--
Midterm Elections and Divided
Government:
An Information-DrivenTheory
of Electoral Volatility
STEPHEN P NICHOLSON, SANTACLARAUNIVERSITY
GARY M. SEGURA, CLAREMONT
GRADUATE
UNIVERSITY
Divided governmentaffects individual choices over how to vote in
midtermelectionsbecauseit increasesuncertaintyin the minds of voters.
Particularly,divided control of government makes blame attribution
more difficultby obscuringcausal connections and reducingthe overall
amount of usable information.As a result, we argue that under divided
government,voters are less likely to vote for the House candidatenot of
the President'sparty.Using both NES and election-specificcontextual
data, we examine divided government'seffect on the voters' political
knowledgeand candidatepreferencesin all midtermelections from 1978
to 1994, controllingfor well identified factors that shape outcomes in
House elections. We find, first, that divided government reduces the
amount of politicalinformationheld by voters. Second, divided government helps the President'spartyby loweringthe probabilitythat an individual votes for the out-partycandidate.
The 1994 election resulted in one of the more dramatic changes of political
fortune. In two years, the Democrats went from winning their first presidential
election in 16 years to their largest midterm loss and, more importantly, the loss
of control of the House of Representatives for the first time in 40 years. Was 1994
NOTE: An earlierversion of this articlewas preparedfor presentationat the AnnualMeetingof the
MidwestPoliticalScienceAssociation,April 10-12, 1997, Chicago,Illinois.Authors'names
are presented alphabetically.The authors wish to thank Stacy Gordon, RichardFleisher,
PatrickSellars,and the editors and anonymous reviewersfor their helpful comments, and
NathanWoods for his effectiveresearchassistance.Remainingerrorsare,of course, our own.
Vol. 52, No. 3 (September1999): pp. 609-629
PoliticalResearchQuarterly,
609
PoliticalResearch
Quarterly
just simply the resultof the known cyclicalpatternsin Americanelectionsor was
somethingnew at work?
The events of 1994 are consistent with some existing theories of midterm
It is not our
losses, inconsistentwith others,and open to endless interpretation.1
intentionto add to this. Rather,we wish to examinethe 1994 electionin its political context, comparingit to othermidtermelections of similarand differentcircumstances.Specifically,we want to examine the choices in 1994 as they were
perceivedby the voters.What informationcould voters have about the performance of government,the positions of parties, and the effects of both on their
lives? How does this level of informationand uncertaintycompare with the
choices faced by voters in other midterm elections?What factorsbest explain
these differences?
We hypothesizethat voters in 1994 differedfrom recent previousmidterm
election voters in that they made their evaluationsand arrivedat their decisions
on how to vote in the context of unified government.We believe that divided
governmentobscures causal connections and increasesuncertaintyso that the
overallpropensityto vote againstthe candidateof the President'spartyis lower.
We will show how voters'knowledge of partisangovernmentcontrols changes
acrossperiodsof unified and divided governmentand, further,demonstratethat
these changesaffectthe propensityof votersto punish the President'sparty In so
doing, we examine the preferencesand decisions of voters in all midtermelections from 1978 to 1994, controllingfor other well-identifiedfactorsthat shape
outcomesin House elections.Finally,we offersome thoughtson the implications
of these findingsto issues of electoralstabilityand governance.
THE IMPACTOF DIVIDED GOVERNMENTON VOTERS
Studies of divided governmentfall into two broad categories:causes and
consequences. Scholarsinterestedin causes ask how and why voters produce
divided control of the Presidencyand Congress.Some look to how elites, both
intentionallyand unintentionally,structurevote choices that produce divided
government (Alvarez and Schousen 1993; Jacobson 1990a; Petrocik 1991;
Seguraand Nicholson 1995; Wattenburg1991). Othersarguethat voters intentionallychoose dividedgovernment(Alesinaand Rosenthal1995; Fiorina1992).
In contrast,researchon the consequencesof divided governmentexplores
its impact on public policy (Cox and McCubbins1991; Fiorina 1992; Mayhew
1991; McCubbins1991) and the strategicbehaviorof politiciansin the legislative arena (Ginsbergand Shefter 1990; Kernell 1991). Curiously,this research
has seldom asked how unified or divided governmentaffectspolitical attitudes
among the citizenry.The only two exceptionswe know of are Bennettand Bennett (1993), who examine the factualinformationheld by voters over time, and
1 For researchaddressingthis event, see Abramowitz(1995) and Klinkner(1996).
610
Midterm
ElectionsandDividedGovernment
Leyden and Borrelli(1995), who examine state-level economic evaluationsas
endogenous to state-leveldivided government.Examinationsof actualbehavior
are harderto find.
We seek to fill this lacuna by exploringthe question of whether voters are
more, or less, likely to punish the President'spartyat the midtermelection based
on partisancontrol of the Presidencyand Congress.For example, in periods of
unified government(e.g., 1994) do voters attributegreaterresponsibilityto the
President and his party?Are voters less likely to hold the President'sparty
accountableunder divided government?
Thesequestionshave importantimplicationsfor the responsiblepartymodel
of government,which assumesthat voters can hold the partyin power responsible for its actions.Some have found supportfor this type of behaviorwith regard
to economicvoting (Fiorina1981), but othershave found the electorateunableto
meet the demands necessaryto engage in this behavior(e.g., Converse 1964).
Here, we explore whether during unified government, voters more closely
approximatethe behaviorneeded to sustaina responsiblepartymodel.
MIDTERMELECTIONSAND PRESIDENTIAL
POLITICS
Effortsto explainthe phenomenonof midtermlosses by the President'sparty
have contributedsubstantiallyto our generalunderstandingof Americanelections (Alesinaand Rosenthal1995; Campbell1966; Cover 1985; Erikson1988;
Jacobson1990a;Jacobsonand Kernell1983; Kernell1977). Explanationsof this
phenomenonfall into threebroadcategories.
The turnout explanation,generallyreferredto as "surgeand decline" (A.
Campbell1966;J. Campbell1993), presupposesno inconsistentvoting behavior
on the partof citizens,nor any changeof preference.Rather,part-timevoterswho
turned out for the presidentialelection, who Angus Campbellinfers disproportionatelyfavorthe ultimatevictor,fail to turn out in the less salientmidtermyear.
Theirvotes, centralto the election of some of the President'sco-partisansin the
presidentialelectionyear,are sorelymissed in the midtermelection.
In "evaluative"
explanations,the argumentis that some voters will vote in
midtermelectionsin a mannerinconsistentwith theirvote for Presidentjust two
yearsbefore,at the expense of the President'sparty,because of economic evaluations (Tufte1975), the personal(un)popularityof the President(Kernell1977),2
or a desire to balance the ideological composition of the elected branches
2
That Presidentshave systematicallyappeared at the nadir of their popularityas the midterm
approachesis an empiricalregularityupon which the theory was built. The effects describedare
asymmetric(Kerell 1977), meaning the benefits accruedto a Presidencyfrom a good economy
and favorablepolicy evaluationsare substantiallysmallerthan the costs associatedwith the converse. When the Presidenttakes office his popularityand the expectationsof the electorateare
invariablyhigh. Declining evaluationsrepresentan aggregatelevel regressionto the mean and
should be expected under most circumstances(but see Erikson1988).
611
PoliticalResearch
Quarterly
(Alesina and Rosenthal 1995; Fiorina 1992).3 The midterm election, then, is
driven by the voter'sevaluationof the President,economic conditions, or the
desire to achievemoderatepolicy outcomes.
A final explanationof midtermoutcomes is providedby Jacobsonand Kernell's(1983) "strategicpoliticians"model, which relies on the sophisticationand
awarenessof potentialcandidatesand donors,who makedecisionsaboutwhether
to run, or to whom to donate,based on their subjectiveestimationof the probability of winning. This estimate, in turn, is driven by national-levelpolitical
forces-macro-economic performance,issues, presidentialpopularity-thereby
associatingthese forceswith outcomes.The resultis thatthe President'spartydoes
poorly-not because voters are sophisticatedconsumersof national trends but
becausepotentialcandidatesand politicalelites behaveas though they were.
underDividedGovernment
MixedMessagesand theAttribution
of Responsibility
These three explanationsof midterm losses are not mutually incompatible
and each is likely working at some level. Evaluative(and likely strategic)explanations of midtermelection losses are, to a greateror lesser extent, built upon
two assumptions.First, voters must hold the Presidentaccountable,or at least
politicians,potential candidates,and political elites must believe that they do.
Thatis, blame must be attributed-by voters or by potentialcandidates-to the
Presidentfor policies and economic circumstancesdeemed unfavorable.
But, as Fiorina(1992) has observed, the attributionof blame is tricky in a
system of separatedpowers. The Presidentis only a single participantin a much
larger structureso the opportunity exists for the attributionof blame to be
mixed, with the Presidentpointing his fingerat the Congresswhile the majority
there does the reverse.Co-partisans,at the very least, distancethemselvesfrom
the President. This is the second critical assumption. For these models of
midterm losses to work, voters must draw connections between the President,
whom they must blame, and membersof Congressfromhis party.
Accountabilityacrosselectoralbrancheswould be enhancedby a responsible partymodel of government,but decliningpartisanshipamong the electorate,
coupled with the advantagesof incumbencyand the personalizationof the representative-constituency
relationship,make such a model unlikely.In short, partisan conflictbetween the branchesintroducesuncertaintyinto any calculations
over the attributionof blame.
3
In Alesinaand Rosenthal's(1995) model of institutionalbalancing,moderatevoters achievemoderate policy outcomes through balancingpartisancontrol of the Presidencyand Congress.The
ability to do so, however,is enhanced at the midtermelection when voters' decisions are made
under greatercertainty-they know which partycontrolsthe PresidencyWe have expressedskepticism about this theoryelsewhere(Seguraand Nicholson 1995). Nevertheless,the argumentsuggests that moderatepolicy desiresare the drivingforcebehind midtermvoting.
612
Midterm
ElectionsandDividedGovernment
Dividedgovernment
exacerbates
theuncertainty
problemsin blameattribution.
wasproblematic
in American
[I]fresponsibility
politicsevenwhengovernment
wasunified,the problemis compounded
whengovernment
is divided.PresidentsblameCongressfor obstructing
crafted
whilememsolutions,
carefully
bersof Congressattackthe Presidentforlackof leadership.
Citizensgenuinely
cannottell who is to blame,and the meaningof electionoutcomesbecomes
confused.(Fiorina1992:109-10)
increasingly
In a periodof dividedgovernment,partisanshipfailsto bridgethe inter-branch
gap
when assessingwhom one wishesto hold accountableforthe failingsof government.
Furthermore,when governmentis divided, press coverage critical of the
administrationis almost certain to be tempered with an "either/or"balance,
where the Congressof the other partyis offered-usually by an administration
spokesman-as an alternativesource of difficulties.With their focus and format
tailoredto politicalconflict,nightlynews and Sundaymorningnetworkpolitical
magazineswill, ratherthan offeringcritiquesof policies authoredwith the cooperation of both branches, weigh the merits and demerits of policy positions
offeredby each institutionand the partywhich controls it. Prominentpolitical
voices fromboth ends of the spectrumwill be audible,competingin the attribution of responsibilityfor all things good and bad. In short, the messages upon
which the citizen must rely in formingjudgmentswill be mixed.
of MixedMessages
CognitiveImplications
Divided governmentboth muddies the informationalwaters and, at some
level, reducesthe amount of usable information.As a result,it is harderto relate
the relativeapprovalor disapprovalof the Presidentand the political partiesto
public perceptionsof policy outcomes or economic performance.Under divided
government,plausible cases can be made for holding either institution and/or
either political party responsible for outcomes. This increase in noise with a
decreasein usableinformation,in turn, generatesconsiderableuncertaintyin the
minds of voters and potentialcandidatesalike who are approachingdecisions.
For the economic voter,blame or creditfor good or bad economic performance is hard to offer.While it is generallyassumed that the Presidentbears the
greaterburden, divided governmentmust amelioratethe effects. Similarly,presidential popularityis likely to decline at a much slower rate if the public perceives (or is successfullyconvinced)thatthe apparentfailingsof the Presidentare
the productof a hostile Congress.Under virtuallyany circumstances,and holding otherthingsequal, the aforementionedeffectson the Presidentand, by extension, the electoralprospectsof his co-partisansshould be smallerin a period of
divided government.4
4 Similarly,for challengersand incumbentsalike, who must decide whetherto run for (re)election,
this uncertaintyis equally vexing. For Jacobson and Kerell (1983), quality challengersof the
613
PoliticalResearch
Quarterly
We do not mean to imply thatnone of the public holds the Presidentresponsible in a divided governmentregime. The Presidentis the focal point of government for most citizens (Lowi 1985) and we would expect that to remainthe
case for many people even in a divided governmentenvironment.But the portion of the electorateattributingblame to the Presidentis smallerthan it might
be, as is the share of the total blame assigned to the President.This hedging in
attributionof responsibilitymay be sufficientto damp the electoralswings we
would normallyexpect in midtermelections for any given set of economic and
politicalcircumstances.
TESTINGTHE INFORMATION
EFFECTSOF DIVIDED GOVERNMENT
We have arguedthat the presenceof dividedgovernmentaffectsinformation
levels among voters, who, in turn, have less informationabout which partyis to
blame for whateverdissatisfactionsthey have with government.The effect, we
believe, benefits the President'sparty during midterm elections. Beforewe test
this hypothesis,we want to demonstratethe occurrenceof the informationeffects
that drive our argument.Specifically,we turn now to whether voters really do
possess less electorallyuseful informationin periods of divided government.
A simple way to test this question is to ask exactlyhow many respondents
actuallyknow whethergovernmentis divided (a useful piece of informationthat
we shall employ in anothercontext momentarily).Bennettand Bennett(1993)
asked a similarquestion.Theirdependentvariablewas a three-pointscale which
was the sum of two dichotomies:the correctidentificationof the pre-electionand
post-electionHouse majorities.Using 17 regressors,all but one significant,they
wereableto explain28 percentof the variance.Withrespectto the effectof divided
government,they found that"peoplefind dividedgovernmentconfusing,but unifiedpartycontrolof nationalelectiveinstitutionsboostspoliticalinformation"
(76).
Our hypothesis is simpler and our measure focused on the pre-election
period of interest in this effort.We expect that during periods of divided government,a smallerpercentageof the electoratecan identifywhich politicalparty
controlsthe House beforethe election. This is a good questionby which to judge
the level of informationsince the answeris both simple and, for the time period
of this study,alwaysthe same-the Democrats.5In short, this is a very low highbar over which we have asked the respondentsto jump.
President'spartyusuallypreferto sit out midtermelectionswhile good challengersfromthe opposition smell blood and qualify Incumbentsdeciding whether to retire face similar choices and
incentives.Butif these potentialcandidatesare less certainabout the tilt of nationalforcesand the
prospectsof success, both effectsare almost certainto be damped.
5 We assumethat no one who correctlyidentifiedthe partycontrollingCongresswould be unaware
of the President'spartisanidentification.Unfortunately,the NES, to our knowledge, has never
asked this question during the yearsof our study.
614
Midterm Elections and Divided Goverment
Using data from the National Election Studies for midterm elections from
1978 to 1994, pooled, we comparedthe responses on this question, grouping
1978 and 1994 as unified governmentperiodsand 1982, '86, and '90 as divided
is coded one (1) if the pargovernmentperiods.The variableDividedGovernment
tisan control of the two elective branchesis divided prior to the midtermelection, and zero (0) if the governmenthad been under unified partycontrol.
As is clearlyapparentin Table 1, divided governmentdramaticallyreduces
the level of usableinformation.When governmentwas unified, 72.02 percentof
all respondentscould correctlyidentifythe Democratsas the partycontrollingthe
House. When governmentwas divided,only 38.33 percentof respondentscould
performthe same task. That is, over 60 percentof respondentsgot this question
wrongin 1982, '86, and '90 despitethe factthat the answerhad been unchanged
for 30-40 years prior to the respondent'sinterview.The differenceis statistically
significant,with Chi-square=1009.44 and an associatedp-valueof .000.
TABLE1
THEHOUSEBYPARTISAN
CROSSTAB
OF PARTYCONTROLLING
CONTROL
OF KNOWLEDGE
IN MIDTERMELECTIONS:
OF GOVERNMENT
1978-1994
Knowledgeof Controlof the House
PartisanControl
of Government
Unified
Divided
Total
No
Total
Yes
N
%
N
%
N
%
1037
3416
4453
27.98
61.67
48.17
2669
2123
4792
72.02
38.33
51.83
3706
5539
9245
100.0
100.0
100.0
PearsonChi2(1) = 1009.4378 Pr = .000
To see how importantdivided governmentis to producing this difference,
we ran a Probitmodel where the abilityto identifycorrectlythe partycontrolling
we
the House was the dependent variable.In addition to DividedGovernment,
control for two predictorsof voter knowledge. First, Levelof Informationis the
NES interviewer'sassessmentof the respondent'spoliticalknowledge, a variable
found elsewhereto be both reliableand valid (Bartels1996). Second, Partisanis
coded one (1) for both weak and strong partisanidentifiersand coded zero (0)
for Independentsand independentLeaners(towardone of the majorparties).We
estimatedmodels for all NES respondents,as well as for only those respondents
who voted.
In Table2, the model examiningall respondentspredicts72.7 percentof the
cases correctly,and has a ProportionalReductionof Error(Lambda-p)of .433.
Not surprisingly,those respondentsthe interviewerassessedas politicallyknowledgeableand strongpartisanswere likely to get it right,and the effectsarehighly
615
PoliticalResearch
Quarterly
TABLE
2
PROBITCOEFFICIENTS
ESTIMATING
OF RESPONDENTS
EFFECTSON THELIKELIHOOD
KNOWINGPARTISAN
CONTROLOF THEHOUSEOF REPRESENTATIVES
BEFORE
MIDTERM
ELECTIONS:
1978-1994
Variable
All Respondents
Partisan
MidtermVotersOnly
.193***
(.033)
-.984* *
Divided Government
(.030)
.131**
(.045)
-1.104***
(.046)
Levelof Information
.555***
(.015)
.508***
(.024)
Constant
-.485***
(.036)
-.148*
(.063)
N
9164
Chi Square
2729.81
.000
Significance
PercentPredictedCorrectly
4334
72.7
PRE(Lambda-p)
.433
1085.91
.000
73.4
.246
*Significantat p < = .05 **Significantat p < = .01 ***Significantat p < = .001
significant.But more to our point, even when we control for these obvious and
powerful predictors,DividedGovernmentstill suppresses the level of information-more respondentsget it wrong. When we look at the changes in the predicted probabilitiesassociatedwith changes in the values of the independent
variables,DividedGovernment
actuallyoutperformsPartisan.The model predicting the knowledgelevel of midtermvoterspredicts73.4 percentof the cases correctlyand has a PREof .246. The resultswere consistentwith the earliermodel,
the only differencebeing a slight reductionin significancefor Partisan,almost
certainlya product of the self-selectioneffect of those who turned out. Divided
is once againan extremelypowerfulpredictor.
Government
Sinceprobitcoefficientsarenot directlyinterpretable,it is helpfulto examine
the changein the predictedprobabilitiesresultingfromthe model acrossthe range
of values of each independentvariablewhile holding all othersconstant. For all
decreasedby 37.4 percent the likelihood that
respondents,DividedGovernment
the respondentwould correctlyidentify the party controllingthe House, ceteris
paribus.Among actual midtermvoters, the decreasewas 40.1 percent. In both
was clearlystrongerthan strengthof partisanship.Parcases, DividedGovernment
more
tisanswere 7.6 percent
likely to know who controlledthe House among all
6
We thankJ. Scott Long for his innovativeprogramto generatepredictedprobabilitychanges.
616
Midterm
ElectionsandDividedGovernment
respondents,and only 5.1 percent among actual voters. Not surprisingly,the
largesteffect was from the interviewer'sassessmentof the respondent'sinformation level, with an increaseacrossthe rangeof valuesof 73 percentfor all respondents and 67.3 percentfor midtermelectionvoters.Nevertheless,the importance
of DividedGovernment
to the knowledgeof the respondentis demonstrable.
We wanted to check if this informationdeficit appearedin other contexts,
perhapsmoredirectlyrelatedto the congressionalelectionsin which we areinterested.We look, therefore,at questionsaboutwhetherrespondentscould recallthe
name and partyof both candidatesrunningin their House district.Even in unified periods,only 8.67 percentof all respondentscould recallthe name and party
of both candidates.Butwhen governmentwas divided,the numberrecallingboth
names decreasedto 5.65 percent.The Chi-square(38.17) was againsignificantat
p = .000. The conclusionhere is simple. Duringperiodsof divided government,
votersclearlyhave less politicallyand electorallyuseful information.
IN CONTEXT
ELECTIONS
THE 1994 CONGRESSIONAL
Having demonstrated,we think, that the amount of useful informationand specificallyinformationrelevantto blame attribution-varies significantly
from divided to unified electoralenvironments,we arrive,again, at the starting
point of this effort.The 1994 congressionalelections took place in a period of
unified government.Under such circumstances,it is our expectation that the
attributionof blame for unsatisfactorypolicy or economic performanceis a simpler task for most citizens and reinforcedby relativelyconsistentmessagesto be
found in both the media and the public discourse.7The results in 1994, while
they might have been worse than we might otherwise have expected for the
Democraticincumbents given the contemporaryperformanceof the economy,
were driven in part by the unified nature of nationalpolitics. All policymaking
brancheswere in the hands of the same politicalparty.8
In order,then, to get a handle on the information-drivenprocessesat work,
we hypothesize that at the individual level, ceterisparibus,the likelihood of a
respondentpreferringa congressionalcandidatenot of the President'sparty in
midtermelections is smallerwhen the nationalgovernmentis divided. Specifically, using the National Election Studies for midterm congressionalelections
Our explanationis also consistentwith Jacobson and Kemell'sstrategicmodel, since politicians
and donors may also have difficultyperceivingthe prevailingdirectionof nationaltrendsin a poor
informationenvironment.
8 Takento its naturalconclusion, we would hypothesizethat, at the aggregatelevel, the size of the
midtermloss by the President'spartywill be smallerthan it otherwisewould have been (controlling for economic and politicalconditions)should an election take place in an era of divided government.Such a hypothesisis difficultto test given the relativelysmall N and the inabilityto control for importantfactorssuch as campaignspending prior to 1972.
7
617
PoliticalResearchQuarterly
from 1978-94, we will examine the vote choice of respondentsin their district's
race for the House of Representatives.9
Controllingfor incumbency,spending,
respondent'sparty,presidentialapproval,economic evaluations,and other factors usually associatedwith midtermvoting, we will estimatethe probabilityof
voting for the out-party-the party not currently in control of the White
House'0-between periods of unified and divided government.In so doing, we
take a close look at the 1994 election to assess if it is an "outlier"or consistent
with the rhythmsand dynamicsof congressionalelections.
The effectsof unified or divided governmentmay not be the same acrossall
voters,especiallywith regardto politicalknowledgeand degreeof partisanidentification. First, as for political knowledge, we are curious as to whether this
effectvariesacrossrespondentswho know,or do not know,whethergovernment
is unified or divided. Those knowledgeableabout the partisancontrol of governmentshould be more susceptibleto the hypothesizedeffectsthan those without this knowledge. However,those with high informationlevels might also be
sufficientlypartisanor awareof the issues at hand thatblame attributionremains
a straightforward
task,even in an era of dividedgovernment.' Therefore,the relativestrengthof the relationshipswe hypothesizewill varyacrossstrengthof partisanship.Those stronglyidentifiedwith a partyareless likely to cross partylines
to vote, regardlessof presidentialpopularity,economic evaluations,or the quality of their party'scandidate,thus reducingthe availablevariancefor which the
informationenvironmentand the divided controlof governmentmight account.
Independentsand partisanleaners should be more subject to the forces traditionallyperceivedto be at work in midtermelections, and specificallythe factors
identifiedhere, though their lower level of informationand sophisticationmay
confound these effects.
9 Theinclusionof campaign
seemedessentialbutlimitsouranalysisto
financeas a controlvariable
Act.Theelectionof 1974wasexcluded
electionsthatpost-datetheFederalElectionsCommission
effectof Watergate.
documented
becauseof thepreviously
overwhelming
10Bycustomin thestudyof American
refersto the politicalpartynot occupypolitics,"out-party"
Thatis, in periodsof dividedgovof theirstatusin the Congress.
regardless
ing the Presidency,
is themajority
ernment,theout-party
partyin theCongress.
11Anotherconfounding
factoris thatthehypothesisdoesnot,strictlyspeaking,requiretheresponof thepolitunaware
is dividedorunified.Evenrespondents
dentsto knowwhethergovernment
icalcontrolof the electoralbranchesmightmanifestdifferences
if, as we haveargued,theinforA
circumstances.
andthenatureof politicalmessagesvariesacrossdifferent
mationenvironment
in 1982or 1990,forexample,maynot knowthatpoliticalcontrolis divided,butis
respondent
whilerespondents
to conflicting
stillsubjected
messagesof creditclaimingandblameattribution,
environin 1978and 1994couldbe sensitiveto an information
unawareof unifiedgovernment
We
willtest,
is
the
the
Democratic
that
a
consistent
mentcharacterized
problem.
party
message
by
is relatedto the size and
then,to see if factualknowledgeof unifiedor dividedgovernment
strengthof theeffectwe hypothesize.
618
MidtermElectionsandDividedGovernment
DATA AND VARIABLES
The dataused for our analysisare a mergedset of two distinctcollectionsof
information.First, as we have indicated, we pooled responses to the 1978-94
midterm National Election Studies. We then merged this individual-leveldata
with contextualvariablesappropriateto the congressionalrace in that respondent'shome district:the presenceor absenceof an incumbent,the qualityof the
challenger,the relativespending of the two candidatesinvolved (as reportedto
the FEC), and the partisanidentificationof the incumbent and challenger.In
addition, we included several dummy variablesto account for the presence or
absenceof dividedgovernment,by yearand in total. The resultingdataset, then,
containsboth individualand contextuallevel variablesthat collectivelytake into
considerationmultiple explanationsof voting at the midterm.
DependentVariable
The dependent variableis a simple dichotomy of whether the respondent
candidatein the district'sHouse election. This variable
voted for the "out-party"
was obtainedby recoding each respondent'sself-reportedvote, where Outparty
equals one (1) if the respondentclaimed to have voted for the candidatenot of
the President'spartyand zero (0) if the vote was cast for the President'sco-partisan. Those who report having not voted are-for obvious reasons-excluded
fromthe analysis.The resultingN from these five midtermelections is 3636.
Variables
Independent
We control for three individual-levelvariablesof importance:PartyID, IdeApproval.PartyID is a seven-pointscale that incorporates
ology,and Presidential
both partisanpreferenceand strengthof preference.It is recodedfor each yearin
a mannersuch that the highest value (6) correspondswith that of the out-party
candidatewhile the lowest value (0) implies strongidentificationwith the President'sparty.Naturally,we expect that strong identificationwith the out-partyis
a positive predictorof castinga vote for that out-partycandidate.
Ideologyis the respondent'sself-reportedideology,againrecodedin a manner
such that the highest value indicates an ideologicalbias for the out-partyand
againstthe President.Our reason for including this variablealongside partisanship is that committednon-partisanswho have an ideologicalself-identification
and a traditionallypartisanvoting preferencewould be missed by the previous
variable.These two variablesarenot collinear,correlatingat only .31, well below
the usual thresholdfor concern.We againexpect a positive relationship.
Presidential
Approvalis the respondent'sassessmentof the President,coded
as a simple dichotomywhere one (1) indicatesthe approvalof the Presidentand
zero (0) the disapproval.This variableis included to accountdirectlyfor the negative voting thesis previously discussed. Ceterisparibus,we expect Presidential
619
PoliticalResearch
Quarterly
Approvalto be negativelyrelatedto the probabilitythat a respondentvotes for an
out-partycongressionalcandidate.'2
We include two variablesto account for the specific context in which the
voting decision was taking place-that is, district-specificvariables:Spending,
and Quality.Spendingis the ratioof dollarsspent by the two candidates.To reach
this figurefor each race,we divide the spendingof the out-partycandidateby the
spendingof the candidateof the President'sparty.In-partycandidateswho spent
nothing have been recodedto one dollarto avoid undefinedterms.The measure
that resultsshould be positively associatedwith the likelihood that the respondent votes for the out-partycongressionalcandidate.The impact of money on
congressional elections has been documented on numerous occasions (e.g.,
Fenno 1982; Jacobson 1980, 1990b; Sorauf1988 among many others) and our
expectationsdo not vary.Significantspending is the only way challengerscan
accumulatethe necessaryname recognitionand is a well-recognizedpart of the
incumbent'sadvantage,given his/her generallyhigher access to contributions
fromboth individualsand groups.A substantialspending differentialin favorof
the President'sco-partisancould preventa loss of a seat, even in the presenceof
other factorstending in the opposing direction.We control for Spending,therefore, in order to prevent its effect from potentiallyoverwhelmingthe information-drivenbehaviorwe want to identify.
Quality,like Spending,is a relationalmeasure of the two candidates.The
quality of each candidateis coded as two (2) if the candidateis an incumbent,
one (1) if holding (or havingheld) other politicaloffice, and zero (0) if the candidateis a politicalneophyte.'3The resultingqualityscore of the candidatefrom
the President'spartyis then subtractedfromthat of the out-partycandidate.The
resultingvariable,Quality,ranges from 2 to -2 and should be positively correlated with the likelihood of a voter casting a ballot for the out-partycandidate.
Without doubt, incumbency,and the advantagesit accrues to a candidate,is a
huge advantagein an election (Alfordand Hibbing 1981; Cain, Ferejohn,and
Fiorina 1987; King 1991). Similarly,the literatureon the importanceof challenger quality is large (Abramowitz1988; Jacobson 1990a; Squire 1992) and
with few exceptions (e.g., Seguraand Nicholson 1995) seems to suggest that the
relativepolitical experienceof candidatesis stronglyassociatedwith predicting
congressionalelection votes and outcomes. Qualitychallengersrunning against
incumbentsof the President'spartyare farmore likely to producea midtermseat
loss than those with no politicalexperience.
Somerespondentsdid, in fact,refuseto eitherapproveor disapproveof the President,but they representedonly 1.15 percentof all respondentsand less than 1 percent(.82 percent)of thosewho voted.
For the purposeof simplicity,they were droppedfromthe analysis,but theirinclusiondoes not, in
any way,alterthe size or significanceof the otherestimatesnor the predictivepowerof the model.
13 We thank GaryC. Jacobsonfor providingus with these data.
12
620
Midterm
ElectionsandDividedGovernment
In additionto these district-specificcontextualvariables,we account for the
differentpolitical circumstancethrough the inclusion of two variables:Divided
Government,and EconomicEvaluation.Again, DividedGovernmentis coded one
(1) if the pre-electionpartisancontrol of the elective branchesis divided, and
zero (0) if it is unified. Ceterisparibus,DividedGovernment
should damp the attribution of blame to the President'spartyand thereforebe negativelyrelatedto the
probabilitythat a respondentvotes for the out-partyHouse candidate.
Economic
Evaluationis used here as a controlvariablesince it has been widely
suggested(acobson 1990b; Kramer1971; Tufte1975) that the voter'sperception
of the macro-economy'sperformanceleading up to the midterm election has
much to do with the evaluation of the President as well as the vote in the
midtermHouse elections.We measureEconomicEvaluationusing a retrospective
evaluationquestion from the NES. Specifically,respondentsindicated whether
they thoughtthe economyhad improved,worsenedor stayedthe same in the last
year.14 The variableis recoded so that poor economic assessmentsare coded as
one (1), a perception of an improvingeconomy as negative one (-1), and no
changecoded zero (0). The resultingmeasureshould be positivelyrelatedto the
likelihood of voting for the out-party
TESTINGAND RESULTS
Table3, column 1 presentsestimatesfroma multivariateprobitanalysis.The
chi-squaredstatisticis highlysignificantwhich meansthatwe can rejectthatnullhypothesisthat all the independentvariablesin our model areunrelatedto voting
for the out-partyin these midtermelections. Furthermore,the model correctly
predicts81.8 percentof the caseswith a ProportionalReductionof Error(Lambdap) of .609 over the null model. Takentogether,these tests indicatethat our model
is well-suitedto explainingvoting for out-partycandidatesat the midterm.
Turningto individualpredictors,we see that our hunch was correct:voters
are less likely to vote for the out-partyduring periods of divided government.
Controllingfor other factors, divided governmenthas a statisticallysignificant
negative effect (p = .001) on voting for the out-partyHouse candidate at the
midterm. To make any judgments about this effect in the 1994 congressional
elections, in column two we reestimatedthe model by disaggregatingthe two
periods of unified governmentin our sample by election. Both the 1978 and
1994 dummyvariablesare statisticallysignificant(p < .05 for 1978, p < .001 for
1994) and the signs of the coefficientsare positive, meaning that unified gov14
For 1978, the specific questionwordingreferredto the "businessconditions."The exact question
wordingused in the other four surveyswas not employedin the 1978 survey,so the discrepancy
was unavoidable.Resultson this variable,then, should be read with caution but, we think, are
suggestive.Nevertheless,we think this approachis superiorto an aggregatelevel measureof inflation and unemploymentwhich can only vary acrossthe five elections.
621
Political Research Quarterly
TABLE
3
PROBIT
COEFFICIENTS
ESTIMATING
INFLUENCE
OFPARTISAN
CONTROL
OF
GOVERNMENT
ONVOTING
FORTHEOUT-PARTY
IN MIDTERM
ELECTIONS:
1978-1994
Variable
Model 1
Model 2
PartyIdentification
.366 **
(.015)
.366 **
(.016)
Ideology
.195***
(.042)
.195***
(.042)
EconomicEvaluation
.048
(.035)
.055
(.035)
9.16e-07*
(4.68e-07)
8.84e-07*
(4.68e-07)
Quality
.371***
(.017)
.373***
(.017)
PresidentialApproval
-.274* **
(.063)
-.264 **
(.064)
Divided Government
-.199***
(.063)
Spending
1978
.128*
(.074)
1994
.278***
(.077)
-.996* **
(.087)
Constant
3417
N
3417
Chi Square
1998.67
Significance
PercentPredictedCorrectly
.000
81.8
PRE(Lambda-p)
-1.203***
(.090)
.609
2001.97
.000
82.0
.614
*Significantat p < = .05 **Significantat p < = .01 ***Significantat p < = .001
ernments in these two elections increased the probability that a voter chose the
out-party candidate. To be sure, the effect in 1994 is larger, but it does not alone
drive our findings on the importance of divided or unified government.
To estimate the approximate effect of the independent variables, we again
turn to the changes in the predicted probabilities associated with each. In
column 1, Divided Government decreases the likelihood of a voter choosing an
out-party congressional candidate by 7.8 percent, a substantial change given the
presence of important controls like partisanship, incumbency (as part of Quality),
and Spending. In the second column, the year dummies also have a substantial
effect. Since each of the years represent elections under unified government, we
622
MidtermElectionsandDividedGovernment
would expect the change to be positive. The 1978 dummy increasesthe likelihood of voting for the out-partyby 5.1 percent,while the 1994 dummyincreases
the probabilityby 10.8 percent.As we shall see momentarily,when we subdivide
the electorateby their strengthof partisanshipand level of informationso that
we can look at potentialswing voters, these effectsare even stronger.
Most of the control variablesworked as we had expected. The impact of
party identification,ideology, spending, and the quality of candidates (taking
into account both challengersand incumbents) all had statisticallysignificant
coefficientswith signs in the predicted directions.Like most researchon congressionalelections, partisanidentificationand candidateattributesappearto be
the most importantdeterminantsof voting.5
An unexpectedresultis the marginalinsignificanceof the respondent'sevaluation of the economy. Recallwe expected that voters offeringa more critical
assessmentof recenteconomic performancewould be more likely to choose candidatesfromthe out-partyThe coefficientsarein the predicteddirectionbut they
fail to reachsignificance,with a p-value of .083 for the model with DividedGovernmentand .061 for the model with the two year dummies.
WhoDecideson theBasisof DividedGovernment?
Whose choices are affectedby the presence or absence of divided government?Earlier,we hypothesizedthat not all voters should behave in this manner.
Indeed,we expect that many voters do not care, or even know, if governmentis
divided. For those voters who do not care, we expect them to have partisan
attachments.In this group, we would expect them to resist informationthat is
inconsistentwith their partisandispositions.Democratsblame Republicansand
vice versa.Basedon these considerations,we breakthe sampleinto two groups:
(1) partisans,which includes both strong and weak party identifiers;and (2)
Independents,which includes both Independentswho lean toward either the
Democraticor Republicanparties and "pure"Independents. Given that both
groups include respondentswith varyinglevels of commitmentto either party,
we still must include the partisanidentificationvariableto take into account the
very differentcandidatepreferencesof these groups.
We furthersubdividedthe sampleby whetheror not the respondentwas able
to identifycorrectlywhich party controlledthe House of Representatives
(again
assumingthey know the partyof the President).Thoseunableto performthis task
since theylackthe specific
correctlyarefarmoreproneto blameattribution"errors"
15
The most significantdangerin our generalapproachis the possibilityof serialautocorrelation.We
tested for such a possibilityby including the full arrayof time dummies. Since divided government only variesacrosstime, it by necessity falls out of the equation.The resultingmodel, however, does not appreciablyaffectthe remainingcoefficientsnor improvethe predictivecapacityof
any of the models we present,suggestingthat autocorrelationis unlikely.
623
PoliticalResearch
Quarterly
knowledgethatthe policymakingbrancheswerecontrolledby the sameor different
parties.We expectthatthe effectof dividedor unifiedgovernmenton the probability that a respondentvotes for an out-partyHouse candidatewould be weakeror
missingamongrespondentswho do not know whethergovernmentis divided.'6
We tested our model in each of these four groups,expectingthat only pure
Independentsand independentLeanerswho correctlyidentifiedthe partyin controlof the Housewould be influencedby dividedor unifiedgovernments.Able to
make the necessarycausalattributionsof responsibilityand unhinderedby partisan disposition,these votersshould be influencedby clearor unclearattributions
of responsibilitythat are based on control of the two branchesof government.
Independentsuninformedaboutcontrolof Congress,for obviousreasons,cannot
bring this informationto bear on theirvote choice. One could arguethat even if
these voters "gotit wrong,"they may neverthelessvote like the group that "gotit
right."We think this scenariounlikely given that these respondentsare not generallywell informedaboutpolitics.'7Finally,since causalattributionsof responsibility areless importantto these partisans,we also expect that this groupwill not
show much effectfromdivided or unified governmenton theirvote choice.
The analysis largely supports our hypotheses. Looking at the first two
columns of Table4, partisanswere less affectedby the unified governmentsin
1978 or 1994. There was no support for a unified governmenteffect among
strong partisansin 1978, whether or not they had correctinformationon congressionalcontrol.For 1994, thereappearsto be a significanteffectamong those
who could correctlyidentifythe partycontrollingthe House, but no effectamong
those less informed. Instead, it appears that the vote choice for partisansis
affected primarilyby the strength of their partisanidentification,presidential
approval,and candidatequality
Columns 3 and 4 depict the results for Independentsand Leaners.Those
knowledgeableabout the partisancontrol of Congress(column 3) voted on the
basis of unified governmentin the 1978 and 1994 midtermelections. What is
most striking about this finding is its strong effect. When compared to other
groups of voters, Independentsand Leanerswho correctlyidentified the party
controllingCongressappearto be highly influencedby unified government.
16
17
We suggestedearlierthat this assumptionmight not be correctif the informationaspects of this
dynamic,i.e., the media'spresentationof either/orexplanationsfor policy failureor economic distress, might have a sufficienteffecton even uninformedvoters to maintainthe differencebetween
voting behaviorsunder unified and divided governments.If this is correct,we would expect that
the DividedGovernment
coefficientwould remainsignificantacrossthe two groups(albeit,perhaps,
of a smallermagnitudeamong those unawareof partisancontrol).
The correlationbetweenthe interviewers'assessmentsof the respondents'politicalknowledgeand
Knowledgeof Congressis (rs = .32). While this correlationis moderate,it likely severelyunderestimatesthe associationbetween generalsophisticationand the dichotomousmeasuregiven that
some respondentsprobablyguessed correctlywhich partycontrolledthe House.
624
MidtermElectionsand Divided Government
TABLE
4
THEINFLUENCE
PROBITCOEFFICIENTS
ESTIMATING
ON
OF UNIFIEDGOVERNMENTS
IN MIDTERMELECTIONS,
AND
VOTINGFORTHEOUT-PARTY
BYPARTISANSHIP
OF HOUSE CONTROL:1978-1994
KNOWLEDGE
VotersKnowledge
of House Control
Strongor Weak
PartyIdentifiers
Correct
Incorrect
and
Partisan"Leaners"
Pure Independents
Correct
Incorrect
PartyIdentification
.357***
(.022)
.339***
(.027)
.450***
(.079)
.622***
(.107)
Ideology
.270***
(.066)
.031
(.091)
.189*
(.086)
.185
(.132)
EconomicEvaluation
.070
(.055)
.041
(.076)
.030
(.074)
.088
(.108)
9.25e-07
(8.22e-07)
8.56e-07
(7.22e-07)
5.10e-07
(1.06e-06)
2.60e-06
(2.21e-06)
Quality
.359***
(.027)
.347***
(.033)
.405***
(.037)
.471***
(.052)
PresidentialApproval
-.266**
(.105)
-.285*
(.128)
-.260*
(.139)
-.076
(.186)
1978
.030
(.119)
.106
(.172)
.383**
(.158)
.236
(.210)
1994
.264*
(.116)
.233
(.422)
.414**
(.163)
.323
(.505)
-1.624***
(.292)
-2.049***
(.396)
Spending
Constant
-1.238***
(.142)
-.908***
(.169)
330
N
1568
851
601
Chi Square
1087.98
516.07
229.78
Significance
% PredictedCorrectly
PRE(Lambda-p)
.000
84.6
.000
82.6
.676
.583
.000
76.7
.524
149.41
.000
77.6
.507
*Significantat p < = .05 **Significantat p < = .01 ***Significantat p < = .001
When we look at changes in the predictedprobabilities,this impressionis
supported.Among Independentswho knew which partycontrolledthe House,
our potentialswing voters, unified governmentsignificantlyincreasedthe likelihood that they would vote for the out-partyby 15 percentand 16.2 percentfor
1978 and 1994, respectively.Moreremarkableis how potent this effect is given
the presenceof the controlvariableslike spending and ideology.In sum, unified
and divided governmentsare criticalfactorsfor these so-called "swingvoters"in
midtermelections.
625
PoliticalResearch
Quarterly
Although partisansin 1994 were influenced by unified government, the
effectis little more than half the size of the effect on Independentsand Leaners.
Thus, despite the generaleffectof unified governmentidentifiedhere, it appears
that the elections of 1994 did have a greatereffecton partisansthan in the elections of 1978. This insight regardingpartisaninfluencescertainlymerits further
investigationin studies of the 1994 elections.
CONCLUSION
We began by hypothesizing that divided governmentshaped the choices
made in elections. Specifically,the muddied informationalenvironmentwould
enhance the prospects of candidatesof the President'sparty,who normallyare
held accountablefor all that ails the society but, when governmentis divided,
can blame the other side.
Our findings are supportive of our claims. In the first analyses, we show
that respondents and voters alike are less likely to be able to identify which
party controls the House of Representativesunder divided government, even
controlling for other important predictors including overall assessments of
political informationlevels and strong partisanship.In the second set of analyses we examinedvote choice for all respondentswho voted in the midtermelections from 1978 to 1994, and saw thatunified governmentwas a significantpredictorof voting for out-partycandidateswhile divided governmentdampedthis
effect, holding constant other importantfactors.Unpacking differentsegments
of the electorate,we saw that divided government'simpact was greatestamong
pure Independentsand independentLeanerswho knew about the partisancontrol of Congress.Among these voters, causalattributionsof responsibility,blaming who is firmly in charge, can be achieved when institutionalcircumstances
permit.
It could be suggestedthat our information-driventheoryis similarto extant
evaluativeexplanations,particularlythe policy-balancingmodel (Fiorina 1992;
Alesinaand Rosenthal1995), insofaras voting for House membersis linked to
the President.But our argumentis distinct in that the causal mechanismis the
institutionalcontext of the evaluation,i.e., unified or divided government.More
importantly,our findings would not support policy-balancingas an alternative
explanation.The causal mechanism in the balancing model is the ideological
proximityof House candidatesto the President.Votersemploying this decision
rule would vote alwaysfor the out-partyHouse candidatewho is more ideologically distant from the President.The presence of divided government, then,
should make no differenceat the individuallevel of analysis.Since the balancing
voter "wants"divided government after the election, the institutionalcontext
before the election is irrelevantto their preferences.Of course, the number of
in the electoratecould, itself, be endogenous to divided and
"policy-balancers"
unified government.But even in this event, the inclusion of divided government
626
Midterm
ElectionsandDividedGovernment
as a predictorof midtermvoting behaviorrepresentsan explanatorycontribution
distinct fromwhat the policy-balancingargumentoffers.
It is worth noting that we have not attemptedan examinationof aggregate
outcomes. The effect of divided government estimated here is on individual
behavior.How this behavioraggregatesinto the partisandivision of the House of
Representativesis a complicatedquestion.A full model of seat loss would need
to account for individualpreferences,which we examine here, as well as individual decisions over whether to vote, the recruitmentand funding of quality
candidates, retirements,and the distributionof partisansacross electoral districts,none of which we areable to addressfullywith these data.For this reason,
we cannot definitivelyanswerwhy 1994 was such a debacle for the Democrats.
Unified government,however, certainlymade the Democraticparty an easier
target-blame attributionwas simpler.
What are the broaderimplicationsof our study?First, the responsibleparties model of governmenthas long held sway among political scientists. This
model depends on the abilityof citizens to link policies and performanceto particular political actors. Divided partisancontrol of governmenthas significant
implicationsfor the workabilityof such an ideal type since it deprivesvoters the
opportunity to judge only one party for policy outputs-the sine qua non of
responsible party government-and it appears to have a significanteffect on
voters'decisions.
Second,our findingsseem to suggestsomethingthat politicalobservershave
speculatedaboutfor some time-Presidents mightbenefitpoliticallyfromdivided
government.Thoughpartisandivisionof the electedbranchesdoes seem to shape
the electoratein a mannercontraryto the President'sinterests,it seems also to
reduce the likelihood that any given voter holds the Presidentat fault and punishes his partisancolleagues.This logic could conceivablybe extended to presidentialapprovaland disapproval,a subjectbeyond the specificscope of this article but certainlyworthyof study.In addition,dividedgovernment'seffectson the
composition of the electorate,a subjectjust beginning to be examined at the
aggregatelevel (Franklinand Hirczy de Mifo 1998), should be more closely
examinedat the individuallevel. Findingthatthe President'spoliticalfortunesare
not only divorcedfromthose of his partisancolleaguesin the legislaturebut, perhaps, inverselyrelatedto them would suggestthatprospectsfor responsibleparty
governmentwill remainpoor, and the futureof divided governmentlengthy.
REFERENCES
Abramowitz,Alan. 1988. "ExplainingSenateElectionOutcomes."AmericanPolitical ScienceReview82 (2): 385-403.
. 1995. "TheEnd of the DemocraticEra?1994 and the Future of ConPoliticalResearch
48 (4): 873-89.
Quarterly
gressionalElectionResearch."
627
PoliticalResearch
Quarterlv
Alesina,Alberto,and HowardRosenthal.1995. PartisanPolitics,DividedGovernment,and theEconomy.New York:CambridgeUniversityPress.
Alford,John R., and John R. Hibbing. 1981. "IncreasedIncumbencyAdvantage
in the House."Journalof Politics43 (4): 1042-61.
Alvarez,R. Michael, and MatthewM. Schousen. 1993. "PolicyModerationor
AmericanPoliticsQuarterly21 (4): 410-38.
ConflictingExpectations?"
Bartels,LarryM. 1996. "UninformedVotes:InformationEffectsin Presidential
Elections."American
Journalof PoliticalScience40 (1): 194-230.
Bennett,StephenE., and LindaL. M. Bennett.1993. "Outof Sight, Out of Mind:
Americans'Knowledge of Party Control of the House of Representatives,
1960-1984." PoliticalResearchQuarterly46 (1): 67-80.
Cain, Bruce,John Ferejohn,and MorrisFiorina. 1987. ThePersonalVote:ConCambridge,MA: HarvardUnistituencyServiceand ElectoralIndependence.
versityPress.
Campbell,Angus. 1966. "Surgeand Decline:A Study of ElectoralChange."In
Campbell,Converse,Millerand Stokes,eds., Electionsand thePoliticalOrder.
New York:Wiley
Elections.LexingPulseof Congressional
Campbell,JamesE. 1993. ThePresidential
ton: Universityof KentuckyPress.
Cover,Albert.D. 1985. "Surgeand Decline in CongressionalElections."Western
PoliticalQuarterly38 (4): 606-19.
Converse,Phillip E. 1964. "TheNature of Belief Systems in Mass Publics."In
David E. Apter,ed., Ideologyand Discontent.New York:Free Press.
Cox, GaryW, and MatthewD. McCubbins.1991. "DividedControl of Fiscal
Policy"In GaryW Cox and SamuelKernell,eds., ThePoliticsof DividedGovernment.Boulder,CO: Westview.
Erikson,RobertS. 1988. "ThePuzzle of MidtermLoss."Journalof Politics50 (4):
1011-29.
Fenno, RichardE 1982. TheUnitedStatesSenate:A BicameralPerspective.Washington, DC: AmericanEnterpriseInstitute.
Fiorina,MorrisP. 1981. Retrospective
Votingin AmericanNationalElections.New
Haven:YaleUniversityPress.
New York:Macmillan.
. 1992. DividedGovernment.
Franklin,MarkN., and WolfgangP Hirczyde Mino. 1998. "SeparatedPowers,
Divided Government,and Turnoutin U.S. PresidentialElections."American
Journalof PoliticalScience42 (1): 316-36.
Ginsberg, Benjamin, and Martin Shefter. 1990. Politicsby OtherMeans:The
of Electionsin America.New York:BasicBooks.
DecliningImportance
Elections.New Haven:YaleUni1980.
Jacobson,GaryC.
Moneyin Congressional
Press.
versity
1990a.TheElectoral
Boulder,CO:Westview.
OriginsofDividedGovernment.
in
of
"The
Effects
990b.
1
CampaignSpending House Elections:New
Science34 (2): 334-62.
EvidenceforOldArguments."
AmericanJournal
ofPolitical
628
Midterm
ElectionsandDividedGovernment
Jacobson,GaryC., and SamuelKernell.1983. StrategyandChoicein Congressional
Elections,2nd.ed. New Haven:YaleUniversityPress.
Kernell,Samuel. 1977. "PresidentialPopularityand NegativeVoting:An Alternative Explanationof the MidtermCongressionalDecline of the President's
Party."AmericanPoliticalScienceReview71 (1): 44-66.
. 1991. "Facingan Opposition Congress:The President'sStrategicCircumstance."In GaryW Cox and SamuelKernell,eds., ThePoliticsof Divided
Government.
Boulder,CO:Westview.
King,Gary.1991. "ConstituencyServiceand IncumbencyAdvantage."British
Journalof PoliticalScience21 (1): 119-28.
Klinkner,PhilipA., ed. 1996. Midterm:TheElectionsof 1994 in Context.Boulder,
CO:Westview.
Kramer,Gerald H. 1971. "Short-termFluctuations in U.S. Voting Behavior,
1896-1964."AmericanPoliticalScienceReview65 (1): 131-43.
Leyden,KevinM., and StephenA. Borrelli.1995. "TheEffectof StateEconomic
Conditions on GubernatorialElections:Does Unified GovernmentMake a
Difference."PoliticalResearchQuarterly48 (2): 275-90.
Lowi, TheodoreJ. 1985. The PersonalPresident:PowerInvested,PromiseUnfulfilled. Ithaca,NY:CornellUniversityPress.
andInvesMayhew,DavidR. 1991. DividedWeGovern:PartyControl,Lawmaking,
tigations,1946-1990.New Haven:YaleUniversityPress.
McCubbins,MatthewD. 1991. "Governmenton Lay-Away:Federal Spending
and DeficitsUnderDividedPartyControl."In GaryW Cox and SamuelKerBoulder,CO:Westview.
nell, eds., ThePoliticsof DividedGovernment.
Petrocik,John R. 1991. "DividedGovernment:Is It All in the Campaigns?"In
GaryW Cox and Samuel Kernell,eds., ThePoliticsof DividedGovernment.
Boulder,CO:Westview.
Segura,GaryM., and StephenP Nicholson. 1995. "SequentialChoices and Partisan Transitionsin U.S. Senate Delegations:1972-1988."Journalof Politics
57 (1): 86-100.
Sorauf,FrankJ.1988. Moneyin AmericanElections.Glenview,IL:Scott,Foresman.
Squire,Peverill. 1992. "ChallengerQuality and VotingBehaviorin U.S. Senate
Elections."Legislative
StudiesQuarterly17 (2): 247-61.
R.
1975.
of the Outcomes of MidtermCongresEdward
"Determinents
Tufte,
sional Elections."AmericanPoliticalScienceReview69 (3): 812-26.
Wattenburg,MartinP 1991. "TheRepublicanPresidentialAdvantagein the Age
of PartyDisunity."In GaryW Cox and SamuelKernell,eds., ThePoliticsof
DividedGovernment.
Boulder,CO:Westview.
Received:March5, 1998
Accepted:February,1999
snicholson@scu.edu
gary.segura@cgu.edu
629