Model 3): the governing coalition's share of polled vote intentions,
which we term"popularity,"and the numberof by-election losses that
that governmenthad suffered to date. Both variableswere lagged to
avoid endogeneity.The popularitymeasureis straightforwardenough,
but the calculationof the by-election variablerequiresa brief explana-
tion. By-elections are an opportunity for voters to offer midterm
judgments on the government's performance (Mughan 1988). By-
elections tendto be called in batches,the governmentexpectingto lose
some (because they are in opposition strongholds)and win others.To
avoid overstating the government's unpopularity,we counted only
by-electionlosses in seatspreviouslyheld by the government,thatis, in
seats that the government could have been expected to win. Our
variable tracksthe sum of these losses over the course of the parlia-
mentarytermin a mannerthatproxiesa coalitionof minoritiesdynamic
(Mueller 1970).
FIGURE 1A
The Impactof LeadershipSelection and CabinetManagementRules
on the Timingof InitialCabinetReshuffles
1.00
-o
E 0.75
cn
> 0.50
0
z
Z
.0
_.
. 0.25
0
0.00
n 6 12 18 24 30 36
Inter-event Time (Months)
FIGURE 1B
The Impactof LeadershipSelection and CabinetManagementRules
on the Timingof SubsequentCabinetReshuffles
1.00
o
s -,Convention-based Leadership
-o - Caucus-basedLeadership
E 0.75-
\\ .... --,t--- Cabinet Management
i CLimited
0.
v
0.50 --?
o ---_ _
z
0
0
0.00 T ---- - ---- --
I0 0.25 -
0 00
0 6 12 18 24

The initialresultsdo not providestrongsupportfor the hypothesis

thatreshufflesbecome more likely as the governingparty's(or parties')
popularitydeclines. The hazardof a cabinet reshuffle does appearto
rise as governmentpopularitydeclines, but the effect is (marginally)
statisticallyinsignificant.As for by-elections, we expected mounting
by-election losses to induce reshuffling. Our statistics suggest the
opposite, however: reshuffles become less likely (that is, the hazard
ratiodeclines) as by-electionlosses rise (althoughthe substantiveeffect
is not large because governmentsin this sample tend to lose only one
seat per term on average). One might explain these weak and
counterintuitivefindingsin a numberof ways, butthe broadermessage
of these resultsis thatthe connectionbetween the overallpopularityof
the governmentand reshuffles is not straightforward.22 This finding
points to an avenue for futureresearch.
Model 4 adds the PM's (lagged) personal approvalrating. (We
dropped the informal leadership selection dummy because prime
ministerialapprovaltends not to be availableuntil the late 1970s, by
which time all partieshad adoptedformal leadershipselection rules.)
Primeministerialapprovalcombineswith the government'spopularity
in an interestingway. Governmentpopularityis positivelyrelatedto the
hazard of a reshuffle (the hazard ratio is greaterthan 1), but prime
ministerialapprovalis negatively relatedto the hazardof a reshuffle
(the hazard ratio is less than 1). This result appearsto suggest that
popularcabinetsaremorelikelyto be reshuffledthanunpopularcabinets,
a counterintuitivefinding because it is unclearwhy a PM would alter
an ostensiblysuccessfulcabinet.A morecoherentinterpretation emerges
if the marginaleffects of governmentpopularityandprimeministerial
approvalare consideredin tandem,namely, if we hypothesizethat, at
any given level of governmentpopularity,reshufflesbecomemorelikely
as the PM's approvaldeclines. (Supportingthis interpretationis the
fact that when prime ministerialapprovalis not in the model-as in
Model 3 andin otherspecificationsnot shownhere-government popu-
larityoperatesin the hypothesizeddirection,with decliningpopularity
sparking reshuffles.) In other words, when PMs become electoral
liabilities, that is, when theirpersonalpopularitybegins to lag behind
their government's,reshufflesbecome more likely.
The substantiveeffects of this relationshipare very strong and
areillustratedin Figures2A and2B. The figuresshow the survivaltime
(timeuntilreshuffle)of a single-partymajoritycabinetwith convention-
basedleadershipselection(theCanadianandBritishnorm)givenvarying
levels of primeministerialpopularity.Partypopularityis fixed at40%in
thesegraphs(thesamplemean),while primeministerialpopularityvaries
FIGURE2A
PrimeMinisterialPopularity
and the Timing of InitialCabinetReshuffles
1.00
10
E 0.75
a)
04
0.50
o
zZ
0
0I
0.25
0.00
0 6 12 18 24 30
FIGURE2B
PrimeMinisterialPopularity
and the Timing of SubsequentCabinetReshuffles
1.00
- - - - Unpopular PM
0
E 0.75 ----- - -- -- --- - Popular PM
> 0.50 \
- - --- - - - - - - - - - --_ - - - - - - - - - - - - _ - -
o
zZ
_,_ *
*.0
. 0.25
0.00 I
0 6 12 18
from 47.5% ("popularPM") to 32.5% ("unpopularPM"). This 15%

decline in primeministerialapprovalreducesthe mediansurvivaltime
to an initial reshuffle from 21 to 14 months, and the median survival
time to a subsequentreshuffle from 12 to 6 months. This is a robust
patternthat persists in every subsequent specification that includes
both governmentalpopularityandprimeministerialapproval.
We addeda fourthpopularityvariable,the popularitygapbetween
senior and (largest)junior coalition partners,to Model 5.23Our prior
expectations regarding this variable were not strong-we merely
expected that changes in it would be associated with changes in the
risk of a reshuffle.The resultsindicatea fairlystrongnegativerelation-
ship between the popularitygap between the senior coalition partner
and largestjunior partnerand the hazardof a reshuffle. The marginal
effects are illustratedin Figures 3A and 3B. Both figures depict the
time until a majoritycoalition cabinet (with caucus-basedleadership
selection) is reshuffledas the popularitygap between partnersshrinks
from 30% to 20%.24 This 10% shift in popularitybetween coalition
partnersreduces the median time to an initial reshuffle from 19 to 13
months, and the median time to a subsequentreshuffle from approxi-
mately 16 monthsto 12 months(grantingthatthe plateauin the survival
curvelimitsprecision).Any such shiftin the relativepopularityof senior
andjunior coalition partnersis due eitherto the seniorpartybecoming
less popularor thejuniorpartnerbecoming more popular(or both).As
PMs are almost always membersof the seniorcoalitionpartner(in our
datasetthey always are),what we observe is thatPMs tend to reshuffle
as theirintracoalitionpositionbecomes weaker.To ourknowledge,this
is the first demonstrationof an idea hinted at by Laver and Shepsle,
namely, that PMs might reshuffle the cabinet to "impound ... the
public opinionshock while avoidingthe necessity of an earlyelection"
(Laverand Shepsle 1998, 39, n. 12). WhetherPMs use these reshuffles
to shoreup theirparty'spositionwithinthe coalitionor to acknowledge
the junior partner'snewfound status cannot be discerned from our
aggregatedata,butit remainsan interestingquestionfor futureresearch.
The finalvariablewe addedto the modelwas the running(lagged)
sumof dissentingdivisions,thatis, divisionsin whichthe governingparty
(or coalition)failedto maintainperfectcohesion.(We do not have these
data for Irelandand so Model 6 excludes Ireland.)Backbenchdissent
on the parliamentaryfloor is one barometerof the parliamentaryparty's
satisfactionwith the cabinet'spolicies. The hypothesis is that the risk
of a PM reshufflingthe cabinetis positivelyrelatedto intrapartydissent,
but naturallywe do not see reshuffles being sparkedby an isolated
rebellion. Instead, we see rising levels of parliamentarydissent
FIGURE3A
The Relative Popularityof CoalitionPartners
and the Timingof InitialCabinetReshuffles
1.00
i 0.75
3
ca
a4
> 0.50
o
z
'_
o0
. 0.25
0
0.00
0 6 12 18 24 30
FIGURE3B
The Relative Popularityof CoalitionPartners
and the Timing of SubsequentCabinetReshuffles
1.00
E 0.75
a)
0
> 0.50
o
Z
o
0
0.25
0
0.00
0 6 12 18
indicatingthat an ever-growingproportionof the parliamentaryparty

opposesthePM.Thevariable'srunningsumconstructionhelpsto capture
this dynamic.The resultsvindicateour hypothesis:the hazardof a re-
shuffle increasesby approximately1%for every additionaldissenting
division. The substantiveeffect is largefor Britishgovernments,which
endure97 dissentingdivisionspertermon average,butmuchsmallerfor
AustralianandNew Zealandgovernments,whichexhibitvirtuallyperfect
cohesion,sufferingfewerthan5 dissentingdivisionspertermon average.
Determiningwhetheror not reshufflesbecome more likely as the
parliamentarytermprogressesis not as simple a matteras estimating
andinterpretingthe marginaleffects of the othervariablesin the model.
True,there is rarelya statisticallysignificant connectionbetween the
hazardof a reshuffle and the numberof months left in the CIEP,but
this lack of a trendtells us only thatthe mean level of the hazarddoes
not shift up or down over the course of the parliamentaryterm. The
hazardrate itself might still rise (or fall) over the course of the term, a
possibilitybest assessedby graphingthe baselinehazardrateover time.
Figure 4 presents the (smoothed) hazard rates for initial and
subsequentreshufflesgeneratedfromModel 3. The baseline hazardof
initialreshufflesrises noticeablyover time whereasthatof subsequent
reshuffles is fairly flat. Rising hazards signal a systematic decay of
cabinet stability over time, whereas constant (flat) hazards are
commonly interpretedas evidence thatfailureis due to randomevents
ratherthan systematicprocesses (Alt and King 1994; Warwick1994,
7-9).25 The steadily rising hazardof initial reshuffles indicates, then,
that reshuffles (initial ones, at least) become ever more likely as the
election nears, a pattern that is consonant with the argument that
reshuffles are elements of an electoral strategy. The flat hazard of
subsequentreshuffles suggests that laterreshuffles are not partof any
such electoral cycle. It would be in keeping with coalition theorists'
interpretationof constanthazardsto speculatethat the flat hazardrate
of subsequentreshuffles is evidence that they are sparkedor under-
takenin responseto randomlyoccurringpolitical events (for example,
leadershipchallenges,seriousscandals,andso on). Keep in mind,how-
ever, that every reshufflereducesthe hazardof a subsequentreshuffle
by 10 to 20% (dependingon the model). Thus, althoughthe pictureis
clear for initialreshuffles,which become more likely as time passes, it
is less so for subsequentreshuffles.
Fully exploringthe substantivemeaningof the baselinehazardsis
a matterfor anotherarticle,however.At this stage, it is most important
to note that neither of the baseline hazardsis declining, a signal that
unobservedheterogeneityis not an issue here.Ourmodels also conform
FIGURE4
HazardRates of Initial and SubsequentReshuffles
o Initial Reshuffles
- -- Subsequent Reshuffles
2-
ct +
+1- +.... +
-+
0-
0 6 12 18 24 30
to the proportional-hazards assumption (the assumption that the baseline

hazards are proportional in the covariates), although this is not true
when time is measured in elapsed time. The coefficients are neverthe-
less remarkably robust to this sort of change.
5. Conclusion
The central assumption of our theoretical model is that PMs are

driven by a desire to maintain power. Consequently, we see reshuffles
as tools that PMs use to help them attain this end, by undercutting the
activities of internal rivals for power (chiefly their ministers, but also
their coalition partners), and by improving their odds of winning
reelection. This framework leads us to predict that PMs will reshuffle
their cabinets as their intraparty, coalitional, and electoral positions
become more precarious.
Three main statistical results of our work bear out this prediction.
First, the hazard of a reshuffle is highest in cabinets where institutional
rules both limit the PM's management of the cabinet and place leader-
ship selectionunderthe controlof the parliamentaryparty.This finding

demonstratesthat reshuffles are most likely in situations where the
PM's agency loss to cabinetministersis theoreticallygreatest.Second,
the hazardof a reshuffle rises whenever parliamentaryand electoral
popularitydecline.Specifically,we show thatPMsreshuffleas intraparty
dissent increasesand the prime minister'sown approvalratingsbegin
to lag behind the government'spopularity.Third, we show that the
hazardof a reshuffle rises whenever the popularitygap between the
PM'spartyandthejuniorcoalitionpartnernarrows.The commontheme
here is that PMs reshuffle their cabinets whenever their intraparty,
parliamentary,or electoral positions deteriorateand the PMs become
identifiableto partymembers,coalitionpartners,and voters as political
liabilities.Theseconditionsinviteleadershipchallenges,threatsof coalition
termination,andelectoraldefeat,and,in turn,theyappearto elicitcabinet
reshuffles.Takentogether,these resultscomportmore closely with the
strategicinterpretation of reshufflesthatwe have put forwardthanwith
the argumentthatreshufflesare standardparliamentary operatingproce-
duresor technocraticdevicesusedto recruitpoliticaltalentintocabinet.
What do these results mean for the broaderstudy of parliamen-
tarypolitics?RobertMenzies,Australia'slongest-servingprimeminister,
is reputedto have remarkedto his ministersthat his cabinet's central
policy was to maintainMenzies, himself, as primeminister.Certainly,
the theory and results we have laid out here underscorethe central
positionof primeministers-and primeministerialsurvival-in parlia-
mentarypolitics. This is not to say thatwe espouse a presidentialview
of parliamentarypolitics or automaticallyassume that parliamentary
governmenthas given way to prime ministerial government(as per
Foley 2000 or Savoie 1999, for example). On the contrary,our model
does not place PMs above parliamentarypolitics but in their midst,
where PMs struggleto maintaincontrolof theircabinets,theirparties,
and their coalition partners,as well as theirreelection fortunes.
Ourresults,moreover,should not be reducedto the epigramthat
"weak PMs reshuffle, strong PMs don't." We do not take PMs to be
intrinsicallyweak or strong.Instead,we have arguedthatthe natureof
parliamentarygovernmentis such that there is tension between PMs
and cabinet ministers and coalition partners. Some institutional
configurationsand circumstances-constraints on the PM's ability to
managethe cabinet or the PM's personalunpopularity,for example-
exacerbatethis tension and invite challenges to the PM's authority.
Themorefrequentandseverethesechallenges,the strongerthe incentive
will be for the PM to reshufflethe cabinet.It is the incentiveto reshuffle
that is strongeror weaker,not the PM.
Our article also serves to open up a researchagenda on cabinet

reshuffles.A largenumberof issues remainto be explored.Several of
these strike us as particularlyprovocative. First, the link between
governmentpopularityand reshuffles is not straightforward.It is not
yet evident, for example, whether the effects of changing levels of
governmentpopularityon reshuffles are institutionallyor temporally
contingent. These sorts of relationshipsare importantto investigate
because they have the potentialto shed light on a second question,to
wit, do reshuffles make a difference? The historical anecdotes and
interview excerptsthatwe have presentedhighlightthe fact that PMs
use reshuffles to try to achieve some end: a desired policy, a resur-
gence of politicalpopularity,the sideliningof a rival, andthe like. Our
statistical evidence comports with this view but does not speak to
whether or not reshufflingworks as PMs intend. This efficacy is not
an easy matterto establish,yet it is a vital one because demonstrating
thatreshufflesalterelectoraloutcomesorpoliciesfurtherjustifiescabinet
reshufflesas a topicworthyof attention.The natureandplace of cabinet
reshufflesin coalitiongovernmentsis also worthfollowingup. Wehave
shown thatreshufflesbecome more likely as the seniorparty'sposition
in the coalition weakens. It is not clear, however, whether these
reshufflesare"defensive,"designedto appeasean increasinglypopular
junior partner,or "offensive," intendedto prevent the junior partner
fromtranslatinggrowingpoliticalpopularityinto cabinetpower.This is
an especially importantquestionbecauseit promisesto link the nascent
work on reshuffles to the far more establishedliteratureon coalition
formationandtermination.
To fully explore these, and otherrelated,issues demandsa larger
and more refineddatasetthat is, for now, beyond our reach.Thatsaid,
ourwork indicateshow importantit is to thinkabouthow institutional
rules and political conditions affect the strengthof the PM's position
withinthe governingcoalition.This emphasison intrapartypolitics and
on the pivotal role of the prime ministerstandsin contrastto much of
the existingworkon coalitionpolitics.Thus,althoughthereareparallels
(especially methodologicalones) betweenthe workwe have presented
here and the existing literatureon cabinet survival (for example, Alt
andKing 1994, Diermeierand Stevenson 1999, LupiaandStrom1995,
and Warwick 1994), there are enough differences to suggest that the
way forwardrequirescarefulthinkingaboutthe similaritiesanddiffer-
ences between reshuffles and other forms of cabinet instability.
Christopher Kam is Assistant Professor of Political Science,

University of British Columbia, 1866 Main Mall, Buchanan C472,
Vancouver,BC V6T 121, Canada. (This article was written while
Kam was on faculty at the University of South Carolina.) Indridi
Indridason is Assistant Professor of Political Science, University
of Iceland, 101 Reykjavik, Iceland.
NOTES
We thankBradGomez, ThomasHansford,CarlaHudson,andthreeanonymous
reviewers for their helpful comments.Many thanksto Joseph Wearing,Jack Vowles,
Michael Marsh,Gail McElroy,Philip Cowley, LyndaErickson,and Eric Belangerfor
contributingdata to our project.ChristopherKam also thanksthe WalkerInstituteof
InternationalAffairs at the University of South Carolinafor fundinghis fieldwork in
Ottawa.
1. There is a small literatureon ministers'tenure(e.g., Headey 1974 and Rose
1987). A few works examine reshufflesdirectly(e.g., Alt 1975, Dewan and Dowding
2002, Herman1975, HuberandMartinez-Gallardo 2001, and White2000), and Sayers
and Moon (1999) investigate the reorganizationof Australianministerialportfolios
over time. In contrast,Grofmanand van Roozendaal(1997) cite dozens of works on
portfolioallocationand cabinetsurvival.
2. Kam interviewed 10 Canadianparliamentarians,including 1 ministerand 3
former ministers, on the subject of cabinet reshuffles. The interviews took place in
Ottawa,May 14-26, 2003. All subsequentinterviewexcerptscome fromthese conver-
sations.
3. Take Britain as an example. In Britain, there are two grades of ministers
outside the cabinet, Ministers of State and Under-Secretariesof State. Ministers of
Stateare consideredseniorministersin theirown rightandareactuallyquitepowerful:
they areoften chargedwith implementingandadministeringpolicy, andmay have some
power to set policy in narrowareas. Under-Secretaries,in contrast,have no power
independentof their cabinetminister.
4. This was our minimum criterion for a reshuffle, and it rarely resulted in
ambiguitybecausemost reshufflesinvolved farmore extensive changes.
5. For example, one might argue that changes made to the incumbentcabinet
following a successful reelectionbid shouldbe consideredreshuffles,not installations
of wholly new cabinets.
6. A numberof historicalexamplesunderscorethe factthatPMs cannotcounton
ministersto follow ordersdutifully.Selwyn Lloyd, as we have seen, refusedto imple-
ment the expansionistfiscal policy that Macmillandesired.Rex Connor,Ministerof
Mineralsand Energyin Gough Whitlam'sAustralianLaborgovernment,engaged in a
covertscheme to raisecapitalto fundAustralianminingefforts,even afterWhitlamand
the cabinetforbadeConnorfromdoing so (Bolton 1996, 239; Edwards1996, 119-20)!
As Minister of Finance in Jack Lynch's Fianna Fail government,Charles Haughey
conspiredwith several otherministersto fund and transportarms to NorthernIreland
(Collins 2000). Haughey's clandestine actions ran directly against Lynch's
nonconfrontationalapproachto the "Troubles"in NorthernIreland(and verged on
being illegal, althoughHaugheywas acquitted)but were neverthelesspopularwith the

extraparliamentary party.MichaelHeseltine,MargaretThatcherherself argues,pushed
his own line on a Europeantakeoverof Britain'sWestlandhelicoptercompany,making
concertedefforts to precommitthe cabinet to his, not Thatcher's,preferredoutcome
(Thatcher1993, 423-32). Roger Douglas, PeterLange's FinanceMinisterin the New
Zealand Labour governments of the 1980s, refused to water down his staunch
neoconservativepolicies, eventuallyforcing a showdown with Lange that resultedin
both men leaving the cabinet (Boston and Holland 1987). In Canada,Paul Martin,
Chretien'sFinanceMinisterand chief rival,used his influence in the partyto push for
a formalreview of Chr6tien'sleadership,an opening gambitin a bid to force Chretien
from the premiership.In sum, whetherbecause ministersare incompetent,ideologi-
cally opposed to the PM, willfully intent on unseating the PM, or have become
departmentalciphers,PMs frequentlystruggleto maintaincontrol of their ministers.
7. The claim here is simply thatreshufflesare one (important)means of elimi-
nating the agency loss associatedwith cabinet government.There are other ways of
doing this, of course. Vanbergand Martin(2004), for example, arguethat legislative
mechanismsareused to monitorministers'actionsandaddressthe moralhazardpresent
in ministerialdelegation.
8. The firstconditionis more intuitivethanthe second;it merelyposits thatthe
cabinet's political fortunes are shared to some extent. For example, if a minister is
caughtin a scandal,thenthe assumptionis thatthe negative effects of the scandalpartly
adhereto his or hercabinetcolleagues.As an exampleof the second condition,consider
a ministerwhose overspendingin his or her portfolioenables successorsto overspend
to an even greaterextent, as would be the case if budgetingwere incremental.Alterna-
tively, one might imaginethata ministerwho gets away with overtlycriticizingthe PM
createsa precedentfor his or her successors.Thus,the second assumptionis that"bad"
ministerialbehaviorin the presentenableseven worse ministerialbehaviorin the future.
9. We are concernedhere only with the leadershipselection rules in the prime
minister's party because even if a coalition governmentis in place, only one party
controlsthepremiership.Of course,the existenceof a coalitionmay complicatematters,
but we deal with these issues laterin the article.
10. The list of PMs removedby theirpartiesis longer than one might think:In
the UnitedKingdom,MargaretThatcher;in Australia,JohnGortonandBob Hawke;in
New Zealand,Peter Lange, GeoffreyPalmer,and JamesBolger. (One might arguably
place Ireland'sAlbertReynoldsin this groupbecausehis resignationwas clearlydue to
the fact that he had lost the confidence of the Fianna Fail parliamentaryparty.)A
number of other PMs were formally challenged for the leadershipby one of their
ministers,includingMalcolmFraser,CharlesHaughey,andJohnMajor.Tellingly,all of
these PMs were pushed out or challengedby partiesin which the parliamentaryparty
alone voted on the leadership.In Canada,where leadershipselection is in the handsof
a partywideconvention,only Chretiencan be said to have faced a formal(but aborted)
challenge.Of the 46 PMs in oursample,then, 11 (24%)were formallychallengedby an
internalrival and 6 (13%)were removedfrom the premiershipas a result.
11. In Australia, for example, the National party vetoed the Liberal party's
selection of William McMahonas HaroldHolt's successor, allowing John Gortonto
emerge as a compromisechoice for the premiership(Davis 1998).
12. Indeed,Alt (1975) andWhite (2000) have investigatedthis hypothesisusing

Britishand Canadiandata,respectively.In both cases, the conclusionwas the same: if
there is a cyclical componentto reshuffles,it is weak. Alt notes, for example, a recent
(1950-70) tendency for shuffles to occur earlierin the parliamentarytermthan in the
past, but no more thanthat.One shouldnot be put off by these weak results;PMs may
use reshufflesboth as responsesto randomevents andas partof a systematicreelection
program.Until one controlsfor the impactof randomevents (andneitherAlt norWhite
did this), any systematicpatternin the timing of reshufflesis likely to remainobscure.
13. Parliamentarydissent,thatis, backbenchersvoting againsttheirown parties,
is a more frequentoccurrencethan one might think. Philip Norton (1975, 1980) and
Philip Cowley (2002) have documenteda surge of dissent in the British Parliament
fromthe early 1970s onward.The percentageof divisions witnessingintrapartydissent
on the governmentside rose from a 1945-70 averageof 5%to a post-1970 averageof
15%. Governmentdefeats increased in lockstep with this surge of dissent. British
governmentssuffered65 defeats between 1970 and 1979 (many due to internalrebel-
lion); the previous 25 years had witnessed only five (Norton 1985, 27). Dissent has
never been as threateningto the governmentin Canadaas in Britain,but it has been
nearly as frequent(Wearing1998). The 1993-97 Chretiengovernment,for example,
saw its backbenchersvote againstit on 16%of divisions. Dissent has been much less
frequentin New Zealand,Australia,and Ireland,but work by Lucy (1985), Hobby
(1987), Mitchell (1999), and Kam (2002) indicates that it has not been absent. Our
point, in any case, is not that PMs cannot get their legislative programs through
Parliament(they can), but ratherthat high levels of dissent signal that the policy
contentof the PM's programsis disliked.This disapprovalis significantbecause PMs
arenot presidents;theirpolitical success and survivaldependon maintainingthe active
supportof both their ministersand their backbenchers.
14. The 74 censoredspells reflect the fact thatthere are 74 governmentsin our
sample,each of which was censoredwhen it gave way to a new government(as per the
discussionin Section 1). Plannedelections, those for which the governmentostensibly
controlledthe timing of dissolution,were responsiblefor most instancesof censoring
(57 of 74 cases).Another11cases were censoredbecauseearlyelectionswere called by
or forced on minoritygovernments.Of the remaining6 cases of censoring,3 were due
to coalitionfailure(i.e., a coalitionpartnerleavingthe coalition),1 to a minoritygovern-
ment being displaced by a newly formed coalition during the term (the short-lived
Reynolds governmentin Ireland,which was pushed out by the Rainbow coalition of
Fine Gael, Labour,and ProgressiveDemocrats),1 to the dismissal of the government
by the Head of State(the 1975 Whitlamgovernmentin Australia),and 1 to a caretaker
governmentcalling elections (the Frasergovernment,which assumedpower on Gough
Whitlam'sdismissal).
15. We were carefulto code exceptionsto this rule. For example,the Australian
Labor Party was so thrilled by Paul Keating's 1995 election victory that the party
departedfromits usualpracticeandallowedKeatinga freehandin selectinghis ministers.
16. This section is heavily influencedby Box-Steffensmeierand Zorn 2002.
17. Practicallyspeaking,this methodmeans that there are as many recordsper
unit as time periods(months,in this case) understudy,but recordsfor the nth reshuffle
enterthe risk set only afterthe unit (i.e., the governmentin question)has undergoneits
nth-1 reshuffle.
18. Clearly,monthlyobservationsof the same governmentare not independent;

this is why the standarderrorsareclusteredby government.(Standarderrorsmight also
be clusteredby primeminister,butthis possibilitymakeslittle differenceto the results.)
What is perhaps less obvious is that the occurrenceof earlier events may alter the
hazardof laterevents. The best way of dealingwith event dependenceis to stratifythe
model by event orderso as to allow the estimationof wholly separatehazardsfor each
stratum.Event-by-covariateinteractionscan then be used to avoid the restrictionof
identicalcoefficients across strata.Of course, adoptingthis approachmeans thatsome
grouping of higher-orderevents is requiredbecause the risk set shrinkswith every
reshuffle. The most obvious and logical distinction to make in this application is
between initial and subsequentreshuffles:there can only be one initial reshuffleof a
cabinet, but possibly many subsequentreshuffles. (Stratificationof the first, second,
third,and fourthand higherreshufflesreturnedsubstantiallysimilarresults.)
19. JulyandAugustdummies(for the summerbreak)had no statisticallysignifi-
cant effects.
20. The inclusion of a prior-event-countvariableallows the hazardfunctionto
shift up or down (as the case may be) with every additionalreshuffle.This is a very
rough means of dealing with event dependencethat should not be used in place of
stratification(Box-Steffensmeierand Zorn2002, 1086). When used alongsidestratifi-
cation, however, a prior-event-countvariable can help to discern differences in the
hazardsof higher-orderevents that have been groupedtogether, in this instance, of
second andhigher-orderreshuffles.
21. We ran models in which all institutionalvariableswere interactedby event
order, but only the coefficients for caucus-based leadership systems and cabinet
managementlimitationsshowed any statisticalsignificance.
22. The popularityhypothesiscan be operationalizedin a numberof ways. For
example, one might use monthlychanges in popularityand examinethe propensityof
PMs to reshuffle given drops in popularity.A more elaboratehypothesis is that PMs
aremore likely to reshufflewhen the partyexperiencesa largedropin popularitywhile
at a low level of popularitythanwhen the partyexperiencesa small dropat a high level
of popularity.Similarly,one might combine our electoralcycle andpopularityhypoth-
eses and arguethat late-termshifts in popularityhave a greaterimpacton the risks of
a reshufflethando early-termshifts,which areso distantfromthe electionthatthey are
of little concernto the PM. We exploredthese hypothesesand some of them do have
traction.For simplicity, however, we have focused on the relative power of institu-
tional rules and changingpolitical conditionsin sparkingreshuffles.We ranadditional
models thatexploredthe relationshipbetweenreshufflesandchangingeconomiccondi-
tions (e.g., unemployment,inflation,exchange rate volatility,and strikeactivity), but
these resultswere so weak that we did not think it efficient to include them here.
23. For coalitionscomprisingthreeor moreparties,we measuredthis popularity
gap variableas the differencebetween the senior(or major)party'spopularityand the
most popular(minoror junior) coalition partner'spopularity.In our dataset,the most
popularjuniorcoalitionpartnerwas alwaysthe second-largestpartyin the coalition(after
the seniorpartner).This need not always be the case, of course,and futureresearchers
workingwithbroaderdatasetswill have to decide whetherthe popularitygap shouldbe
measuredas the differencein popularitybetweenthe two largest coalitionpartiesor as
the difference in popularitybetween the two most popular coalition parties.
24. The mean popularitygap for coalitiongovernmentsin our sample is 26.5%

with an interquartilerange of 12.3%. Of the 74 governmentsin our sample, 20 are
coalitions.
25. We note thatthis is an interpretationof constanthazards.Strictlyspeaking,
a constanthazardindicatesonly that the hazardis independentof time.
REFERENCES
1998a. "Britain'sNew Cabinet."TheEconomist,30 July.

1998b. "Intothe Reshuffle Season." TheEconomist,7 May.
1999. "TheCabinetReshuffle:Blair to Move and Win."TheEconomist, 14 October.
Alt, James. 1975. "Continuity,Experience,andTurnoverin the BritishCabinet,1868-
1970." In Cabinet Studies. A Reader, ed. Valentine Herman and James Alt.
Basingstoke:Macmillan.
Alt, JamesE., and GaryKing. 1994. "Transfersof GovernmentalPower:The Meaning
of Time Dependence."ComparativePolitical Studies27: 190-210.
Bolton, Geoffrey. 1996. The OxfordHistory of Australia, Volume5. 1942-95. The
Middle Way.2d ed. Melbourne:OxfordUniversityPress.
Boston, Jonathan,andMartinHolland. 1987. TheFourthLabourGovernment.Radical
Politics in New Zealand.New York:OxfordUniversityPress.
Box-Steffensmeier,Janet,andChristopherZorn.2002. "DurationModels for Repeated
Events."Journal of Politics 64: 1069-94.
Bueno de Mesquita, Bruce, James D. Morrow, Randolf M. Siverson, and Alastair
Smith.2002. "PoliticalInstitutions,Policy Choice,andthe Survivalof Leaders."
BritishJournalof Political Science 32: 559-90.
Butler,David, andGarethButler.1994. BritishPolitical Facts, 1990-1994. New York:
St. Martin's.
Clarke, Harold D., Karl Ho, and MarianneC. Stewart.2000. "Major'sLesser (Not
Minor) Effects: Prime MinisterialApproval and Governing Party Supportin
Britainsince 1979."Electoral Studies 19: 255-73.
Collins, Stephen.2000. ThePower Game:Fianna Fail since Lemass.Dublin:O'Brien.
Cowley,Philip.2002. RevoltsandRebellions:ParliamentaryVotingunderBlair.London:
Politico's.
Cox, GaryW.1987. TheEfficientSecret:TheCabinetand theDevelopmentofParties in
VictorianEngland.Cambridge:CambridgeUniversityPress.
Davis, James. 1998. LeadershipSelection in Six WesternDemocracies. Westport,CT:
Greenwood.
Dewan, Torun,andKeith M. Dowding. 2002. "HeHas to Go! Analyzingthe Effects of
MinisterialResignationson GovernmentPopularity."Presentedat the annual
meeting of the AmericanPolitical Science Association,Boston.
Diermeier,Daniel, and Randall Stevenson. 1999. "CabinetSurvival and Competing
Risks."AmericanJournalof Political Science 43: 1051-68.
Edwards,John. 1996. Keating: TheInside Story.New York:Viking.
FitzGerald,G.A. 1991. All in a Life:An Autobiography.Dublin:Gill and Macmillan.
Foley, M. 2000. The British Presidency. Manchester:ManchesterUniversity Press.
Franks,C.E.S. 1987. TheParliamentof Canada.Toronto:Universityof TorontoPress.
Gomez, BradT., andJ. MatthewWilson.2001. "PoliticalSophisticationandEconomic

Voting in the American Electorate:A Theory of HeterogeneousAttribution."
AmericanJournalof Political Science 45: 899-914.
Grofman,Bernard,and Peter van Roozendaal. 1997. "ModellingCabinetDurability
andTermination." BritishJournalof Political Science 27: 419-51.
Headey, Bruce. 1974. British CabinetMinisters. TheRoles of Politicians in Executive
Office.London:Routledge.
Herman,Valentine.1975."Comparative Perspectiveson MinisterialStabilityin Britain."
In CabinetStudies.A Reader,ed. ValentineHermanandJamesAlt. Basingstoke:
Macmillan.
Hobby, Michael. 1987. "The Crack of the Whips: Party Cohesion and Institutional
Consensusin the New ZealandParliament,1936-85." Master'sthesis. Univer-
sity of Canterbury.
Huber, John D. 1996. RationalizingParliament: Legislative Institutionsand Party
Politics in France. New York:CambridgeUniversityPress.
Huber,JohnD., andCelia Martinez-Gallardo. 2001. "CabinetInstabilityandthe Accu-
mulationof Experienceby CabinetMinisters:The FourthandFifthRepublicsin
ComparativePerspective."Presentedat the annual meeting of the American
PoliticalScienceAssociation,SanFrancisco.
Indri6ason,Indri6i,andChristopherKam.2003. "CabinetReshufflesandBureaucratic
Capture."Paperpresentedatthe annualmeetingof the MidwestPoliticalScience
Association,Chicago.
Kam, Christopher.2002. "Parliaments,Parties,and MPs: A ComparativePerspective
on BackbenchDissent, PartyDiscipline, and Intra-partyPolitics." Ph.D. diss.
University of Rochester.
Katz, RichardS., and PeterMair. 1995. "ChangingModels of PartyOrganizationand
PartyDemocracy."Party Politics 1: 5-28.
King, Gary,JamesAlt, Nancy Bums, and Michael Laver. 1990. "A Unified Model of
CabinetDissolutionin ParliamentaryDemocracies."AmericanJournalofPolitical
Science 34: 846-71.
Laver,Michael,andNormanSchofield. 1990. MultipartyGovernment.ThePolitics of
Coalition in Europe.New York:OxfordUniversityPress.
Laver, Michael, and Kenneth Shepsle. 1996. Making and Breaking Governments:
CabinetsandLegislaturesin ParliamentaryDemocracies.New York:Cambridge
University Press.
Laver, Michael, and Kenneth Shepsle. 1998. "Events, Equilibria,and Government
Survival."AmericanJournalofPolitical Science 42: 28-54.
Leubberts,Gregory. 1986. ComparativeDemocracy. Policy Makingand Governing
Coalitions in Europeand Israel. New York:ColumbiaUniversityPress.
Lucy, Richard.1985. TheAustralianForm of Government.Melbourne:Macmillan.
Lupia,Arthur,andKaareStrom.1995. "CoalitionTerminationandthe StrategicTiming
of ParliamentaryElections."AmericanPolitical Science Review89: 648-65.
Macmillan,Harold.1963. At the End of the Day, 1961-63. London:Macmillan.
Martin,Lanny.2000. "PublicOpinionShocksandGovernmentTermination." Working
Paper,Departmentof Political Science, FloridaState University.
Mershon,Carol. 1999. "The Cost of Coalition:A Five Nation Comparison."In Party

Discipline and ParliamentaryGovernment,ed. Shawn Bowler, David Farrell,
and RichardKatz. Columbus:Ohio University Press.
Mitchell, Paul. 1999. "CoalitionDiscipline, EnforcementMechanisms,and Intraparty
Politics."InPartyDisciplineandParliamentaryGovernment,ed. ShawnBowler,
David Farrell,and RichardKatz. Columbus:Ohio UniversityPress.
Mueller,John E. 1970. "PresidentialPopularityfrom Trumanto Johnson."American
Political ScienceReview64: 18-34.
Mughan, Anthony. 1988. "On the By-Election Vote of Governments in Britain."
LegislativeStudies Quarterly13: 29-48.
Muller,WolfgangC., andKaareStrom. 1999. "PoliticalPartiesand HardChoices."In
Policy, Office, or Votes?How Political Parties in WesternEuropeMakeHard
Choices, ed. WolfgangC. Mullerand KaareStrom.New York:CambridgeUni-
versity Press.
Norton,Philip. 1975. Dissension in the House of Commons:Intra-PartyDissent in the
House of Commons'Division Lobbies, 1945-1974. London:Macmillan.
Norton, Philip. 1980. Dissension in the House of Commons,1974-1979. New York:
Oxford University Press.
Norton, Philip. 1985. "BehaviouralChanges.BackbenchIndependencein the 1980s."
In Parliamentin the 1980s, ed. PhilipNorton.Oxford:Basil Blackwell.
Rose, Richard.1987. Ministersand Ministries.Oxford:OxfordUniversityPress.
Savoie, Donald. 1999. Governingfromthe Centre.Toronto:Universityof TorontoPress.
Sayers, Anthony,and JeremyMoon. 1999. "The Dynamics of GovernmentActivity:
A Long-RunAnalysisof the ChangingScope andProfileof AustralianMinisterial
Portfolios."AustralianJournal of Political Science 34: 149-67.
Strom,Kaare, and Wolfgang C. Muller.2000. "The Keys to Togetherness:Coalition
Agreementsin Parliamentary Democracies."In TheUneasyRelationshipsbetween
ParliamentaryMembersand Leaders,ed. LawrenceD. Longley andReuvenY.
Hazan. London:FrankCass.
Strom,Kaare,and StephenM. Swindle. 2002. "StrategicParliamentaryDissolution."
AmericanPolitical ScienceReview96: 575-91.
Thatcher,Margaret.1993. The Downing Street Years,1979-90. New York: Harper
Perennial.
Thies,Michael.2001. "KeepingTabson Partners:The Logic of Delegationin Coalition
Governments."AmericanJournalof Political Science 45: 580-98.
Vanberg,Georg,andLannyMartin.2004. "Policingthe Bargain:CoalitionGovernment
andParliamentaryScrutiny."AmericanJournalof Political Science 48: 13-27.
Warwick,Paul. 1994. GovernmentSurvivalin ParliamentaryDemocracies.Cambridge:
CambridgeUniversity Press.
Wearing,Joseph. 1998. "Guns, Gays, and Gadflies: Party Dissent in the House of
Commons underMulroneyand Chretien."Presentedat the annualmeeting of
the CanadianPolitical Science Association, University of Ottawa.
Weller,Patrick. 1994. "PartyRules and the Dismissal of PMs: ComparativePerspec-
tives from Britain,Canada,andAustralia."ParliamentaryAffairs47: 133-43.
White, Graham.2000. "MinisterialMusical Chairs: Patternsof Provincial Cabinet
Shuffles." Presentedat the annual meeting of the CanadianPolitical Science
Association, Quebec.

On (2005) Timing and Durability of Cabinets

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

On (2005) Timing and Durability of Cabinets

Uploaded by

Copyright:

Available Formats

The Timing of Cabinet Reshuffles in Five Westminster Parliamentary Systems

Author(s): Christopher Kam and Indriđi Indriđason

Despite theirpolitical prominence,cabinetreshuffleshave not attracteda great

A cabinetreshuffleinvolves the promotion(or demotion)of

LEGISLATIVESTUDIES QUARTERLY,XXX, 3,August 2005 327

demonstratingthat parliamentarypolitics are manifestly strategic in

Our model generatesa series of hypotheses about the timing of

1. Defining Cabinet Shuffles

The Extent of Cabinet Change

We follow White (2000) and define reshuffles as any change in

Q: Why does the PM not simply change one underperformer

A: Well, let me give you an example. In our first Parliament,

A variety of constraintsmake it difficult, then, for PMs to respondto

Alt (1975) adds a temporaldimensionto reshuffles,arguingthat

The Organizational Level

The titles of the variousministerialoffices vary by country,but it

2. The Causes of Cabinet Reshuffles

Cabinet Ministers, PMs, and Reshuffles

Underpinningourmodel of reshufflesaretwo assumptionsabout

PMs can use reshuffles to address these problems.7 Quite

The Relationship between

leadershipselection.9Rules governing leadershipselection influence

Moreover,many backbenchmemberswill not have the PM's interests

Cabinet Management Rules, Coalition Government,

Rules on leadershipselection primarilyaffect the moral hazard

than do PMs who have complete authorityover the composition of

Hypothesis1: Reshufflesaremorelikelyif the PM is selected

Hypothesis2. Reshufflesaremorelikelyif the PM'sabilityto

Hypothesis 3: The effects of cabinet management and

Hypothesis4. Reshufflesare less likely if the cabinetis a

Changing Political Conditions and Reshuffles

So far we have connected the PM's propensityto reshuffle to a

Reshuffles, Electoral Cycles, and Political Popularity

2002). It is reasonableto hypothesize(asjournalistsoften do) thatPMs

A: Abouttwo to threeyearsin, by aboutthe midpointof the

Q: Now somebodytoldme thatthesemid-to late-termshuffles

A: Well,they'rewrong.Politicalpartiesarein the businessof

This cabinet minister suggests that reshuffles are part of an electoral

political popularityis due to scandalousministerialbehavior,then the

Do PMsreshufflein responseto declininglevels of partypopularity

Reshuffles and Coalition Politics

The relationshipbetween coalition partnersoccurs at the inter-

Reshuffles and Parliamentary Politics

Parliamentaryevents may also induce PMs to reshuffle their

Hypothesis 5: Reshuffles become more likely as the

Hypothesis 6: Reshuffles become more likely as the governing

Hypothesis7. Reshufflesbecome more likely as the PM's

Hypothesis9. Reshufflesbecomemorelikelyas parliamentary

The Cost of Reshuffles

Ourargumentis not intendedto give the impressionthatreshuffles

3. Data and Methods

Our data come from five countries: Britain, Ireland, Canada,

concentratedscope is also reflectiveof the fact thatwe triedto improve

We tested ourhypotheseswith a variantof the Cox regression,or

reshuffledat a given instantin time, conditionalon not yet havingbeen

Table 1 presentsour results.In additionto countrydummies,the

Institutional Determinants of Reshuffles

Models 1 and 2 test our institutionalhypotheses. There is little

Covariates Model 1 Model 2 Model 3 M

Majority Status .66 .72 .61*

N Cabinet & Senior 1.03* 1.03 1.01

initialreshufflesand increasingthe hazardof laterreshuffles.But once