parliamentary elections and voting behaviour

Upload: rumi-roy

Post on 05-Jul-2018

220 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    1/14

    PARLIAMENTARY ELECTIONS AND POLITICAL PARTIES

    Jayesh umar

    We discuss here the probability of incumbent art.\"s winning he election rom the same constit-uency..We st-imate he probabilitl, of winning by the uling parl,, irrespective f whether hat parb,formsa coalition during election, has been ct member of the last government n the centre. We stimate theaverage robabilin' of winning n India's general elections rom 1971 to 1999, under he given electoralsystem, sing a panel data Binary Choice Model.

    India having he largest lectoral mandate ndmultiparty le ctoral ollege n the world s the mostsuitable andidate or analysing lections. ndia ha sthe arge base f politicalparties over 670) and, aspolitical arties re ndispensable br the successfulworking of any democracy, hey offer a ot of choiceto voters. This article ries o explain he behaviourof electorates. We try to model the chances of anincumbent getting re-elected n the election. nparticular, we attempt o answer whether he sex ofcandidates ffects he outcome of any election?Does he dent ity of a party n terms of national s .regional matter'l

    At the heart of any electoral democracy s theidea of 'one person, ne vote , and one value ' , . e . ,anyone's ote should not have higher weight hanthat of another. n India, he discrepancy etweenelectoral onstituencies s enormous. he largestconstituency n 1999, orexample, Outer-Delhi, adnearly 86 imes more voters han Lakshadweep, hesmallest constituency. t shows hat the vote ofLakshadweep's oters as more weight han hat ofthe Outer-Delhi's n choosing heir epresentativesin Parliament. Multiparty elections or Parliament,state egislature, nd local legislature re a key

    vehicle br th e expression f ideological iversity.Elections also provide a forum for advances nwomen's political participation. Women have ha dthe right fbrever, but the recent rends oward morepluralist elections have made their participa tionsomewhat more mcanineful. ndian women ar c

    woefully underrepresented n elected office, andfemale candidates re few. However. women arenow competing iom almost every constituency.and n some constituencies he number of womencontestants ivals men's. Ruling parties nd theirsupporters swept almost every electoral contestduring 1950-90, only a handful of times haveopposition candidates won a plurality in nationalelections. Several factors explain the poor per-formance. he non-Congress pposition acked hegrassrootsupportuntill he 1980s. he ulingpartiesenjoy extensive patronage networks and havesuperior ccess o the national/regional edia. Th e'winner-takes-all' system, sed n most constilu-

    encies . enef , i tncumbent ar t ies .

    In parliamentary systems with proportionalrepresentation nd multiple parties, he primeminister and hisftrer abinet re dependent n thesupport and confidence f more than one parlia-mentary party. Such coalition governments reformed anew after each election, n which voterschoose arties nd not coalitions r governments.Coalitional ptions re iscussed uring he electioncampaigns ut not all options urn out to be easibleonce he elections-returns ecome nown. Contrary

    to wo-party parliamentary ystems, he electorate'ssignals or a new government reambiguous, ivingthe party leaders ome eeway n coalition bar-gaining. n order o fulfil most of there promises.parties ave not only to enter a winning coalition,but they have o reach a coalition agreement hat

    Jay esh u mar s Resea rch ch o la r, n d i raGan d h i n s t i tu te f Dev e lo p men t e sea rch u mb a i . eMa i l : j ay e sh @ig id r. a c . inJEL Classification: C25. N45. Keywords: coalition, electoral ystem, nd ruling party.

    This paper s a modified version of the paper presented t the 9th Intemational Conference f Forum for Interdisciplinar.yMathematics n Statistics. ombinatorics nd Related Areas SCRA-FIM-lX) at Allahabad, December2l 23.2002. Author'thank.sall panicipants or their comments. n addition many hanks re extended o Kausik Chaudhuri orhelpful commenrs nd suggesrrons.Special hanks are due to the Journal's Editor and an anonymous eferee or detailed comments and suggestions. hiih helpedsu b s tan t ia l lyn imp ro v in g h e p re sen ta t io n . h e u su a l i s c la i rn e r p p l ie s .

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    2/14

    puts the promised policies on the governmentagenda. n this context, parliamentary lectionsprovide an mportant xperimental etting o studythe voter 's choice behaviour n choosing heir ep-resenlatives n Parliament or the functioning ofdemocratic nstitutions. Moreover, n segmentedsocieties ike India developments ertain n the1990s have changed he context n which

    theyfunction. Performance f democratic nstitutions ssubstantially function of networks of trust andcivic engagements mong itizens y participatingin coordinated ollcctive ction. or any democracywith such a large electoral mandate, lections anhardly be over-emphasised. lections and electoralanalysis ave always been of high priority n an ydemocratic ountry. n this context, olit icalecon-omics with public choioe has a lot to offer rnincreasing ur understanding f the voter's chorcebehaviour n elections. n the arena f multi partyelections, he presence f electoral lliances s themainstay 1' emocratic overnance.

    In order o test br the basics f the presence fvoter's choicc behaviour, we flrsl look at theprobability f an ncumbent etting e-elected romthe same constituency . urther, n this study, wemake an attem pl at examining he behaviouralpattern ollowed by other electoral ariables pan-ning the electoral erms. By means of extensiveempirical nalysis , he study explores irst, whetherthe national parties have better odds of getting e-elected han the regional parties or vice-versa.

    Second, does the sex of candidates make anydifference in electoral outcome? Third, doesincreased oter-turnout n any election mean anti-incumbency oting behaviour? ourth, oes a argenumber of candidates ut on the vote share f therncumbent, esulting n losing he election?

    The remainder 1'the tud y s organised s bl-lows: We discuss he existing esearch n this areain Section , ollowed by the ndian scenario nd tselectoral ystem n Section . We propose simpleeconometric odel and explore conometric ssues

    JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY JULY.SEPT', 003

    that arise n the specification f the model n Section4, lbllowed by the empirical esulrs n Section .with discussion n some nteresting mplications.Section concludes he work.

    LiteratLrre eview

    In this study, we model the winnings o1' anincumbent arty. We assume hat all he candidatesare qually ikely o be elected i.e., here s a andomselection rocedure f candidates), nd he el'fectson outcome are not individual specific but partyspecific Which s not a very estrictive ssumption,as most of the electoral esults are driven by theincumbent's party rather than the incumbentthemselves, arring some xceptions.).We estimatethe probability f winning of the ncumbent arty(i.e. , he party, which has also won he ast lection .irrespective f whether t has been a member of 'coalition (during/afier elections), he governin_eparty any. or one of the members f the governingparty, n case f a coalition overnment. or, we ar einterested n taking account of constituency-wiseruling winning) by parties ather han he uling atthe center. We estimate he average ro bability fwinning, given that the party has won the lastelection, sing onstituency-wise ata f he gener-aleleclions iom 1967 to 1999 under rhe Indianelectoral ystem, sing a Panel Data Binary ChoiceModel (Generalised stimation quations opula-tion Averaged Model). This exercise ives us theidea f he ncumbent's erformance nd ts chancesto come nto power again, nder he assumption frational oters n the constituency whrch s verywidely used n existing political economy itera-ture).

    The probability hat a party wins the elecriondepends n various actors, iz. , he state o whichthe constituency elongs, he nherent haracteris-tics of the population, and various other socio-economic ndcultural actors caste, eligion. tc.).One of the potential actors hat might aff'ect hewinning of a party s that he same party has wo n

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    3/14

    vot.. t5 No. 3

    the ast election. his seems uite a probable aseand ndia, with its socio-politi cal et-up, rovidesan ncentive o est he hypothesis hathaving ccessto power aff-ects he outcome f elections n futureor does not.

    Researchers ave done significant amount ofwork in this area mostly with data he

    US or fromthe European ountries. atz, Honaker, nd King[2001] mprove pon heir originalmodel, Katz an dKing, 1999, p . 5-321 n bias , ar iance, umer icalstability, and computational speed, making itquicker o estimate. he dif{'erence ere being hatthe problem of ef1'ective oting, which is treated smissing ata n the original model. They argue hatordinary east squares OLS) regression s inap-propriate when, the dependent ariable measuresthe share of the vote going to each party, andrecommend superior echnique. Unfbrtunately,

    theKatz-King

    model requires high level of sta-tis ticalexpertise, nd s computationally mpracti-cal fbr more than three political parties, s thcirmodel s based n the principle of the wo politrcalparties ompetin-{ 'or he constituency eat. ackson[2001] ries o give a simpler stimation rocedure,using a seemingly unrelated egression SUR)model with a multinomial ogistic unction).

    There are some elatively ecent tudies hat gobeyond he examination f particular ases, ndassess he impact on national elections. Conley

    [2002],using he 1990s ensus nd electoral ata fthe Canadian idings, inds significant ncumbencyeffects or members of parliament MPs), and noevidence f any significant ncrease n incumbencyadvantage n the last decade. Pappi and Gabriele[1998. Pp. 229-55], using survey data or the Eastand West Germany, roposes ow party preferencesof voters n a multiparty system are measured,compared, nd aggregated. ormal, tal, 11998,Pp.251 93) considered stochastic model of vorerresponse, ased n multinomial probit estimation,using Euro-barometer urvey data and European

    P ARLIAM ENTA RY ELECTI ONS AN D P O L|TI CAL P A RTIES

    elite survey data iom 1979. They argue hat max-imisation of expected ote, n general, s not arational party strategy in multiparty politicalsystems, hich are based on proportional eprc-sentatlon.

    The development f methods o examine mul-tiparty elections as clearly caught he attention fthe profession/discipline, udging by the number frecent papers n the opic ollowing the nitial Karzand King [999, Pp. 15-32] paper. Tomz, et ol,[2002, Pp. 66-83], make importanr sreps n thisdirection, s heir model s based n standard inearmodel and offers a sophisticated et convenientalternative that involves seemingly unrelatedregression SUR). Gelman and King [1990, 1994,Pp.5 4-541 nalyse lections n all state egislatureselected rom a single member district over he period1968-88 n the US and conclude har edistrictingincreases lectoral esponsiveness , nd t leads oless ias owards ma jority of voters han would beexpected fno redistricting ad occurred. efore heredistricting, n case of large legislatures, herepresentation f voters was biased owards pow-erful representatives, hich leads o unequal rs-tribution of power among voters. Katz and KingI I 999, Pp. 1 32],King, er al, [200 , Pp. 49 69]an dGelman nd Katz 200 develop methodologies oestimate he effbctive' vote n partially contested(re-elections), ultiparty lections. hey rear ar-tially contested lections s missing data probleni.and use an algorithm procedure to recover

    contesting arties' stimated ote shares. e don'tsee any reason o expect hat, had a party contestedelections, ther arties, ote hares ould have eenchanged ecause, fa party s not contesting iom aparticular onstituency, t must be the case hateither he party s contesting he election ndirectly(may be a coalition), r n the party's assessment, tdoes not have any chance of winning from thatconstituency. errill [1994] rovides videnoe hatboth the pure proximity and the pure directionalmodels fall short of accounting br the spatialdistribution f voters by party. Gelman and Katz

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    4/14

    [2001], gives he strength f voting, giving us aninsight of the nter-link ages etween oting, coali-tions and he electoral olleg e. However, no workhas so far been done n greater etail o explain hewinnings of any party based on their base an dperformance n the constituency -ruling ather ha nthat n the ruhng at center.

    Though survey esearch as enormous dvan-tages or studying ndividual-level ref'erences, etas analyses of random selections of isolatedindividuals rom unknown geographical ocations,they necessarily iss much of electoral olitics. Assuch, hey are often best omplemented ith studresof aggregatc lectoral eturns.

    Indian Scenario

    The electoral ystem flndia is argely based nthe British pattern. The Constitution id not provide

    any details about he electoral system, and eft it toParliament o determine. ccordingly, Parliamentpassed number of laws to regulate he electoralsystem. Some prominent aws enacted or thispurpose nclude he Representation f People Act,1950 and 1951, the Presidenti al and Vice-Presidential lection Act. 1952. and he Delimita-tion Act, 1952, etc.

    India emains y far he argest emocracy n heworld. with almost 600 million voters. ndia'sparliamentary overnment nd the First-Past-the-

    Post FPTP) electoral ystem are a legacy of theBritish colonialism, which ended n 1947. TheIndian Constitution rovides hat all adult citizens,who are l8 years or more of age, and who are nototherwise debarred rom voting, can exercise heirright to the franchise. oters elect a 543-memberLok Sabha the ower house), rom single-memberconstituencies. nd each of India's states hasadopted similar system. y contrast, embers fthe lower house of Parliament nd of the statelegislative ssemblies embers f indirectly electRajya Sabha r he Council f States, swell as hose

    JULY.SEPT.2OO.]

    of the corresponding pper houses f th e states.There are also he non-executive) resident nd heVice-President, lected by the members of Parlia-ment and state egislative ssemblies.

    General lections re held once very ive years,but he President ay dissolve he Lok Sabha n headvice f the prime minister efbre ts erm s over.as n 1971 nd n 1997 r, fhe or she s convincedthat no stable overnment an be ormed, s n 199and 1998. he prime minister holds office as ongas he or she an command majority n Lok Sabha.All the successive overnments f the Congressparty, which ruled ndia continuously ntil 1977.servcd a full term in office. Since 1977, govern-ments ave been ess table, nd a number f primeministers ave ad o es ign, saresult fparty plitsor no-confidence otes efore omoletine heir ullterm.

    The major effect of the electoral ystem, t eastuntil 1977, was o guarantee ajority governmentsbased n a majority of voter suppor-r. he FPTPelectoral ystem esulted n the ruling Congressparty securing stable majorities n Lok Sabha.usually gainst fragmented pposition. owever.since 1977, when he opposition arties ombinedto form coalitions and started putting up commoncandidates gainst he Congress andidates as wa sthe case nrhe 1911 nd 1989 eneral lections). heCongress ajorities ave vanished. oleover, henature f the system s that small changes n vote

    share often have a dramatic mpact upon he shapeof the resulting Parliament.

    The overall results of elections o Lok Sabhahave never been proportional. ecause he candi-date, who obtains he most votes polled but notnecessarily majority of voters s declared lected,support an often be divided by setting andidatesof the same aste, eligion, or legion against achother. However, despite he divided nature o1India's multi-ethnic emocracy, he electoral ys-tem has etained considerable egree f support,

    JOURNAL F INDIAN CHOOL F POLITICAL CONOMY

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    5/14

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    6/14

    6 3 8

    economic, nd cultural bundations, hat distinguishthe ndian states nd egions iom one another, mayexert a s ignificant nfluence n electoral utcomes.

    The model proposed ere s:

    (wins),, cx.0 Br(national arty),, Br(reserved),+ p.,(sex.y,, B.(electors), , B.(valid otes) ' ,+ p(,(vorers), , pr(candidates),,+ Br(duration),, Br(state), Br0(time), u, l

    We have taken the dependent ariable as adichotomous ariable wins), which akes wo val-ues, .e. , if the ncumbent arty wins he election,and 0 otherwise. he explanatory ariables sed nthe model are:

    L Voters' urnout n the constituency poll per-centage) s the percentage urnout of voters n thatelection, howing he effect of the public desire ora change r the extent of anti-incumbent oting.

    2. Categorical variable or the category of theconstituency reserved).

    There s general erception hat eservation f aparticular eat can give undesired avour o somepolitical parties ompared o others, or example,regional parties may have better penetration nreserved eats han national parties. This variabletakes alue of 1, f the seat s reserved, otherwise.

    3. Categorical ariable or the sex of the candi-dates scx) .

    Indian case, t is assumed hat males have mclreaccess o power than women, which gives he mundue avour. This variable controls he eff'ect ofsex discrimination. t takes alue of l , if the winnerand oser are of the same ex, 0 otherwise.

    JULY.SEP7 . . OO. ]

    4. Dummy variable or the party being a nationalparty or not (national party).

    This s to control he national arty's ability oget more esources nd ts access o power, whichcan nfluence n the esult . One may argue hat f acandidate s from National party, his/her promiscsare more ikely to be ulfil led, compared o others.or candidates may get the thvour due to thefavourable cenario or the oartv.

    5. Duration s the gap between wo consecutiveelections easured n years, o control or he effectof time gap o fulfil l one's promises efore goingfor the next election.

    6. Valid votes comprise he ratio of total validvotes olled o he size felectorate, .e. . he variableto control br hc education knowledge,)evel n theconstituency. We assume here that the knowl-

    edgeable erson would caste his/her vote withproper aution, educing he atio of invalid votes.It is a proxy used o control for the knowledge ofvoters n the constituencv.

    7. Candidates omprise he number 1'candidatescontesting. his variable ontrols br the eff 'ect fcompetition n the constituency. ne may reat hlsvariable s a control br the dummv candidates.

    8. Time dummies o capture he effect of currentsocio-economic-political cenario t the time of

    elect ion n he outcome.

    This is to capture he eflect of the currentscenario, ather han he effect of the uli ng period(which s captured hrough uration). or example,in year 1980 we find that the year dummy issignificant nd negative his could be explained sthe after effect ofthe emergency eriod hat esultedin the esults gainst he uling party Congress.

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

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    7/14

    voL. 5 No 3

    9. State ummies o control he unobserved tatecharacteristics.

    As the dependent ariable s dichotomous, eare using the Binary Choice (Logit) model. Weestimate population-averaged anel-data modelsusing GEE. This approach stimates ross-sectionaltime-series odels. n particular, t estimates en-eral linear models, and allows specrfying withingroup corelation structure fbr panels. Thisproduces alid standard rrors, v en f the correla-tions within group are not as hypothesised y thespecified orrelation tructure.

    For a thorough discussion f GEE in the esti-mation of GLM (Generalised inear Models), see,Zeger and Liang, 1986(a)(b), Pp. 13-22, Pp .121- 301. urther nformation n inear models anbe found in Nelder and Wedderburn 972, Pp .370-3841. inally, there have been a number ofilluminating rticles n various pplications f GEEin Zeger, Liang, and Albert [988, Pp. 1049-60],Zeger and Liang [1986(aXb), Pp. 13-22, Pp .121-1301, nd Liang [987, Pp. 695-702]. ender-gast tal. , I996, Pp. 89- I8] provide nice surveyofthe current methods or analyzing clustered atain regard o binary esponse ata. Our implemen-tation bllows hat of Zeger and Liang 986(a)(b),Pp. 3-22,Pp. 2 l - I 301 hichhasbeen ummarisedin Appendix. br the convenience f the eader. Wegeneralized inear models f Y, , with covariates 1,,

    Empirical Analysis

    This section s divided nto two Sub-sections:sub-section focuses n he data whereas moiricalresul ts reprcscntecin Sub-sect ion .

    Data

    The data used or the analysis s obtained romlhe Reports of Election Commission of India. Thedataset consists o[ general election data from1961-1999. he first election-year or our study s1967. The data from 1967 s used o folm rhe

    PARLIAM ENTARY ELECTIONS AND POLIT]CAL PARTIES

    required ariable ulns, s he prior election ata snot readiiy available or each constituency, ue oreconstruction f the constituencies. t leaves uswith the election ata rom 1971-1999 i.e.. 1771,1911, 1980,1984 1985*) , 1989, 1991 1992*) ,1996, 1998 and 1999). his, accounts or l0 elec-tions out of to tal 3 held so far. We haven't akendata befbre 1967 lections, s here as been majolrestructuring f the boundaries f constituenciesbefore 97l election.

    Thus, we end up with 4,143 constituency-elections, which consist of minimum five andmaximum nine elections er constituency, s lnsome onstituencies, lection ad not been held artimes. n some tates ad not been held elections tthe ime of general lections ut later on we havetreated; hose elections nly have aken place at hesame ime. For example, n Punjab hat he electionfor 1984 ook place n 1985. n such ases. e havetreated hem as fthey had been hcld n 1984 only.

    Entpirical Results

    Table represents he basic haracteristics f theIndian Lok Sabha elections n terms of number ofcandidates, ontestants orfeiting deposits, umberof women candidates, andidates elonging o dif-f'erent political parties in terms of national orregional). he number of candidates ontesting nelections has increased ill 1996. after which itstarted eclining. n 1957, here were ,5 9 candi-dates ontesting or the election f 494 seats, hilein 1996 or 543 seats otal of 13.952 andidatescontested. hese ncreasing umbers f contestantswere mainly rom ndependent ategory, nd mostof them could not save heir deposits. eposits refbrleited f the candidate s not able o bag even 3percent f votes polled. One may algue hat hesecandidates ere used sdummy candidate o cut onrival candidates ote share n the base f religion,sex, caste and other socio-economi c actors. n1951, 360 candidates rom independent ategorylost heir deposits rom the otal of 533 candidates;while in 1996, out of 13,952 andidates. 2,688candidates' eposits ere orfeited.

    6-19

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    8/14

    64 0 JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY

    Tab le

    JULY-SEPT.2AO3

    Can d id a te s0 )

    Election Year( 2 )

    N P(3 )

    S P(4 )

    RUP( 5 )

    IND( 6 )

    WC(7 )

    Total( 8 )

    ContestedWo nForfeited Deposit

    ContestedWo nFort'eited Deposit

    ContestedWo nForfeited Deposit

    ContestedWo nForferted Deposit

    ContestedWo nForf-eited eposit

    Contested

    Wo nForfeited Deposit

    ContestedWo nForfeited Deposit

    ContestedWo nForfeited Deposit

    Co n te s tedWo nForfeitcd Deposit

    ContestedWo n

    Forfeited Deposit

    ContestedWo nForfeited Deposit

    ContestedWo nForl'eited Deposit

    ContestedWo nForfeited Deposit

    1 9 5

    1957

    t962

    t96'7

    197

    t97'7

    t9 8 0

    I984*

    I9 8 9

    l 9 9 l

    1996

    l 99u

    t999

    | 1 t ' 7

    4 1 8344

    9 t94211 3 0

    1.269440362

    t,34244039 0

    t .2 2 34 5 135 9

    1.060

    4 8 11 0 0

    1 , 5 4 14 8 5444

    l ,3 0 74623 8 7

    1 , 3 7 811421

    I , 8 9 24908 5 2

    l , 8 l4038 9 8

    t.49338163 1

    | ) ao36 943 1

    1 4 32112

    52 1) l

    35 6

    4 5 11 2 7167

    17 1t 0 l20 7

    75 01 5 84 l t

    7068

    l 326

    203l 3

    1 5 6

    70

    360

    1 5 6I

    l 5 t

    1260

    120

    926l 9

    8 6 8

    8421

    82 0

    1.0481

    I,020

    8 7 119

    74 4

    b) 4

    l 0o /+

    53 331

    36 0

    4 8 1r' 1

    324

    47 920

    37 8

    86 63 5

    747

    I 1 3 4t 4

    1 ,0 6 6

    t . 2 2 4

    9I , 1 9 0

    2,8269

    2,794

    3,894l 3

    3 .8 3 0

    3 , 7 3l 2

    3.672

    5.606I

    5.590

    1 0 ,6 3 59

    10.603

    1 , 9 t 56

    l , 8 9 8

    I,9456

    l , 9 2 8

    452 8

    6

    663 lt 9

    o /

    29I 4

    80

    43

    70

    l 93 t

    I 132 879

    t7 l+-1

    1 0 9

    1 9 82 9

    l l 5

    3 1 84)

    l 8 l

    5 9 940

    5 1 8

    2 1 4;+3

    l u 4

    28449

    1 8 3

    I ,8 7 4,+8914 5

    1 , 59191494

    1 . 9 8 54948 5 6

    ) 1 6q

    5201.20_1

    2..7845 1 8

    t . 7 0 7

    2.439

    5421 .3 5 6

    1.629529

    3 .4 1 7

    5 4 q l

    5.r4 .3 8 2

    6 , 1 6 0529

    5.003

    8 . 8 6 1547

    7 .6 1

    1 3 .9 5 254 3

    1 2 .6 8 8

    4.75054 3

    3.,+86

    4,64854 3

    3,400

    12434

    l l 93 l40

    z t 72 8

    1 0 8

    1 4 8

    5 8

    ) a A

    40t26

    8 5

    496

    1 0 6t+

    28

    1 6 5664 5

    * Includes he data or the elections happened n later years n some states.NP: National Parties. SP: State Panies, RUP: Registered Unrecognized) arties, ND: Independents. C: Women Candidates

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    9/14

    vqL. 15 NO.3 PARLIA M ENTARY EI,ECTI O NS AN D P O .I7'I CAL P ART ES

    ooo,oG

    o

    os

    1 0 09080-7rlO U504030

    Figure l Winnings y National arties

    1970 t v / 5 1 9 8 0 1 9 8 5 1 9 9 0

    Election Year

    1995 2000

    oGo@

    clo

    obS

    Figure 2 Incumbent's Winning

    1970 1975

    OU

    40

    zv

    0

    1IL

    1 9 8 0 1 9 8 5 1 9 9 0

    Election Year

    199 5 2000

    oGoo(

    oo;s

    1 0 51 0 09590B5BO

    Figure 3 Sex-wise istribution

    1980 1985 1990

    Election Year

    197 0

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    10/14

    We also observe ncreasing participation offemale contestants , lthough till the participationis limited n nature. n 1951 election. 5 womencontested or the election, ut of which 28 won an donly 8 losr rheir deposits . n 1999 election 28 4women ontested,49 on and 183 ost heirdepositsi .e. out of 4,648 candidates nly 284 were womenand only 49 got elected n 543 seats arliament.

    Figure shows hat he national arties s. othershave performed well nitially winning about 80 percent of the otal seats . hich dropped ramaticallyin 1998 lection, hcy ecovered n 1999 lecrion o70 per cent oftotal seats. n Figure 2, we show hepattcrn f incumbents etting e-eleoted, he cycli-cal behaviour of the percentage f seats won by theincumbent s apparent ere, his also eflects n theregression esults. we flnd that he year dummiesfor the third and folth period s negative nd srg-nificant while for all other years s positive an dsignificant. Which shows, there wereanti-incumbency ias in those periods. n 1984election, Congress won almost all the seats n theparliamentary lectron hich may be nfluenced ythe sympathetic behaviour of the voters due toassassination f Indira Gandhi. his effect may no tbe captured through the duration variable. Asduration would be same or 1989 lecrion 1984-89)as wel l as for 1996 elect ion 1991-96) ut thesocio-economic-political scenario would beentirely differcnt. n this case also he coefficientfor the year dummy is negative nd significant,

    implying an anti-incumbency otingbehavior f hevoters y and arge. or rest ofthe election ears hecoefTicient or the year dummy is significant ndpositive, which implies that ther e were positiveincumbency ias or the ruling parties. his is asexpected ince he parties n the power have bettercontrol over the f'actors which may influence heoutcome f the elections. his rend n the electionalso helps n explaining he cyclical ehavior f th epercentage f seats on by he ncumbent arty.Onecan clearly see that the coefficients ol' the timedummy with reference o the year 1977 or the years

    JULY-SEPT. OO.]

    1980 and 1984 s negative nd significant. hi simplies here s an anti-incumbency ias n theseelection ears, which esulted n the cyclical beha-viour of the seats won by the national parties. n1980 election ue o the emergency n the earlierperiod ed to negati ve mpact on the incumbentparties' performance. One other explanation canalso be provided o explain his cyclical

    behaviourof the ncumbent's inning. One may argue hat nthe nitial elections he winning parties enjoyed hepower but didn't perform well which esulted n thenegative mpact on their performance nd resultedin the downward trend in the winning of theincumbent arties. Whereas n the ater period fter-1980 election he incumbents ained heir voters ,trust and worked or the well being of the voters.This resulted n an ncreased rend owards winningof the ncumbent. his rend ontinued ill the 199election. n 1996 election, due ro high volatilepolitical and economic cenario esults are againagainst he ncumbents. ne may argue hat oncethe ncumbent arties re n power br onger er-iodtheir performance oes down and then n the nexrelection hey oose. he voters ive he power o heothers or providing better governance. igure 3shows, ow the sex of opposing andidates iffer.ln 1911 election, most of the winning and osing(here we define osing andidate s unner up can-didate) candidates re of the same sex, which hasreduced o 87 per cent rom 100 per cent n 1999election.

    We now shift our attention o the outcome f ourregression nalysis . he esults fthe egression rereportcd n the Table 2. We find thar he sign of thecoefl'icients f sex. duration nd poll percentage snegative, nd he sign ofnational artyand eservedcandidates re positive. All of the state dummieshave negative coefficients except Jammu an dKashmir, Mizoram and West Bengal.r he coeff -cient of state ummy or Jammu nd Kashmir s notsignificant, hereas t is signifioanr or Mizoramand West Bengal, this significant and posttivecoefficient ould eflect he persistence fthe sarne

    JOURNAL OF INDIANSCHOOL OF POLIT CAL ECONOMY

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    11/14

    Y O L. 5 N O .3 PARLIAM ENTA RY EI,ECI' O NS AN D P OL]T CAL P ARTI ES

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

    61 3

    Vanables( l )

    Odds Ratio( 2 )

    p > lz l( 3 )

    Coefficients( 4 )

    p > lz l( 5 )

    National PanyReservcdSexDurationPoll Percentage

    Valid VotesCandidatesConstant_Andhra PradeshArunachal Pradesh

    AssamB i h a rCaprtal TerntoryDe lh rGo a

    GujratHaryanaHimachal PradeshJammu and KashmirKarnataka

    KeralaMadhya PradeshMaharashtraManipurMeghalaya

    MizoramNagalandOrissaPunjabRajasthan

    Tamil NaduTripuraUttar PradeshWes t Ben g a lYear I 97

    Yea r I 98 0Yea r 1 9 8 4Yea r 9 8 9Year I 99Year I 996Year I 998

    2.121444*1 .2 8 3 8 9 *0 .9 5 7 8 3 20.1033292*0 .8 2 9 6 3 6 1

    1.36E+08*1.000599

    0.6805947*0.3884401

    0 90857.10 3 I 6 5 0 0 1 *0 6 645990.33764t7*0 l 5 3 l 4 +

    0.4793286*0.2 59504*0.2825454*14605520.7541362

    0.9820250.49 0497*0.4564442*0.569052-5*0.9128536

    I 96 654+0.57466010.35074*0.261497'7*0.23t741t*

    0.92858220.10157 4*

    0.2483291*2.323487*| .7469424

    0.093048*0.18691t?-42.079232*1.720078x3.6 6483*l . 7 2 4 l t ]E

    0.0000 0040 7 l 90.0000.001

    0.0020.847

    0 0040.000

    0.1790.0000 .3 8 90.0000.000

    0.0010.0000.0000 . 1 9 90 l t 3

    0 9 3 60 0 0 10.0000.0000.904

    0.0000.0000.0000.0000.000

    0.5720.007

    0.0000.0000.000

    0.0000 0000 0000.0000.0000.000

    0.75209680.2498942

    -0 0430828-0.35 9302-0. 86768

    l 8.72680.00059840.'7265'703

    -0.3 47 82-0.9456164

    -0.0958789, l I 5 0 4 3 2-0..183769-1 08577-t.876572

    -0 73-53689-t 532106-t.2639t60.378847

    -0 282 23

    - 0 0 1 8 I 3 8 4-0.7 l2099-0 7842888-0.5637825-0.02752t7

    0.6737882-0.-5-s39766-t.041t- 1 . 3 4 1 3 3-t.462t34

    -0.0140964-0.3544255-13929980.8430690.557867

    -2.374639- 67680 .7 3 9 9 8 40.54236961 2855020.54483 I

    0 0000 0040 . 7 1 90.0000 .0 0 1

    0.0020.8470 0 4 10 0040.000

    0.1790.0000 .3 8 90 0000 000

    0 0 0 10.0000.0000 . 1 9 90 l t l

    0 .9 3 60 0 0 10.0000 0000.904

    0.0000.0000.0000.0000.000

    o.5720.007

    0.0000.0000.000

    0.0000 0000 0000.0000 00 00 000

    The exchangeable orrelation the observations ithin a subject are assumed o be equally correlated) oes not in generarprovide random effects but this gives the same kind of structure hat anses n a normal inear model with random ntercepts. tactually its an equal-correlatron opulation averaged model that s equal o the random effects model for linear equation. Th edistribution ofdependent ariable s assumed o be ofbinomial family with logit ink. It acknowledges hat correlations xrstwlthlna subject, ut it assumes hat he correlations re constant ver lme.* indicates ignificance t 5% level of sisnificance.

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    12/14

    VOL. 5NO.3 PARL|AM ENTARY ELECTIONS AND POLITICAL PARTIES

    P ( X , , 1 = P ( Y, , = 1 X , , ) P ( Y r , = I X r , , . , , . , , . . . . , k , )

    The underlyrng atent ariable model s:

    Y "= B'X + c , w h ere Y = I i f Y- > 0

    Here Y is the dependent ariable winning), nd e s a continuously istributecl ariable ndependent fx anclthe distribution f e is symmetric bout ero.

    In the GLM approach see McCullagh nd Nelder, 19891, e assume ha r

    h1p,,)xi FVar(y,. ,) C(prj)O

    [ , = E (y, ) h - ' rx ] ,F r.n .' (* : , , 8 ] '

    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 - 2 2 , p p .2 l - 1 3 0 1 , e r y, ( y i . r , K y, . , , ) ' r b e r h e n , x

    I vector of outcome values, and et x, = (x,.,,Kx,.",)t e the n, x p matrix of covariate values br the irh suhiect -

    1, . . . ,m. We assume hat he marginal ensity br yi. ;may be written n exponential amily notation s

    | ( y,i )= cx p [{y rr 0 , , a (0 ,, ) = b (y r. , ) 0 l

    where , . , h(L.,) , r; = x,. ,B. nder his brmulation. he irst wo moments regiven by

    E(y ' J = a ' ( 0 , ) , v a r (y, . ' ) a" (0 , ) i0We define he quantities assuming hat we have an n x n working correlation matrix R(a),

    A ,= d i a g { d 0 ,, , d 4 , , t n r n m a t r i x

    n , - d i rg {a t0 ,, t} n . r n mar r ix

    5 , = y, a ' ( 0 , ) n x I m a r r i x

    D = A , A , X , n . p m a r r i x

    v , = A ' t R { c r ) A " n x n m a t r r x

    such hat he GEE becomes

    v ru) D, ' ,v,

    S, = (r

    We then have hat

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    13/14

    64 4

    parties n these states, while the negative oetfi-cients for some other state dummies reflect anti-incumbency ias at work n (allor some f) he otherstates.2 he estimated oefficients f national arty,reserved andidates, ollpercentage ndvalid votesare all statistically ignrficant t 5 per cent evel ofsignificance . e find that he og odds of winning

    are higher f the ncumben t elongs o the nationalparty. One nteresting indings of our analysis s nterms of the coefficient of polling percentage: otespolled enters with a negative ignificant oefficient,which suggests hat as he voters urnout ncreasesin a constituency, he probability f winning of theincumbent ecreascs. .e. , more voter urnout sig-nif ies an ant i - incumbency ol l ing .This might bedue o the fact that hose voters. who would havebeen otherwise ndiffbrent by not going o castetheir vote), have disliked the incumbent 's per-

    fbrmance and get he motivation o vote against heruling party br effecting he change. We also ind

    JULY-SEPT,2OO.]

    that he sex and he number of candidates o nothave significant ffect on the probability of theincumbent inning he elect ion . '

    ConcLusion

    Using he data rom Indian parliamentary lec-tions,we show hatcandidates fthe national artieshave better hance f winning again rom he samcconstituency. n increase n the poll percentage na constituency exerts an adverse etfect on theprobability f the ncumbent inning he election.Voters are willing to cast their votes of others,reflecting he poor performance f the ruling partyin the particular onstituency, hen hey want herequisite hange n the government. ne nterestingway o expand ur work s to relate his analysis nterms f various emographic haracteristics, s ingthe Census ata. One may also ry o explainwhetherthe public goods provision an explain he proba-bility of winning election. However, hese re eftto future esearch.

    JOURNAL OF INDIANSCHOOL OF POLITICAL ECONOMY

    APPENDIX

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

    l, 2, ...n,,where here are n, observations br each group dentilier . In the above, g{ } is calledF the distribution amily. f g{ } is the ogit unction nd Y,, s distribured inomial, we have

    = X,,0 , Y - binomial.o g i t E (Y, )}

    The ink function ogitis defined s n{y/(l-y)}, the natural og of odds. We assume he working correlarionrnatrix or modeling he within-group orrelation, square max{n,} x max{n, matrix, say R. Then he correlationstructure correspondrng o equal-correlation odels) s defined s:

    i f

    R,..=

    otherwise

    fo r = l , 2 , . . n n d =the ink function an d

    In the GEE approa ch he unobserved ft-ectsconditional nly on X,, with the esult hat we have

    binary response model, he response robabilities are specifiedthe fbllowine:

    The response robability s,

  • 8/15/2019 Parliamentary Elections and Voting Behaviour

    14/14

    6.t6 JOURNAL OF INDIAN SCHOOL OF POLITICAL ECONOMY JULY.SEPT 00.]

    B,-, F,where he term

    is what we call the IRLS variance estimate iteratively eweighed east squares). t is used o calculate he standarderrors.

    I oto,,o,'0:n,t0:i'l,iol,B,,o'',p.rrs0r

    'lioltF:v;'o,D,(i)i

    NOTES

    l. We have also performed the regression with regionaldummies instead of state dummies, but qualitatively resultsreported n Table 2 remain unaltered.

    2. Thanks to the anonymous referee for providing thisexplanation.

    3. Introduction of the lagged values of the dependent arr-ables o not change he results ignificantly. The only noticeablechange s the significance of duration, which turns out to beinsigniticant afier ntroduction of lagged values n the rnodel.

    RI,FI'RENCES

    Co n ley, S .R . . 2 0 0 2 ; Co n s t i tu en cyCo n tex t s , n cu rn b en cy, n dParty Competition n Canadian Ridings n 1990s', Univerrr4,ol Murthnd Working Puper.

    Gelman. A. and G. King, 1990; 'Estimating the ElectoralConsequences f Legislative Redistricting', Jountul rlAme icun Stu s it ul As soc ut kttt.

    Ce lman . . an d G. Kin g . 9 9 4 1 A Un if iedMeth o d fEv a lu a t in gElcctoral Systems and Redistricting Plans', AtneritonJo t r rn u l . f Poli t t tu l S t ien ce . o l . 3 8 . No . 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 Vo t eCount? Voting Power, Coalitions and heElectoral College',Sot:ittlScience Working Puper I ,l2l , California nstitute ofTechnology.

    Hin ich ,M.; J . Valer i , P.C . Ord e r sh o o k , 9 9 9 ; Uk ra in e ' s 1 9 9 8Parliamentary Elections: A Spatial Analysis', Post-SoyietA l J h i r s , Yo l . 5 ,N o . 2 .

    Jack so n . . 8 . . 2 0 0 1 I A Seemin g ly n re la ted eg re s s io n o d e lfbr Analyzing Multiparty Elections', Annual SumnterP o Lt: L M e hotlo o gt' Co nle e nce, Emory University.

    Ka tz . J .N . l J . Ho n ak e r an d G. Kin g , 2 0 0 11 An lmp ro v edStatistrcal Model for Multiparty Electoral Dara', . locialSt'rence Working Puper I I I | , Califbrnia Institute Of Tech-n o lo g y.

    Ka tz , .N . an d G. Kin g , 1 9 9 9 : A Sta t i s t i c a l o d e l o r Mu lt ip a r tyElectoral Data' Anrericur Journul of Political Science, ol.9 3 . N o l .

    K in g ,G. ; J . Ho n ak e r, . Jo sep h n dK. Sch ev e . 0 0 I An a ly z in g

    Incomplete olitical Science ata: An Alternative Algorithmfor Multiple lmputation'. Americut Politicul SL:ient:eRev leu ' . o l . 9 5 . No . L

    Lrang, K.Y., 19871 Estimating Functions and ApproxtrnateConditional Likelihood'. Biometriku, Vol. 1.

    McCullagh, P. and J.A. Nelder, 19891 Generalised LinearMo d e ls ' , 2 n d d n . , Ch ap man Ha l l , Lo n d o n .

    Merrill, S. IIl. , 1994; A Probabilistic Model for the SparialDistribution of Party Support n Multiparty Electorates'.J ountu ol A nte i c rn S u s t ' ul As,t c tLt on.

    Ne ld e r, J .A . an d R .W.M. Wed d e rb u rn . 1 9 7 2 ; Gen e ra l i s cu

    Lin ea r Mo d e ls ' , Jo u r ru t l of the Rrl l i l S tu t i s t i cu l o d en .

    Se r ie . r . Vo 1 . 1 3 5 .No rma l . S. ; M.D. An d rew, Q.M. Kev in an d W. An d reu , . 9 9 8 ;

    'Multiparty Electoral Competition n the Netherlands ndGermany: A Model Based on Multinomial Prcbit', PubtkChoice. Yol.9'7

    Pappi, F.U. and E. Gabriele, 19981 Voter's Parry Preferences nMultiparty Systems and Their Coalirional and SpatialIrnplications: Cermany after Unification'. Public Choite,Vo l . 9 7 .

    Pen d e rg as t . .F. lS . J . an g e ,M.A. Newto n , M.J . L in d s to rn i .M.Pa t la an d M.R. F i sh e r, 1 9 9 6 : A Su rv ey o f Me th o d s b rAnalyzing Clustered Binary Response ata', InternutionulSn u i . t t i cd ev re r r, o l . 6 4 .

    Tomz, M.; T.A. Joshua nd J. Wittenberg,2002:'An Easy an d

    Accurate Regression Model for Multiparty Electoral Data'.Politiul Anulysis, ol. 10, No. l.

    Wooldridge J.M., 20021 Econonetri( Anah'sls l Cxtss Setrrortund Panel Dutu, The MIT Press.

    Zeg e r, S .L . ; K .Y. L ian g an d P.S . Alb en . 1 9 8 8 : Mo d e ls fo rLo n g i tu d in a l Da ta : A Gen e ra l i s ed s t ima t in g Eq u a t ro nApproach', Biometric.t, ol. 44.

    Zeg e r, .L . an d K.Y. Lian g , 1 9 8 6 a ; Lo n g i tu d in a lDa ta An a ly s r sUsing Generalised inear Models'. Biometriku. Y ol. 73.

    Zeg e r, .L .an d K.Y. L ian g , 1 9 8 6 b : Lo n g i tu d in a lDa raAn a ly s r sfor Discrete nd Continuous Outcomes' Biometrics. Vol. :12.