Vous êtes sur la page 1sur 14

PARLIAMENTARY ELECTIONS AND POLITICAL PARTIES

JayeshKumar
We discusshere the probability of incumbentpart.\"swinning the electionfrom the sameconstituency..Weest-imate
theprobabilitl, of winning by the ruling parl,, irrespectiveof whetherthatparb,forms
a coalition during election,has been ct memberof the last governmentin the centre. We istimate the
averageprobabilin' of winning in India's generalelectionsfrom 1971 to 1999,under the given electoral
system,using a panel data Binary ChoiceModel.
India having the largestelectoralmandateand woefully underrepresented
in elected office, and
multipartyelectoralcollegein theworld is themost female candidatesare few. However. women are
suitablecandidatefor analysingelections.Indiahas now competingfiom almost every constituency.
the largebaseof politicalparties(over670) and,as and in someconstituencies
the numberof women
politicalpartiesareindispensable
fbr thesuccessful contestants
rivals men's. Ruling partiesand their
working of any democracy,they offer a lot of choice supportersswept almost every electoral contest
to voters.This article tries to explainthe behaviour during 1950-90, only a handful of times have
of electorates.We try to model the chancesof an oppositioncandidateswon a plurality in national
incumbentgetting re-electedin the election.In elections. Several factors explain the poor perparticular,we attemptto answerwhetherthe sex of formance.The non-Congress
oppositionlackedthe
candidatesaffects the outcome of any election? grassrootsupportuntill
the 1980s.Therulingparties
Doesthe identityof a party in termsof nationalys. enjoy extensive patronagenetworks and have
regionalmatter'l
superioraccessto the national/regional
media.The
'winner-takes-all'
system,used in most constiluAt the heart of any electoraldemocracyis the e n c i e sb. e n e f , i n
t c u m b e npt a r t i e s .
i d e ao f ' o n e p e r s o no, n e v o t e ,a n do n e v a l u e ' ,i . e . ,
anyone'svote shouldnot havehigherweightthan
In parliamentary systems with proportional
that of another.In India, the discrepancy
between representationand multiple parties, the prime
electoralconstituencies
is enormous.The largest ministerand hisftrercabinetare dependenton the
constituency
in 1999,forexample,Outer-Delhi,
had supportand confidenceof more than one parlianearly86 timesmore votersthanLakshadweep,
the mentary party. Such coalition governmentsare
smallestconstituency.It shows that the vote of formed anew after each election, in which voters
Lakshadweep's
votershasmoreweightthanthatof choosepartiesand not coalitionsor governments.
the Outer-Delhi'sin choosingtheirrepresentativesCoalitionaloptionsarediscussed
duringtheelection
in Parliament.Multipartyelectionsfor Parliament, campaignsbut not all optionsturn out to be feasible
state legislature,and local legislatureare a key oncetheelections-returns
becomeknown.Contrary
vehiclefbr the expressionof ideologicaldiversity. to two-partyparliamentarysystems,theelectorate's
Elections also provide a forum for advancesin signalsfor a new government
areambiguous,
giving
women'spoliticalparticipation.Women havehad the party leaderssome leeway in coalition barthe right fbrever,but the recenttrendstoward more gaining.In order to fulfil most of therepromises.
pluralist electionshave made their participation partieshavenot only to entera winning coalition,
somewhatmore mcanineful. Indian women arc but they have to reacha coalitionagreementthat

J a y e s hK u m a r i s R e s e a r c h
S c h o l a r I, n d i r a G a n d h iI n s t i t u t eo f D e v e l o p m e nRt e s e a r c M
h u m b a i .e M a i l : j a y e s h @ i g i d r . a c . i n
JEL Classification:C25. N45. Keywords:coalition, electoralsystem,and ruling party.
This paper is a modified version of the paper presentedat the 9th IntemationalConferenceof Forum for Interdisciplinar.y
Mathematicson Statistics.Combinatoricsand RelatedAreas(SCRA-FIM-lX) at Allahabad,December2l-23.2002. Author'thank.s
all panicipantsfor their comments.In additionmany thanksareextendedto KausikChaudhuriforhelpful commenrsandsuggesrrons.
Specialthanks are due to the Journal'sEditor and an anonymousrefereefor detailedcommentsand suggestions.whiih helped
s u b s t a n t i a l liyn i m p r o v i n gt h e p r e s e n t a t i o nT.h e u s u a ld i s c l a i r n ear p p l i e s .

JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY

JULY.SEPT',
2003

puts the promised policies on the government thatarisein thespecification


of themodelin Section
agenda.In this context, parliamentaryelections 4, lbllowed by the empiricalresulrsin Section5.
providean importantexperimentalsettingto study with discussionon some interestingimplications.
the voter'schoicebehaviourin choosingtheir rep- Section6 concludesthe work.
resenlativesin Parliamentfor the functioningof
democraticinstitutions.Moreover, in segmented LiteratLrreReview
societieslike India developmentscertain in the
1990shave changedthe context in which they
In this study, we model the winnings o1'an
function.Performance
of democraticinstitutionsis incumbentparty.We assumethatall thecandidates
substantiallya function of networksof trust and areequallylikelyto beelected(i.e.,thereis a random
civic engagements
amongcitizensby participating selectionprocedureof candidates),
and the el'fects
in coordinated
collcctiveaction.For anydemocracy on outcomeare not individual specificbut party
with sucha largeelectoralmandate,electionscan specific(Whichis nota veryrestrictiveassumption,
hardly be over-emphasised.
Electionsand electoral as most of the electoralresults are driven by the
analysishave alwaysbeenof high priority in any incumbent's party rather than the incumbent
democraticcountry.In this context,politicalecon- themselves,
barringsomeexceptions.).We estimate
omics with public choioe has a lot to offer rn the probabilityof winning of the incumbentparty
increasingour understanding
of the voter'schorce (i.e.,theparty,whichhasalsowon thelastelection).
behaviourin elections.In the arenaof multi party irrespectiveof whetherit has been a memberof'
elections,the presenceof electoralalliancesis the coalition (during/afierelections),the governin_e
mainstayo1'democraticgovernance.
partyany.or one of the membersof the governing
party,in caseof a coalitiongovernment.
For, we are
In orderto testfbr the basicsof the presenceof interestedin taking accountof constituency-wise
voter's choicc behaviour, we flrsl look at the ruling (winning)by partiesratherthantheruling at
probabilityof an incumbentgettingre-elected
from the center.We estimatethe averageprobabilityof
the same constituency.Further,in this study,we winning, given that the party has won the last
make an attempl at examining the behavioural election,usingconstituency-wise
dataof thegener-al
patternfollowed by otherelectoralvariablesspan- eleclionsfiom 1967 to 1999 under rhe Indian
ning the electoralterms. By meansof extensive electoralsystem,usinga PanelDataBinaryChoice
empiricalanalysis,the studyexploresfirst,whether Model (Generalised
EstimationEquationsPopulathe nationalpartieshave betterodds of getting re- tion AveragedModel). This exercisegives us the
elected than the regional parties or vice-versa. ideaof theincumbent's
performance
andits chances
Second,does the sex of candidatesmake any to comelnto poweragain,underthe assumption
of
difference in electoral outcome? Third, does rationalvotersin the constituency(whrch is very
increasedvoter-turnoutin any electionmeananti- widely used in existingpolitical economyliteraincumbencyvotingbehaviour?
Fourth,doesa large ture).
numberof candidatescut on the vote shareof the
rncumbent,
resultingin losingthe election?
The probabilitythat a party wins the elecrion
dependson variousfactors,viz., the stateto which
The remaindero1'thestudyis organisedas fbl- the constituency
belongs,the inherentcharacterislows: We discussthe existingresearchin this area tics of the population,and various other socioin Section2, followedby theIndianscenarioandits economicandculturalfactors(caste,religion.etc.).
electoralsystemin Section3. We proposea simple One of the potential factors that might aff'ectthe
econometric
modelandexploreeconometricissues winning of a party is that the sameparty has won

vot.. t5 No. 3

P ARLIAM ENTA RY ELECTI ON S AN D P OL|TI CAL PA RTIES

the last election.This seemsquite a probablecase


and India, with its socio-politicalset-up,provides
anincentiveto testthehypothesis
thathavingaccess
to power aff-ects
the outcomeof electionsin future
or doesnot.

elite surveydata fiom 1979.They arguethat maximisationof expectedvote, in general,is not a


rational party strategy in multiparty political
systems,which are basedon proportionalreprcsentatlon.

The developmentof methodsto examinemulResearchers


have done significantamount of
work in this areamostly with data the US or from tipartyelectionshasclearlycaughtthe attentionof
judgingby the numberof
the Europeancountries.Katz, Honaker,and King theprofession/discipline,
papers
recent
on
the
topic
following the initial Karz
improve
upon
their
original
model,IKatzand
[2001]
K i n g , 1 9 9 9P
, p . l 5 - 3 2 1i n b i a s ,v a r i a n c en,u m e r i c a l and King [999, Pp. 15-32] paper.Tomz, et ol,
stability, and computationalspeed, making it [2002, Pp. 66-83], make importanrsrepsin this
quickerto estimate.the dif{'erence
herebeing that direction,astheirmodelis basedon standardlinear
yet convenient
the problem of ef1'ective
voting, which is treatedas model and offers a sophisticated
alternative
that
involves
seemingly
unrelated
missingdatain the originalmodel.They arguethat
(SUR).
regression
Gelman
and
King
[1990, 1994,
ordinary least squares(OLS) regressionis inapPp.5
l4-541
analyse
elections
in
all
state
legislatures
propriate when, the dependentvariable measures
elected
from
a
single
member
district
over
theperiod
the share of the vote going to each party, and
1968-88
in
the
US
and
conclude
thar
redistricting
recommenda superiortechnique.Unfbrtunately,
increases
electoralresponsiveness,
and it leadsto
the Katz-King model requiresa high level of stalessbiastowardsa majorityof votersthanwouldbe
tisticalexpertise,and is computationallyimpractiexpected
ifno redistricting
hadoccurred.Beforethe
cal fbr more than three political parties,as thcir
redistricting,in case of large legislatures,the
modelis basedon the principleof the two politrcal
representationof voters was biasedtowards powpartiescompetin-{
l'ortheconstituency
seat.Jackson
erful representatives,
which leadsto unequaldrs[2001]triesto give a simplerestimationprocedure,
power
tribution of
among voters.Katz and King
using a seemingly unrelatedregression(SUR)
I 999,Pp. 15-32],King, eral, [200l, Pp.49 -69]and
I
model(with a multinomiallogisticfunction).
GelmanandKatz [200] I developmethodologies
to
estimatethe 'effbctive'vote in partiallycontested
Thereare somerelativelyrecentstudiesthat go (re-elections),
multipartyelections.They trearparbeyond the examinationof particularcases,and
tially contested
electionsas missingdataprobleni.
assessthe impact on national elections.Conley and use
an algorithm procedure to recover
[2002],usingthe 1990scensusandelectoraldataof contesting
parties'estimatedvote shares.
We don't
the Canadianridings,finds significantincumbency seeany reasonto expectthat,had party
a
contested
effects for members of parliament(MPs), and no elections,
otherparties,voteshareswouldhavebeen
evidenceof any significantincreasein incumbency changedbecause,
ifa partyis not contesting
fiom a
advantagein the last decade.Pappi and Gabriele particularconstituency,it must be the case that
[1998.Pp. 229-55],using surveydatafor the East eitherthe partyis contestingthe electionindirectly
andWestGermany,proposeshow partypreferences (maybe a coalition),or in theparty'sassessment,
it
of voters in a multiparty system are measured, does not have any chanceof winning from that
compared,
andaggregated.
Normal,etal, 11998,Pp. constituency.
Merrill [ 1994]providesevidenoethat
251-93) considereda stochasticmodel of vorer both the pure proximity and the pure directional
response,
basedon multinomialprobit estimation, models fall short of accountingfbr the spatial
using Euro-barometersurvey data and European distributionof votersby party. Gelmanand Katz

JOURNALOF INDIANSCHOOLOF POLITICALECONOMY

[2001], gives the strengthof voting, giving us an


insightof the inter-linkages
betweenvoting,coalitions and the electoralcollege.However,no work
hasso far beendone in greaterdetail to explain the
winnings of any party basedon their base and
performancein the constituency-ruling
ratherthan
thatin the ruhng at center.

JULY.SEPT.2OO.]

of the corresponding
upper housesof the states.
Therearealsothe (non-executive)Presidentandthe
Vice-President,electedby the membersof Parliamentand statelegislativeassemblies.

Generalelectionsareheldonceeverylive years,
butthePresident
may dissolvetheLok Sabhaon the
adviceof theprimeministerbefbreits termis over.
Though survey researchhas enormousadvan- a si n 1 9 7 1a n di n 1 9 9 7o r , i f h e o r s h ei s c o n v i n c e d
tagesfor studyingindividual-levelpref'erences,
yet thatno stablegovernment
canbe formed,asin 199I
as analyses of random selections of isolated and 1998.The prime ministerholdsoffice as long
individualsfrom unknowngeographical
locations, ashe or shecancommanda majorityin Lok Sabha.
theynecessarily
missmuch of electoralpolitics.As All the successivegovernmentsof the Congress
such,theyareoftenbestcomplemented
with studres party,which ruled India continuouslyuntil 1977.
of aggregatc
electoralreturns.
servcda full term in office. Since 1977,governmentshavebeenlessstable,anda numberof prime
Indian Scenario
ministershavehadto resign,asaresultof partysplits
or no-confidence
votesbeforecomoletinetheirfull
The electoralsystemoflndia is largelybasedon term.
theBritish pattern.The Constitutiondid not provide
any detailsabout the electoralsystem,and left it to
The major effect of the electoralsystem,at least
Parliamentto determine.Accordingly,Parliament until 1977,was to guarantee
majoritygovernments
passeda numberof laws to regulatethe electoral basedon a majority of voter suppor-r.
The FPTP
system. Some prominent laws enactedfor this electoralsystemresultedin the ruling Congress
purposeincludethe Representation
of PeopleAct, party securingstable majorities in Lok Sabha.
1950 and 1951, the Presidential and Vice- usuallyagainsta fragmentedopposition.However.
Presidential
ElectionAct. 1952.and the Delimita- since 1977,when the oppositionpartiescombined
tion Act, 1952, etc.
to form coalitionsand startedputting up common
(aswas
candidates
againsttheCongresscandidates
Indiaremainsby far thelargestdemocracyin lhe thecaseinrhe 1911and 1989generalelections).
the
world. with almost 600 million voters. India's Congressmajoritieshavevanished.Moleover,the
parliamentarygovernmentand the First-Past-the-natureof the systemis that small changesin vote
Post (FPTP) electoralsystemare a legacyof the shareoften have a dramaticimpact upon the shape
British colonialism,which ended in 1947. The of the resultingParliament.
IndianConstitutionprovidesthat all adultcitizens,
who are l8 years or more of age,and who are not
The overall results of electionsto Lok Sabha
otherwisedebarredfrom voting, can exercisetheir have neverbeenproportional.Becausethe candiright to the franchise.Voters elect a 543-member date, who obtainsthe most votes polled but not
Lok Sabha(the lower house),from single-member necessarily
a majorityof votersis declaredelected,
constituencies.and each of India's states has supportcan oftenbe dividedby settingcandidates
adopteda similarsystem.By contrast,membersof of the samecaste,religion,or legion againsteach
the lower house of Parliamentand of the state other. However, despite the divided nature o1
legislativeassemblies
membersof indirectlyelect India's multi-ethnicdemocracy,the electoralsysRajyaSabhaor theCouncilof States,
aswell asthose tem hasretaineda considerable
degreeof support,

voL.l5 No.3

PARLIAMENTARYELECT IONS AND POLIT]CAL PARTIES

6-17

due in part to the practice of reserving seatsfor constituency.


It is very difficult to take into consociallyunderprivilegedgroups.The Indian Con- siderationall the factors (especiallythe sociostitution,reserves22 per cent of all seatsfor hrs- culturalfactorsthat onemight think of to be related
torically disadvantaged groups, known as to a party'swinning),dueto lackof dataavailability.
Scheduled
Castes(79 reserved
seats)andScheduled We havetakenonly thosevariables,which seemto
Tribes(41 reservedseats).In theseconstituencies,be the most importantin termsof their effect on the
only a memberof the ScheduledCastesor Tribes log-oddsof an incumbent'swinning,andfor which
may contcstthe polls, althoughall electorshave reliabledatais available.Thereliesthe opportunity
votingrights.It hascnsuredthattheirparliamentary to increase
the richnessofthe datasetandgetmore
representation
is in line with theirproportionof the preciseestimates,
but consideringthe lengthof the
population.A constitutionalamendment,which paper,we restrictourselvesonly to thesevariables.
seeksto reserve33 per cent of seatsfor women For, we are more interestedto enrich our underrepresentatives,
is currentlybeingconsidered.
standingthanpredictingthe probability.
In the first Lok Sabhageneralelection(1951)
Anti-incumbencyeffects may be sensitiveto
therewere singleseated,double seatedand triple voter-turnout. There are competing hypotheses
seatedconstituencies.
There were 401 constituen- aboutthe directionof the turn-outvariablefbr vote
c i e s ,o u t o f w h i c h 3 1 4 w e r es i n g l es e a t e dc o n s t i t u - sharesof a party. A popular incumbentmight
encies( I 0 fbr STs),86 double-seated
constituenciesincreaseloyal voters'desireto go to the polls.Yet,
resultingin 172 seats(71 fbr SCs and l5 for STs), some voters may stay home, if the 'saf-eseat'
and one was triple seated,resulting in three seats incumbentseemssure to win. The potentialfor
(one fbr SCs, one for STs). Thus, there were total turnout to damage candidates' fortunes is most
489 seatsin 195I parliamentaryelections.North likely,whenvotersareunhappywith theincumbent,
Bengal (Wcst Bengal)constituencywas the only and go to the polls to registertheir dissatisfaction
constituency,
which had threeseats.
, o n l e y ,2 0 0 2 ] .
[ H i n i c h ,e t a L ,1 9 9 9 C
In thesecondLok Sabhageneralelection( I 957),
therewere only singleand doubleseatedconstituencies.Therewere403 constituencies,
out of which
312 (16 for STs) singleseatedand 91 were double
scatedconstituencies,
resultingin a totalof 182("16
I'orSCsand l5 for STs).Thus,therewere494 seats
in Parliament.Since the third Lok Sabha(1962)
onwards.therehave beenonly singleseatconstituencies,andtheseseatshavebeenvaryingbetween
5l8 to 543seats.
Model Specffication

Electoralsizeis includedin themodelindirectly


becauseelectoralconstituencies
in India. unlike
thosein the United States,are not of equal size.
Electoral size may be important to candidatesin
several ways: First, the potential obstaclesto
reachingvotersduringthe campaignmay increase
in proportionto total population.An incumbent's
nameandrecognitionmay not be as grearin bigger
constituencies
as in small.Second.as an electoral
sizeincreases,
the MP's ability to provideconstltuencyservicesto votersmay diminish to reducethe
valueof incumbencystatus.

T h e p r o b a b i l i t yo i a p a r t yw i n n i n gc o n s e c u t i v e Dummy variablesfor the statesare brou_9ht


to
electionsin a constituencydependson the party's bearin themodelto controlfor regionaleffectsnot
ruling in that constituency.
We have tried to take capturedotherwise.A detailedtreatmentof differ'intoaccounta setof factorsthatmightbe potentially encesin state/regional political culture is well
responsiblefbr a party's wining in a particular beyond the scope of this work. Yet, historiczrl,

638

JOURNAL OI: IND]AN SCHOOL OF POLITICAI- ECONOM'I

J U L Y . S E P 7 2. .O O . ]

economic,andculturalfbundations,
thatdistinguish
4. Dummy variablefor thepartybeinga national
the Indian statesand regionsfiom one another,may party or not (nationalparty).
exerta significantinfluenceon electoraloutcomes.
This is to controlthe nationalparty'sability to
get more resources
The modelproposedhereis:
and its accessto power,which
caninfluenceon theresult.One may arguethat if a
(wins),,= cx.0
+ Br(nationalparty),,+ Br(reserved), candidateis from National party, his/herpromiscs
+ p.,(sex.y,,
+ B.(electors),,
+ B.(validvotes)', aremore likely to be fulfilled,comparedto others.
or candidates may get the thvour due to the
pr(candidates),,
+ p(,(vorers),,
+
+ Br(duration),,
+ Br(state),
+ Br0(time),
+ u,l favourablescenariofor the oartv.
5. Durationis the gap betweentwo consecutive
We have taken the dependentvariable as a
elections
measured
in years,to controlfor theeffect
dichotomousvariable(wins),which takestwo valgap
of
time
to
fulfill
one's promisesbeforegoing
ues,i.e., I if the incumbentpartywins the election,
for
the
next
election.
and0 otherwise.The explanatoryvariablesusedrn
the modelare:
6. Valid votescomprisethe ratio of total valid
votes
polledto thesizeofelectorate,
i.e..thevariable
L Voters' turnoutin the constituency(poll per(knowledge,)
to
control
fbr
thc
education
levelin the
centage)is the percentage
turnoutof votersin that
constituency.We assumehere that the knowlelection,showingthe effectof thepublicdesirefor
edgeableperson would caste his/her vote with
a changeor the extentof anti-incumbent
voting.
propercaution,reducingthe ratio of invalid votes.
It is a proxy usedto control for the knowledgeof
2. Categoricalvariable for the category of the
votersin the constituencv.
(reserved).
constituency
7.Candidates
comprisethenumbero1'candidates
Thereis generalperceptionthatreservation
of a
contesting.
This variablecontrolsfbr the eff'ectof
particularseatcan give undesiredfavour to some
competitionin theconstituency.
One may treatthls
political partiescomparedto others,for example,
variableas a controlfbr the dummv candidates.
regional parties may have better penetrationrn
reservedseatsthan national parties.This variable
8. Time dummiesto capturethe effect of current
takesvalueof 1, if the seatis reserved,
0 otherwise. socio-economic-political
scenarioat the time of
e l e c t i o no n t h eo u t c o m e .
3. Categoricalvariablefor the sexof the candid a t e s( s c x ) .

This is to capture the eflect of the current


scenario,ratherthanthe effectof the ruling period
Indiancase,it is assumedthat maleshavemclre (whichis capturedthroughduration).For example,
accessto power than women, which gives them in year 1980 we find that the year dummy is
undue favour. This variable controls the eff'ectof significantand negativethis could be explainedas
sexdiscrimination.
It takesvalueof l, if the winner theaftereffectofthe emergencyperiodthatresulted
and loserareof the samesex,0 otherwise.
in the resultsagainstthe ruling partyCongress.

voL. t5 No 3

PARLIAM ENTARYELECTIONS AND POLIT]CAL PARTIES

6-19

9. Statedummiesto controltheunobserved
state requiredvariableyulns,as the prior electiondatais
characteristics.
not readiiyavailablefor eachconstituency,
due to
reconstruction
of the constituencies.
It leavesus
As the dependentvariableis dichotomous,we with the electiondatafrom 1971-1999(i.e.. 1771,
are using the Binary Choice (Logit) model. We 1 9 1 1 , 1 9 8 0 , 1 9 8 4( 1 9 8 5 * ) , 1 9 8 9 , 1 9 9 1 ( 1 9 9 2 * ) ,
panel-datamodels 1996,1998and 1999).This, accountsfor l0 elecestimatepopulation-averaged
usingGEE.This approachestimates
cross-sectionaltions out of total l3 held so far. We haven'ttaken
time-seriesmodels.In particular,it estimatesgen- databefbre1967elections,as therehasbeenmajol
eral linear models,and allows specrfyingwithin restructuringof the boundariesof constituencies
group corelation structure fbr panels. This beforel97l election.
producesvalid standarderrors,evenif the correlaThus, we end up with 4,143 constituencytions within group are not as hypothesised
by the
elections,which consist of minimum five and
specifiedcorrelationstructure.
maximum nine electionsper constituency,as ln
someconstituencies,
electionhad not beenheld ar
For a thoroughdiscussionof GEE in the estitimes.In somestateshadnot beenheldelectionsat
mationof GLM (Generalised
LinearModels),Isee, the time of generalelections
but later on we have
Zeger and Liang, 1986(a)(b),Pp. 13-22, Pp. treated;thoseelectionsonly havetakenplace
at the
121-1301.Furtherinformationon linearmodelscan sametime.For example,in Punjabthatthe election
be found in Nelder and Wedderburn[972, Pp. for 1984took placein 1985.In suchcases.we have
370-3841.Finally, there have been a number of treatedthem as ifthey had beenhcld in 1984only.
illuminatingarticleson variousapplications
of GEE
in Zeger,Liang, and Albert [988, Pp. 1049-60], Entpirical Results
Zeger and Liang [1986(aXb), Pp. 13-22, Pp.
121-1301,
and Liang [987, Pp. 695-702].PenderTable I represents
the basiccharacteristics
of the
gastet al., I I 996,Pp. 89-I I 8] providea nicesurvey Indian Lok Sabhaelectionsin terms of number of
ofthe currentmethodsfor analyzingclustereddata candidates,contestantsforfeiting deposits,number
candidates
belongingto difin regardto binary responsedata.Our implemen- of womencandidates,
tationfbllows thatof ZegerandLiang [ 986(a)(b), f'erentpolitical parties (in terms of national or
contestingin
P p .l 3 - 2 2 , P p .I 2 l - I 3 0 1w h i c h h a s b e esnu m m a r i s e dregional).The numberof candidates
in Appendix.fbr theconvenience
of thereader.We electionshas increasedtill 1996. after which it
generalized
linearmodelsof Y,, with covariates
X1,, starteddeclining.In 1957,therewere I ,5l9 candidatescontestingfor theelectionof 494 seats,while
in 1996 for 543 seatstotal of 13.952candidates
Empirical Analysis
contested.
Theseincreasing
numbersof contestants
were
mainly
from
independent
category,and most
This sectionis divided into two Sub-sections:
of
them
could
not
save
their
deposits.
Depositsare
sub-section
I focuseson thedatawhereasemoirical
fbrleitedif the candidateis not able to bag even3
r e s u l t sa r ep r c s c n t e icni S u b - s e c t i o2n.
percentof votespolled.One may algue that these
candidates
wereusedasdummycandidate
to cut on
Data
rival candidates
vote shareon the baseof religion,
The data usedfor the analysisis obtainedfrom sex, caste and other socio-economicfactors.In
lhe Reports of Election Commission of India. The 1951,360 candidatesfrom independentcategory
dataset consists o[ general election data from lost theirdepositsfrom the total of 533 candidates;
1961-1999.The first election-year
for our studyis while in 1996, out of 13,952candidates.12,688
1967. The data from 1967 is used to folm rhe candidates'
depositswereforfeited.

640

JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY

JULY-SEPT.2AO3

Table l

Candidates
0)

Election Year
(2)

NP
(3)

SP
(4)

Contested
Won
ForfeitedDeposit

1 9 5I

| 1t'7

124
34

533
31
360

Contested
Won
Fort'eitedDeposit

1957

9t9
421
130

ll9
3l
40

481

Contested
Won
ForfeitedDeposit

t962

1.269
440
362

zt7
28
108

Contested
Won
ForfertedDeposit

t96'7

t,342
440
390

148

Contested
Won
Forf-eitedDeposit

197|

t.223
451
359

)aA

Contested
Won
ForfeitedDeposit

t97'7

Contested
Won
ForfeitedDeposit

t980

Contested
Won
ForfeitedDeposit

I 984*

Contested
Won
Forfeitcd Deposit
Contested
Won
ForfeitedDeposit

418
344

RUP
(5)

IND
(6)

r'1

WC
(7)

Total
(8)

I,874
,+89
145
45
28

1,5
l9
191
494

324

70
6
8

479
20
378

66
3l
t9

1.985
494
856

l3
2

o/

) 16q

866
35
747

29
I4

520
1.20_1

203
l3
156

I 134
t4
1,066

80

40
t26

43

2..784
518
t.707

1.060
481
100

85
49
6

70
3
60

t.224
9
I,190

70
l9
3t

2.439
542
1.356

1,541
485
444

106
28

156
I
l5t

2,826
9
2,794

I '13
28
79

1.629
529
3.417

l,307
462
387

165
66
45

126
0
120

3,894
l3
3.830

t7l
+-1
109

5 4ql

5.rI
4.382

I9 8 9

1,378
11|
421

143
21
12

926
l9
868

3 , 7t 3
l2
3.672

198
29
ll5

6,160
529
5.003

l 9 9 l*

I,892
490
852

521

5.606
I
5.590

318

356

842
1
820

l8l

8.861
547
7.68
1

58

t+

)l

4)

Contested
Won
ForfeitedDeposit

1996

l , 8 l8
403
898

451
127
167

1.048
1
I,020

10,635
9
10.603

599
40
518

13.952
543
12.688

Contested
Won
Forl'eitedDeposit

l 99u

t.493
381
631

171
t0l
207

871
19
744

1,9t5
6
l,898

214

;+3
lu4

4.750
543
3.,+86

Contested
Won
ForfeitedDeposit

t999

| )ao
369
431

750
158
4lt

b)4

I,945
6
l,928

284
49
183

4,648
543
3,400

l0
o/+

* Includesthe data for the electionshappenedin later yearsin somestates.


NP: National Parties.SP: StatePanies,RUP: Registered(Unrecognized)Parties,IND: Independents.
WC: Women Candidates

vqL. 15NO.3

PARLI A M ENTARY EI,ECTI ON S AN D P OI.I7'ICAL P ARTI ES

Figure l
o
o
o,
o

Winningsby NationalParties

100
90
80
-7rl

o
o

OU

50
40
30
1970

tv/5

1980

1985

1990

1995

2000

ElectionYear

Figure 2
o
G

Incumbent'sWinning

OU

o
@

cl

40
zv

o
bS 0
1970

1975

1980

1985

1990

ElectionYear

Figure 3

Sex-wisedistribution

o 105
G
o 100
o
(!

o
o

;s

95
90
B5
BO
1970

1980

1985

ElectionYear

1990

1995

2000

JOURNAL OF INDIAN SCHOOL OF POLITlCAL ECONOMY

JULY-SEPT.
2OO.]

We also observe increasingparticipationof 1980 and 1984 is negativeand significant.This


femalecontestants,
althoughstill the participation implies there is an anti-incumbency
bias in these
is limited in nature.In 1951election.45 women electionyears,which resultedin the cyclicalbehacontested
for the election,out of which 28 won and viour of the seatswon by the nationalparties.In
only 8 losr rheir deposits.In 1999 election284 1980electiondue to the emergencyin the earlier
womencontested,49
won and 183losttheirdeposits period led to negativeimpact on the incumbent
i.e. out of 4,648 candidates
only 284 were women parties' performance.One other explanationcan
and only 49 got electedin 543 seatsparliament.
alsobe providedto explainthis cyclicalbehaviour
of the incumbent'swinning.Onemay arguethatin
Figure I showsthatthe nationalpartiesys.others the initial electionsthe winning partiesenjoyedthe
haveperformedwell initially winningabout80 per powerbut didn't performwell which resultedin the
centof the total seats.which droppeddramatically negativeimpact on their performanceand resulted
in 1998election,rhcyrecoveredin 1999elecrionto in the downward trend in the winning of the
70 per centoftotal seats.In Figure2, we show the incumbentparties.Whereasin thelaterperiodafterpattcrnof incumbentsgettingre-eleoted,
the cycli- 1980electionthe incumbentsgainedtheir voters,
cal behaviourof the percentageof seatswon by the trust and worked for the well being of the voters.
incumbentis apparenthere,this alsoreflectsin the This resultedin an increasedtrendtowardswinning
regressionresults.we flnd that the year dummies of theincumbent.This trendcontinuedtill the 199I
for the third and folth period is negativeand srg- election.In 1996 election,due ro high volatile
nificant while for all other years is positiveand political and economicscenarioresultsare again
significant. Which
shows, there were againstthe incumbents.One may arguethat once
anti-incumbencybias in those periods.In 1984 theincumbentpartiesarein powerfbr longerper-iod
election,Congresswon almostall the seatsin the their performancegoesdown and then in the nexr
parliamentary
electronwhich may be influencedby electiontheyloose.The votersgive thepowerto the
the sympathetic behaviour of the voters due to others for providing better governance.Figure 3
assassination
of Indira Gandhi.This effectmay not shows,how the sex of opposingcandidates
differ.
be captured through the duration variable. As ln 1911election,most of the winning and losing
durationwould besamefor 1989elecrion(1984-89) (herewe definelosingcandidateas runnerup cana s w e l l a s f o r 1 9 9 6 e l e c t i o n( 1 9 9 1 - 9 6 )b u t t h e didate) candidatesare of the same sex, which has
socio-economic-politicalscenario would be reducedto 87 per cent from 100 per cent in 1999
entirelydiffercnt.In this casealso the coefficient election.
for the year dummy is negativeand significant,
implyingananti-incumbency
votingbehaviorof the
We now shiftour attentionto theoutcomeof our
votersby andlarge.For restofthe electionyearsthe regression
analysis.
Theresultsofthe regression
are
coefTicient
for the year dummy is significantand reportcdin the Table2. We find tharthe signof the
positive, which implies that there were positive coefl'icients
of sex.durationand poll percentage
is
incumbencybias for the ruling parties.This is as negative,
andthesignofnationalpartyandreserved
expectedsincethe partiesin the power have better candidatesare positive.All of the statedummies
control over the f'actorswhich may influencethe have negative coefficientsexcept Jammu and
outcomeof the elections.This trendin the election Kashmir,Mizoram and West Bengal.rThe coeffialsohelpsin explainingthe cyclicalbehaviorof the cientof statedummyfor JammuandKashmiris not
percentage
of seatswon by theincumbentparty.One significant,whereasit is signifioanrfor Mizoram
can clearly see that the coefficients ol' the time and West Bengal, this significant and posttive
dummy with referenceto the year 1977for the years coefficientcouldreflectthepersistence
ofthe sarne

Y O L .t 5 N O . 3

PARLIA M ENTARY EI,ECI' I ON S AN D P OL]T I CAL P ARTIES

613

Table 2. Generalized Estimating Equations - Population Average Model (GEE-PA Model)


Vanables
(l)

Odds Ratio
(2)

p>lzl
(3)

Coefficients
(4)

p>lzl
(5)

National Pany
Reservcd
Sex
Duration
Poll Percentage

2.121444*
1.28389*
0 . 9 5 7 8 3I 2
0.1033292*
0 . 8 2 9 6 3x6 1

0.000
0 004
07l9
0.000
0.001

0.7520968
0.2498942
-0 0430828
-0.35I 9302
-0.I 86768
I

0 000
0 004
0.719
0.000
0.001

Valid Votes
Candidates
Constant_
Andhra Pradesh
ArunachalPradesh

1.36E+08*
1.000599

0.002
0.847

0.6805947*
x
0.3884401

0 004
0.000

l 8.7268
I
0.0005984
0.'7265'703
-0.3847882
-0.9456164

0.002
0.847
0 041
0 004
0.000

Assam
Bihar
CaprtalTerntory
Delhr
Goa

0 90857.1
03I65001*
0 6 l 64599
0.33764t7*
0 l 5 3 ll 4 +

0.179
0.000
0.389
0.000
0.000

-0.0958789
,l I50432
-0..18376
r9
- 1 08577
-t.876572

0.179
0.000
0.389
0 000
0 000

Gujrat
Haryana
Himachal Pradesh
Jammuand Kashmir
Karnataka

0.4793286*
0.2I 59504*
0.2825454*
1 460552
0.7541362

0.001
0.000
0.000
0.199
0lt3

-0 73-53689
-t 532106
-t.2639t6
0.3788
I 47
-0 2821823

0 001
0.000
0.000
0.199
0ltl

Kerala
Madhya Pradesh
Maharashtra
Manipur
Meghalaya

0.982025
I
0.49I 0497*
0.4564442*
0.569052-5*
0.9128536

0 936
0 001
0.000
0.000
0.904

-0018I384
-0.7il2099
-0 7842888
-0.5637825
-0.02752t7

0.936
0 001
0.000
0 000
0.904

Mizoram
Nagaland
Orissa
Punjab
Rajasthan

I .96t 654+
x
0.5746601
0.35074*
0.261497'7*
0.23t741t*

0.000
0.000
0.000
0.000
0.000

0.6737882
-0.-5-s39766
-t.041tI
-1.34133
-t.462t34

0.000
0.000
0.000
0.000
0.000

Tamil Nadu
Tripura
Uttar Pradesh
W e s tB e n g a l
Year I 97 I

0.9285822
0.10157
64*
0.2483291*
2.323487*
| .7469424

0.572
0.007
0.000
0.000
0.000

-0.0140964
-0.3544255
- 1 392998
0.843069
I
0.557867

o.572
0.007
0.000
0.000
0.000

Year I980
Year 1984
Year I 989
Year I 99 I
Year I 996
Year I 998

0.093048
I*
0.18691t?-4
2.079232*
1.720078x
3.6I 6483*
l.724lt]E

0.000
0 000
0 000
0.000
0.000
0.000

-2.374639
- I 6768
0 . 7 31 9 9 8 4
0.5423696
1 285502
0.54483
II

0.000
0 000
0 000
0.000
0 000
0 000

The exchangeablecorrelation(the observationswithin a subjectare assumedto be equally correlated)doesnot in generar


provide random effects but this gives the same kind of structurethat ansesin a normal linear model with random intercepts.It
actually lits an equal-correlatronpopulationaveragedmodel that is equal to the random effects model for linear equation.The
distributionofdependentvariableis assumedto be ofbinomial family with logit link. It acknowledgesthat correlationsexrstwlthln
a subject,but it assumesthat the correlationsare constantover tlme.
* indicatessignificanceat 5% level of sisnificance.

VOL.15NO.3

PARL|AM ENTARYELECTIONS AND POLITICAL PARTIES

P ( X , , 1 = P ( Y , , =I 1 X , ,=) P ( Y r , = I I X r , ,X . , ,X . , , . . . .X, k , )
The underlyrnglatentvariablemodelis:
Y " = B ' X + c , w h e r eY = I i f Y - > 0
HereY is the dependentvariable(winning),ande is a continuously
distributecl
variableindependent
ofx ancl
the distributionof e is symmetricaboutzero.
In the GLM approachIseeMcCullaghand Nelder,19891,we assumethar

= xiF
h1p,,)
= C(prj)O
Var(y,.,)

' (.* : , , 8 ] '


[ , = E ( y ,=) h - ' r x ] , F rh. n
A, = dia-u{g(p,
,),A, C(p,,,,)}
Cov(y,)= 0A,

lirr

independent observarion

U s i n g t h e n o t a t i o n t i o m Z e g e r a n d L i a n g l l g 8 6 ( a x b ) , p p . 1 3 - l222l -, 1
p3
p0
. 1l ,e r y ,= ( y i . r , K y , . , , ) ' r b e r h e n , x
I vector of outcome values,and let x, = (x,.,,Kx,.",)t
be the n, x p matrix of covariatevaluesfbr the irhsuhiecti 1,. ..,m.We assumethat the marginaldensityfbr yi.;may be writtenin exponentialfamily notationas
| ( y ,i ) = c x p [ { y rr 0 , , - a ( 0 ,, ) = b ( y r . ,}) 0 l
where0,.,= h(L.,),lr; = x,.,B.Underthis fbrmulation.the first two momentsaregiven by
E ( y ' J= a ' ( 0 , ) ,
v a r ( y , . '=) a " ( 0 , ) i 0
We define the quantities(assumingthat we havean n x n working correlationmatrix R(a),
A , = d i a g { d 0, ,,d 4 , , t
n , - d i r g { a t 0 ,, t }
5 , = y ,- a ' ( 0 , )
D=A,A,X,
v, = A'tR{cr)A"

nrn matrix
n.rn marrix
nx I marrix
n.pmarrix
nxnmatrrx

suchthat the GEE becomes


-v r uI
'S,
) D,',v, = (r
We then have that

644

JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY

partiesin thesestates,while the negativecoetficients for some other state dummies reflect antiincumbencybiasat work in (all or someof) theother
states.2
The estimatedcoefficientsof nationalparty,
poll percentage
reserved
candidates,
andvalidvotes
areall statisticallysignrficantat 5 per centlevel of

JULY-SEPT,2OO.]

that the sex and the numberof candidates


do not
have significanteffect on the probabilityof the
i n c u m b e nwt i n n i n gt h ee l e c t i o n . '
ConcLusion

Using the datafrom Indianparliamentary


elecsignificance.
We find that the log oddsof winning
tions,we showthatcandidates
ofthe nationalparties
arehigherif the incumbentbelongsto the national havea betterchanceof winningagainfrom thesamc
party.One interestingfindingsof our analysisis in constituency.
An increasein the poll percentage
in
termsof the coefficientof polling percentage:votes a constituency exerts an adverse etfect on the
polledenterswith a negativesignificantcoefficient, probabilityof the incumbentwinning the election.
which suggeststhat as the votersturnoutincreases Voters are willing to cast their votes of others,
in a constituency,
the probabilityof winningof the reflectingthe poor performanceof the ruling party
when they want the
incumbentdecreascs.
i.e., more voter turnoutsig- in the particularconstituency,
requisitechangein thegovernment.
Oneinteresting
n i f i e sa n a n t i - i n c u m b e n cpyo l l i n g .T h i s m i g h t b e
way to expandour work is to relatethis analysisin
due to the fact that thosevoters.who would have
termsof variousdemographic
characteristics,
using
been otherwiseindiffbrent(by not going to caste the
Censusdata.Onemay alsotry to explainwhether
their vote), have disliked the incumbent'sper- the public goodsprovisioncan explainthe probafbrmanceand get the motivation to vote againstthe bility of winning election.However,theseare left
ruling party fbr effectingthe change.We alsofind to future research.

APPENDIX

g{ E(Yt,)}= X,, F, Y - F with parameters


0,,.
f o r i = l , 2 , . . . na n dt = l, 2, ...n,,wheretherearen, observationstbr eachgroup identilier i. In the above,g{ } is called
the link function and F the distributionlamily. If g{ } is the logit functionand Y,, is distriburedbinomial,we have
l o g i t { E ( Y , ) } = X,, 0, Y - binomial.
The link functionlogitis definedas ln{y/(l-y)}, the naturallog of odds.We assumethe workingcorrelarion
rnatrixfor modelingthe within-groupcorrelation,a squaremax{n,} x max{n,} matrix,say R. Then the correlation
structure(correspondrng
to equal-correlation
models)is definedas:

if
R,..=
otherwise
In the GEE approachthe unobservedeft-ectsbinary responsemodel, the responseprobabilitiesare specified
conditionalonly on X,, with the resultthat we havethe fbllowine:
The responseprobabilityis,

6.t6

JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY

JULY.SEPT200.]

=F,- i oto,,o,'
j
B,-,
r0:
rn,t0:i'
r0r
I
l,iol,B,,o'',p.rrs
wherethe term

'l F:lv;'to,)D,(
Bi)i
iolt
is what we call the IRLS varianceestimate(iterativelyreweighedleastsquares).It is usedto calculatethe standard
errors.

NOTES

K i n g ,G . ; J . H o n a k e rA
, . J o s e p ha n dK . S c h e v e2. 0 0I I ' A n a l y z i n g
IncompletePoliticalScienceData:An AlternativeAlgorithm

for Multiple lmputation'. Americut Politicul SL:ient:e


l. We have also performed the regressionwith regional
R e v l e u 'V. o l . 9 5 . N o . L
dummies instead of state dummies, but qualitatively results
'Estimating
Lrang, K.Y., 19871
Functions and Approxtrnate
reportedin Table 2 remain unaltered.
ConditionalLikelihood'. Biometriku,Vol. 1.
2. Thanks to the anonymous referee for providing this
McCullagh, P. and J.A. Nelder, 19891 'GeneralisedLinear
explanation.
3. Introductionof the lagged valuesof the dependentvarrM o d e l s ' , 2 n de d n . ,C h a p m a n& H a l l , L o n d o n .
ablesdo not changethe resultssignificantly.The only noticeable Merrill, S. IIl., 1994; 'A ProbabilisticModel for the Sparial
change is the significanceof duration, which turns out to be
Distribution of Party Support in Multiparty Electorates'.
insigniticantafier introductionof laggedvaluesin the rnodel.
J ountu I ol A nteri ctrn Stut i st it'ul A s,ttt citLtion.
N e l d e r , J . A . a n d R . W . M . W e d d e r b u r n .1 9 7 2 ; ' G e n e r a l i s c u
RI,FI'RENCES
L i n e a r M o d e l s ' , J o u r r u t lo f t h e R r l l i l S t u t i s t i c u S
l oden.
S e r i e .Ar . V o 1 . 1 3 5 .
'Constituency
C o n l e y ,S . R . .2 0 0 2 ;
C o n t e x t s I, n c u r n b e n c ya,n d
N o r m a l .S . ; M . D . A n d r e w ,Q . M . K e v i n a n d W . A n d r e u , .1 9 9 8 ;
PartyCompetitionin CanadianRidingsin 1990s',Univerrr4,
'Multiparty
Electoral Competition in the Netherlandsand
ol Murthnd Working Puper.
'Estimating
Germany:
A
Model Based on Multinomial Prcbit', Pubtk
Gelman. A. and G. King, 1990;
the Electoral
Choice.Yol.9'7.
Consequencesof Legislative Redistricting', Jountul rl
Pappi,F.U. and E. Gabriele,19981'Voter's ParryPreferences
in
A mericun Stu istit ul Associutkttt.
'
A
Multiparty Systems and Their Coalirional and Spatial
C e l m a n .A . a n dG . K i n g . I 9 9 4 1 U n i f i e dM e t h o do f E v a l u a t i n g
Irnplications:Cermany after Unification'. Public Choite,
Elcctoral Systems and Redistricting Plans', Atneriton
Vol.97.
J o t r r n u lo . fP o l i t t t u l S t i e n c e .V o l . 3 8 . N o . 2 .
G c l m a n .A . a n d J . N . K a t z . 2 0 0 l : ' H o w M u c h D o e s A V o t e P e n d e r g a sJt. F . l S . JG
. a n g e ,M . A . N e w t o n ,M . J . L i n d s t o r n i M
. .
Count?Voting Power,CoalitionsandtheElectoralCollege',
P a t l aa n d M . R . F i s h e r , 1 9 9 6 : ' A S u r v e y o f M e t h o d sf b r
Sot:ittlScienceWorking Puper I ,l2l , California Instituteof
Analyzing ClusteredBinary ResponseData', Internutionul
Technology.
S n u i . t t i c dR e v r e r rV
, ol. 64.
H i n i c h ,M . ; J . V a l e r i , P . C .O r d e r s h o o k1, 9 9 9 ;' U k r a i n e ' s1 9 9 8 Tomz, M.; T.A.
Joshuaand J. Wittenberg,2002:'An Easy and
ParliamentaryElections: A Spatial Analysis', Post-Soyiet
AccurateRegressionModel for Multiparty ElectoralData'.
A l J h i r s , Y o l 1. 5 ,N o . 2 .
Politiul Anulysis,Vol. 10, No. l.
'A
J a c k s o nJ. . 8 . .2 0 0 1 I S e e m i n g l yU n r e l a t e dR e g r e s s i oM
n odel
WooldridgeJ.M., 20021Econonetri( Anah'slstl Cxtss Setrrort
fbr Analyzing Multiparty Elections', Annual Sumnter
und Panel Dutu, The MIT Press.
PoI i t Lt:aLM et hotloI o gt' Conlerence, Emory University.
'Models
for
K a t z . J . N . l J . H o n a k e r a n d G . K i n g , 2 0 0 1 1 ' A n l m p r o v e d Z e g e r ,S . L . ; K . Y . L i a n g a n d P . S . A l b e n . 1 9 8 8 :
L
o
n
g
i
t
u
d
i
n
a
l
D
a
t
a
:
A
G
e
n
e
r
a
l
i
s
e
d
E
s
t
i
m
a
t
i
n gE q u a t r o n
Statistrcal Model for Multiparty Electoral Dara', .locial
Approach',Biometric.t,Y ol. 44.
St'renceWorking Puper I I I | , Califbrnia InstituteOf Techl
Z e g e r ,S . L .a n dK . Y . L i a n g ,1 9 8 6 a ';L o n g i t u d i n aD
a t aA n a l y s r s
nology.
'A
Using GeneralisedLinear Models'. Biometriku.Y ol. 73.
K a t z ,J . N .a n dG . K i n g , 1 9 9 9 : S t a t i s t i c aMl o d e lf o r M u l t i p a r t y
l
ElectoralData', Anrericur Journul of Political Science,Y ol. Z e g e r ,S . L .a n dK . Y . L i a n g ,1 9 8 6 b :' L o n g i t u d i n aD
a r aA n a l y s r s
9 3 .N o l .
for DiscreteandContinuousOutcomes'. Biometrics.Vol. :12.

Vous aimerez peut-être aussi