Unintended Consequences: Anticipation of General Election Outcomes and Primary Election
Divisiveness
Author(s): Jeffrey Lazarus
Source: Legislative Studies Quarterly, Vol. 30, No. 3 (Aug., 2005), pp. 435-461
Published by: Comparative Legislative Research Center
Stable URL: http://www.jstor.org/stable/3598643 .
Accessed: 31/08/2011 17:13
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
Comparative Legislative Research Center is collaborating with JSTOR to digitize, preserve and extend access
to Legislative Studies Quarterly.
http://www.jstor.org
JEFFREYLAZARUS
Georgia State University
Unintended
Anticipation
And
Primary
Consequences:
of
General
Election
Election
Outcomes
Divisiveness
This article offers the first theory to explain the relationshipbetween primary
election divisiveness and general election outcomes that is groundedin candidates'
own behavior.Conventionalwisdom holds thatdivisive primariescause candidatesto
do poorly in generalelections. I show thatprimarydivisiveness does not cause this or
any otherpatternof generalelectionresults.Rather,expectationsaboutgeneralelection
resultscause primariesto be divisive. Non-incumbentsenterraces they thinkthey can
win, and they thinkthey can win where the incumbentis vulnerable.More candidates
enter those races than others, splitting the vote among them. This stampede creates
divisive primariesin whichincumbentsaremost likelyto do poorly,andchallengerswell,
in the generalelections.As a result,divisivenessis associatedwith (but does not cause)
bettergeneralelectionperformancesamongchallengersandworseperformancesamong
incumbents.In this manner,primarydivisiveness is an unintended consequence of
behavior directed towards the goal of winning the general election. I tested these
propositionsusing data from major-partyHouse primariesbetween 1976 and 1998
and found that (a) candidate expectations of victory determine when and where
divisive primaryelections occur, (b) those expectationsdrive the correlationbetween
primarydivisivenessandgeneralelectionresults,and(c) primarydivisivenesscorrelates
with incumbentsdoing poorly, and challengerswell, in general elections.
Common wisdom holds that divisive primaryelections hurtthe
candidateswho emerge from these elections by preventingthem from
running at full strengthin the general election. There are numerous
possible reasonsfor this commonwisdom. Forexample,voters loyal to
a primaryloser might find it difficult to break psychological attachmentswith theirfavorite,butnow-defunct,candidate(KennyandRice
1987; Sullivan 1977-78) or to formattachmentswith the partynominee
(Southwell 1986). Those who worked for a primaryloser's campaign
might not work (or work as hard) for the party nominee during the
generalelection (Comer1976; JohnsonandGibson 1974; Stone 1986).
Finally,the losing candidatein a divisive primarymight inadvertently
give a generalelectionopponentammunitionagainstthe primarywinner
(cf. Alvarez, Canon, and Sellers 1995).
LEGISLATIVESTUDIES QUARTERLY,XXX, 3,August 2005
435
436
Jeffrey Lazarus
The plausibility of these causal mechanisms notwithstanding,
studies investigatingwhether or not divisive primariesactually harm
the generalelectionfortunesof partynomineeshaveproduceddecidedly
inconclusive results. Some studies have found that divisive primaries
hurt candidatesin the general election (Abramowitz 1988; Bernstein
1977; Segura and Nicholson 1995); others have found a mixed
relationship(Born 1981; Hogan 2003; Kenney andRice 1984) or none
at all (Hacker 1965; Kenney 1988; Piereson and Smith 1975). Recent
scholarshiphas even begun to turnthe common wisdom on its head,
finding that divisive primariesactually help U.S. House challengers
(Alvarez, Canon, and Sellers 1995; Arbour and McKenzie 2002;
Herrnson2000).
The lack of consensus in such a long-establishedliteraturecries
out for an explanation.I offer one here, arguingthat the mixed results
arerootedin a failureto take into accountwhy divisive primariesoccur
in the first place, along with an inappropriategroupingof challengers
and incumbentsin the same analyses. I argue that divisive primaries
arise when more than one "serious"candidateof a given partythinks
his or her chances of winning the seat warrantentering the race.1
Because seriouscandidatesof the challenger'spartyaremore likely to
run when they forecast good general election outcomes, divisive
primariescorrelatewith better general election performancefor challengers,despitetheputativelynegativeeffectsof"divisiveness."Further,
bettergeneral election performancesfor challengersoften result from
vulnerabilityin the incumbent,so divisive primariesalso correlatewith
poorer generalelection performancefor incumbentswho survivetheir
primarycontests (as most do).
Thus, the relationshipbetween primarydivisiveness and general
electionperformancedifferssubstantiallyfromthe relationshipposited
by common wisdom, in two ways. First, divisiveness does not cause
poorergeneralelection performance.Instead,candidates'expectations
of theirgeneralelection prospectscause them to enter,thus producing
divisive primaries.Second,the directionof the correlationis not always
thatpredictedby the conventionalwisdom. Rather,divisive primaries
correlatewith worse generalelection performancefor incumbentsbut
betterperformancesfor challengers.In otherwords, divisive primaries
are an unintendedconsequence of behavior directedtowardthe goal
of winning a generalelection.
The rationalperspectiveI have just sketchedexplains the mixed
and contradictoryresults of prior studies. Some of these studies test
the divisiveprimariesthesisusingthe performanceof incumbents,others
use challengers,andstill othersuse both.Since the relationshipbetween
UnintendedConsequences
437
primaryandgeneralelection resultsdiffersamongthe differentgroups,
differentdatasetsare boundto yield differentresults.Indeed,previous
findings in this literature correlate perfectly with the nature of
researchers' datasets. All studies that find that divisiveness hurts
candidatesfocus on incumbentsin theirempiricalanalyses(Abramowitz
1988; Bernstein 1977; SeguraandNicholson 1995) without searching
for the correspondingrelationshipamong challengers.All studies that
findthatdivisivenesshelps candidatesfocus exclusively on challengers
(Arbourand McKenzie 2002; Herrson 2000). All studies thatfind no
relationship(Hacker1965; Kenney 1988; Pieresonand Smith 1975) or
a mixed one (Hogan 2003; Kenney and Rice 1984) use both incumbents andchallengers.Finally,all studiesthatexaminechallengersand
incumbentsindependentlyof one another(Alvarez,Canon,and Sellers
1995;Born 1981) find thatdivisiveness"hurts"incumbentsbut "helps"
challengers.
The two countson which I breakfromthe existing literaturehave
been consideredin only one previous study.Born (1981) controls for
the effects of candidates' strategicbehavior on primarydivisiveness
using two-stage least squares (2SLS), and he divides observations
between incumbentsand challengers.His findings are similarto mine,
but unfortunatelyhis contributionsseem to have gone unnoticed.Nine
studies of the divisive primaryeffect in subpresidentialelections have
been publishedsince Bor's; none have incorporatedhis methodology.2
Thisarticleis anattemptto bringthediscussionof divisiveprimaries
back to Bor's issues, and it offers three distinct advances. First, I
make a strongerclaim about causality than Born does. Born argues
that a researchermust control for candidates'expectations about the
general election when determiningthe effect of divisive primarieson
generalelectionoutcomes.I arguethatthese expectationsare the causal
aspect of the relationshipand that a propercontrol for them washes
out any independenteffect that primarydivisiveness has on general
election results. Second, I provide a theoretical explanation for the
differing results for challengersand incumbentsthat is missing from
Bor's analysis.Althoughhis studyintroducesmethodologicalcontrols
for reciprocalcausality,it neitherdiscusses the causal mechanismsfor
the "new" direction of causality nor offers falsifiable hypotheses by
which to test it. The theoryprovidedin this articledoes both, andit also
accountsforthe contradictoryresultsobtainedby the dozenor so studies
of divisive primariesover the last 40 years. Third, I use an updated
dataset and obtain statistically significant results. Born's data are
complete only through1976 and his findings do not reach the level of
statisticalsignificanceatp < .05 (1981, 656-57). In contrast,I use data
438
Jeffrey Lazarus
on House elections between 1976 and 1998 and find a significant
relationshipbetween primaryand general election results. In addition
to these advances in the divisive primaryliterature,this article also
extends the scope of the rationalmodel of candidateentry,using it to
explain why divisive primariesoccur and, hence, how they correlate
with general election performance.
The article proceeds as follows. In Section I, I consider one
measure of candidate perceptions of the "winability"of a seat: the
numberof candidatescontesting the election. Using a simple gametheoreticmodel, I argue that the numberof candidatesis a sufficient
proxy for candidate perceptions. In Section II, I present tests of
hypotheses derived from that game-theoreticmodel to demonstrate
the connectionbetween the candidates'perceptionsandthe numberof
candidates.In Section III, I discuss how I replicatedpast estimations
of the divisive primary effect, this time controlling for candidate
perceptions.My results indicate that candidateperceptionsdrive the
relationshipbetweenprimarydivisiveness andgeneralelectionresults.
In SectionIV,I concludeby discussingthe empiricalresultsandplacing
this study in the context of the largerdivisive primaryliterature.
I. Divisive Primaries and Rational Politicians
Thereis a correlationbetween primaryelection divisiveness and
generalelection fortunes,butthe correlationis drivenby a confounding
variable, incumbent vulnerability. On one hand, as an incumbent
becomes more vulnerable, more candidatesthink they can beat the
incumbentandmoreof thementerthe race.Thisfloodingcausesprimary
electionsto become moredivisive:the primaryelectionvote is splitinto
smaller and smaller slices, and both the margin of victory and the
winner'svote sharego down. On the otherhand,as incumbentsbecome
morevulnerable,they tendto do worse in the generalelection (because
they win primariesdespitetheirvulnerability,a fact I discuss in Section
III).As a resultof these twin relationships,primarydivisiveness correlates positively with general election results among challengers,but
negativelyamongincumbents.
Any proper estimation of the relationship between primary
divisiveness and generalelection resultsmust take into accountcandidates'perceptionsof incumbentvulnerability;estimationsthatfail to do
so measurenot a causal relationshipbut a spuriouscorrelation.Prior
tests of the divisive primarythesis, includingBor's, measuredivisiveness with some function of the primary'svote outcome, althoughthe
specific function varies from study to study. These measures reflect
UnintendedConsequences
439
how close the primaryelection was, but they do not take into account
candidates'perceptions.Although I also lack a variable that directly
measureshow candidatesthink abouttheir own chances of winning, I
takeadvantageof two variablesthatgive indirectindications:the amount
of money spentby losing candidatesin the primaryelection(introduced
in Section III), and the numberof candidateswho enter eitherparty's
primary.In this section and the next, I demonstratethat entry is an
appropriateempiricalproxy for candidateperceptionsby showing that
candidatesenter races guided by their perceptionsof whether or not
they can win. My demonstrationbegins by exploringwhy candidates
enter races in such a patternas to create divisive primaries.
Divisive primariesarea regularoccurrenceinAmericanelections,3
even though they seem to be in neither parties' nor candidates'best
interests.Partyleaderswanttheirpartyto obtainoffices, so they should
preferto spend their scarce resourceson contests that garer seats for
the party (general elections), ratherthan those that have only indirect
influence on the race's ultimate outcome (primary elections).
Furthermore,to the extentthatyoungpoliticianswith promisearethemselves a scarce resource for the party,leaders should prefer to avoid
pittingthem againstone anotherin a primaryandputtinga loss on one
of theirpolitical resumes.Meanwhile,candidatestypically run for the
purposeof winning office andpreferto rununderconditionsthat give
them the greatestchance of doing so.4 Thus, they preferto avoid running in difficultprimaryelections in favorof runningagainstweak-or
no-opposition.
It seems a paradox,then, thatdivisive primariesoccur at all. The
reason they do is that the importantactorsoften cannotwork together
to avoid them. Within some limits, politicians decide for themselves
which elections to contest (see, for example, Jacobson and Kerell
1981), leaving partyleadersout of the equationaltogether.Candidates'
entry decisions are made atomistically,besetting the decision-making
process with coordinationproblems.All candidateswant to run in the
most attractiveraces, but if they all runin the same race, they interfere
with each others'chances of winning. Further,otherraces that offer a
lower, but still positive, probabilityof victory remainundercontested.
As a result, to ask what causes a divisive primaryis to ask why coordinationfails between politicians from the same partywho run for the
same office.
To addressthis question,I turnto a simple game-theoreticmodel
of politicians' entry decisions. In the "PrimaryElection EntryGame"
(specified more completelyin the appendix),two players,Player 1 and
Player2, simultaneouslymakea single binarychoice:to Entera primary
440
Jeffrey Lazarus
election, or Don't Enter the race. I treatthe outcome of the primaryas
a lottery.The prize for the winner is entry into the GeneralElection,
anotherlottery.Thereare two nonstrategicplayers,Dummy andOther
Party.Both Player 1 and Player 2 are strong candidates,and they are
the only strong candidatesavailable to run for this particularoffice.
Both potential candidatesare strategic, and each will enter the race
only if the expected utility of doing so is positive. If both candidates
enter,thenthey will runagainsteach otherin a primaryelection. If only
one candidateruns,thenthe candidateis eitherunopposedortheprimary
opponentis Dummy,who is nonstrategic(thatis, will runregardlessof
the probabilityof winning), not likely to win the primaryelection, and
not expected to win the general election if he or she makes it there. If
neither candidateruns, then Dummy runs in, and wins, the primary
election. The winnerof the primaryelectionrunsin the generalelection
against Other Party.
Strategies,payoffs, and preferenceorderingsappearin Figure 1.
Thereare only two strategiesand threepayoffs. If both Playerschoose
Enter, then they each receive the payoff of runningin a Hard Race,
since each must run against the other. If either politician decides to
enterthe race when the otherdoes not, then the enteringpoliticianruns
an Easy Race againstthe nonstrategicopponent.Any playerwho opts
out of the race receives the payoff of pursuingthe Next-Best career
option,whateverthatmay be. I makeone assumptionregardingplayers'
preferences over the three payoffs: if a politician wishes to stay in
politics, andwe hold the office constant,then thatpoliticianwill prefer
to runan easy race againsta weak opponentthana more difficult race
againsta strongopponent.Therefore,both players always preferEasy
Race to Hard Race.
Since the player designationsare arbitrary,thereare six possible
preference-orderingpairs:(a, a), (a, b), (a, c), (b, b), (b, c), and (c, c).
Five of the games are dominancesolvable with unique solutions.All
three games in which players' preferences differ have a dominantstrategyequilibrium(DSE) of (Don't Enter,Enter)in which the player
who most prefersNext-Best chooses Don't Enter.In (a, a), in which
both players most prefer Next-Best, the DSE is (Don't Enter, Don't
Enter).In (c, c), in whichbothplayersleastprefertheirnext-bestoptions,
the DSE is (Enter,Enter).The only game thatis not dominancesolvable
is (b, b). Thereare, however,two Nash equilibriain this game: (Enter,
Don't Enter)and (Don't Enter,Enter).Mixed-strategyequilibriaexist
but are not considered.
Takentogether,theseequilibriasuggestseveralempiricallytestable
hypotheses regardingmultiple-entryelections. These hypotheses are
UnintendedConsequences
441
FIGURE 1A
Outcomes to Players in PrimaryElection EntryGame
Player 1
Enter
Not Enter
HardRace
Next-Best Option
Enter
HardRace
Easy Race
Player 2
Easy Race
Next-Best Option
Not Enter
Next-Best Option
Next-Best Option
FIGURE 1B
Possible Payoff OrderingCombinations
in the PrimaryElection EntryGame
(a) Next-Best Option > Easy Race > HardRace
(b) Easy Race > Next-Best Option > HardRace
(c) Easy Race > HardRace > Next-Best Option
the basis for the logit estimationsin the following section. I presentthe
specific predictionssuggestedby each hypothesisin italics; those predictionscorrespondto the independentvariablesin those logit equations.
Broadly,the solutionto the game indicatesthatas the HardRace
grows moreattractive,multipleentrygrows morelikely.Multiple-entry
primaries should be concentrated most heavily in situations that
approximatethe conditionsof games (c, c), in which the HardRace (a
difficultprimaryelection againsta strongopponent)is the equilibrium
outcomebecausebothplayersvalueentryhighly,relativeto exit. Overall,
enteringthe race is most attractivewhen the likelihood of winning is
highest andwhen the awardfor winning (a major-partynomination)is
most valuable.5If we assumethatcandidatesvalue victory in a primary
only because it representsa step towardthe ultimate goal of running
442
Jeffrey Lazarus
for office-officeholding-then we find the value of that awardto be
determinedby (a) the probabilitywith which the partynominee wins
the general election and (b) the value of the seat itself. These three
factors (likelihood of winning the primary,likelihood of winning the
general, and the benefits accruing to the seat's winner) allow me to
state specific hypotheses.6
Hypothesis1: Themorelikely a non-incumbentcandidateis to win
the generalelection, the more likely thereis to be multipleentry.
This could also be called the V.O. Key hypothesis. Key demonstratedthat Democratic primariesin the one-party South witnessed
consistently high levels of competition,and he arguedthat the cause
was the regularitywith which the Democraticnomineewon the general
election(1949, 408-16). Hypothesis1 transformsKey's pointprediction
into a comparativestatics prediction:as the probabilityof a party's
nominee winning the generalelection goes up, so does the probability
of multipleentryinto thatparty'sprimary.A generalelection candidate
has a good chance of winning if he or she is the incumbent;for nonincumbentcandidates,runningin the absenceof an incumbentincreases
the odds of winning (see, for example,Cox andKatz 1996 andGelman
and King 1990). Thus,multipleentryshould occur more often where
the incumbentdoes not runfor reelection.Wherethe incumbentdoes
notrunforreelectionbutthechoiceis nothis orhersto make(thatis, when
the incumbentdies or is arrested),candidatesof the incumbent'sparty
have a very good chance to win the seat, but the out-party'schances
do notnecessarilyimprove(Cox andKatz2002). Inthesecases, multiple
entry should occur more often only for the in-party.
Additionally,a weak showing in the previousgeneralelection by
the incumbentis a signalthatthe districtis receptiveto a new candidate:
the more marginalthe previouselection, the higherthe probabilitythat
a challengerwill defeat the incumbentin this election. Thus, multiple
entry should be more prevalent when there has been a marginal
election in the previous cycle. Finally, as an incumbent's career
progresses and the legislator grows fartheraway from constituents
(Fenno 1978), otherpoliticians should startto line up for the seat. This
progression says nothing, however, about the out-party's ability to
capturethe seat, since it does not affect the partisanbalance in the
district. Thus, as an officeholders career progresses, multiple entry
should occur more frequently in in-party primary elections.
Hypothesis 2: The higher the benefits of office, the more likely
thereis to be multiple entry.
UnintendedConsequences
443
Schlessinger(1966) was the firstto demonstratethatoffices vary
in their appealto candidates.More-desirableoffices might have more
multiple-entryprimaries,butthis relationshipis complicatedby the fact
thatelections for more-valuableoffices are often costlier to run in and
the high costs might detersome candidatesfromrunning.Since I used
data from only one office (a seat in the House of Representatives),I
avoidedhaving to sortout these complications.I conducteda different
test of Hypothesis2, takingadvantageof the factthat,withinthe House,
a seat in the majoritypartyis "worth"more thana seat in the minority
party (Binder 1997; Cox and McCubbins1993). As a result, multiple
entry should occur more frequently in primaries of the party that
expects to be the majorityparty in the upcomingyear.
Hypothesis 3: The more likely a non-incumbentcandidateis to
win theprimaryelection,themorelikelythereis to be multipleentry.
Many of the same factorsdeterminethe abilityof non-incumbent
candidatesto win primaryelections as determinetheir ability to win
generalelections. Forbothgeneralelections andin-partyprimaries,the
probability of victory is largely driven by the vulnerability of the
incumbent;thus, all the predictionsaboutin-partyprimariesthatresult
fromHypothesis 1 also applyhere.This is not the case, however,when
it comes to out-partyprimaries.There is no incumbentin out-party
primaries,so winning does not dependon incumbentattributes.7
Stone andMaisel (2003) introduceone factorthatdistinguishesa
candidate's probability of winning the primary versus the general
elections. Theynotethat,as a district'spartisanbalanceleansmoretoward
thepartyof a givencandidate,thatcandidatehasa worsechanceof winning
the primaryelection(becauseof increasedcompetitionfromothercandidates),but a betterchance of winning the general election if he or she
wins the primary.A candidate'schance of winning office is the result
of multiplyingthese firsttwo subprobabilities
(Stone and Maisel 2003,
952). This overallprobabilityis highestwhen the districthas a relatively
even partisanbalance(964-66). Thus,if we combineHypotheses1 and3,
multiple entry should be higher when supportfor the Democratic
and Republican parties is evenly balanced within the district.
Hypothesis 4. Multiple-entry primary elections occur less
frequentlywhere there are mechanisms in place to coordinate
politicians'actions.
Here I discuss multiple entry resulting from primary election
conditionsthatapproximategame (b, b). These cases differ fromthose
occurringin (c, c) because in (c, c), multipleentryresultsdirectlyfrom
444
Jeffrey Lazarus
candidates' preferences. In (b, b), it results from a failure among
candidatesto coordinatea mutuallybeneficial outcome (thatis, one of
the Nash equilibria).Multipleentrycan be curtailedif candidatescoordinateon a single equilibriumin which one candidateentersandothers
do not. Althoughcandidatesgenerallydo not achieve this coordination
on theirown, in some cases institutionsareableto administerit. Multiple
entry should occur less often when such administrationis possible.
One formthis administration
takesis statepartyendorsements:in several
a particularcandidate for the primary
leaders
endorse
states, party
election nomination. In some states, endorsementoccurs by formal
sanction of state law (resulting in caucus or convention nomination
mechanisms);in others,endorsementoccurswithoutsuch sanction.In
all of these states, the party imposes on any candidate beside the
endorseereal and costly obstaclesto running,althoughthese obstacles
are more costly in stateswhere endorsementsare formal(Jewell 1984;
Jewell and Moorehouse 2001). Thus, multiple entry should occur
less often in states with party endorsement than in states without,
and the effect should be stronger where the endorsement is
sanctioned by law.
II. The Determinants of Multiple Entry
and Divisive Primaries
I tested the hypothesesderivedin the previous section using data
from major-partyHouse primariesfrom 1976 through1998. The tests
consist of a series of logistic regressionsin which the dependentvariable is coded as 1 if multiple entry occurred(that is, if the primary
electionhadtwo or morepeople runningin it), andas 0 if it did not (that
is, if the election had one or zero people running).I separatedprimary
elections into four discrete groups-incumbent-party Democrats,
incumbent-partyRepublicans, out-party Democrats, and out-party
Republicans-and I ran one logit for each group. Each independent
variablecorrespondsto one of the predictedrelationshipspresentedin
italics in the previoussection, with two exceptions.The progressionof
an incumbent'scareer,discussed under Hypothesis 1, is represented
by two variables: IncumbentAge and Incumbent Tenure.Additionally, I separatedpartyendorsementsinto those sanctionedby state law
and those that were not. Finally, I addedtwo control variables:South
controlsfor the lingeringeffects of one-partydominancein the South
as identified by Key (1949),8 and state gained/lost seats controls for
the effects of redistricting.A complete list of independentvariables
appearsin Table 1.9
UnintendedConsequences
445
TABLE1
Variables
Independent
Variable
Definition
Probability of Winning General Election
Incumbent Voluntarily
Leaves
A dummy variable coded as 1 if the district's incumbent
voluntarily decided not to seek reelection (retired from
politics or sought another seat), and coded as 0 otherwise
Incumbent Leaves due
to Death or Arrest
A dummy variable coded as 1 if the district's incumbent
did not run due to circumstances out of his or her control,
including death or legal obstacles to running, and
coded as 0 otherwise
Lagged Incumbent Vote
The incumbent's share of the two-party vote in the
previous election
Incumbent Tenure
The number of terms the incumbent has served in office
Incumbent Age
The incumbent's age in years at the time of the election
Value of the Office
Majority
A dummy variable coded as 1 if the party holding the
primary holds a majority in the House, and coded
as 0 otherwise
Probability of Winning Primary and General Elections
Partisan Balance
Absolute value of (50 - Share of vote received in
district by Democratic presidential candidate)
Coordinating Mechanisms
Legal Party
Endorsement
A dummy variable coded as 1 if state law allows state
party organizations to formally endorse one candidate for
the nomination and restrict other potential candidates'
access to the primary election ballot, and coded
as 0 otherwise
Informal Party
Endorsement
A dummy variable coded as 1 if state party organizations,
without legal sanction, endorse one candidate for each
available seat and afford that candidate resources for use
in the election that it does not afford to other candidates,
and coded as 0 otherwise
Control Variables
State Gained/Lost Seats
A dummy variable coded as 1 if the district is in a state
that gained or lost seats because of reapportionment
since the previous election, and coded as 0 otherwise
South
A dummy variable coded as 1 if the primary occurs
in a Southern state, and coded as 0 otherwise
446
Jeffrey Lazarus
TABLE 2
Estimation
of the Determinants
Logistic
of ContestedPrimaryElections
(standarderrorsin parentheses)
Variable
Predicted
Incumbent- IncumbentRelationship:
Party
Party
Out-Party Out-Party
in-party/out-partyDemocrats Republicans Democrats Republicans
Incumbent Voluntarily
Leaves
+/+
2.08***
(.164)
2.94***
(.197)
1.28***
(.168)
1.37***
(.149)
+/0
Incumbent Leaves
due to Death or Arrest
1.21**
(.410)
.845
(.605)
-.698
(.623)
.120
(.456)
Lagged Incumbent Vote + / +
-.002
(.003)
-.008
(.004)
-.018***
(.003)
-.018***
(.003)
Incumbent Age
+/0
.014**
(.005)
.036***
(.008)
-.002
(.009)
.005
(.005)
Incumbent Tenure
+ /0
.004
(.006)
-.0003
(.010)
-.002
(.009)
.002
(.007)
Majority Party
+/+
1.15**
(.265)
.451
(.339)
1.67***
(.298)
-.444*
(.219)
Legal Party
Endorsement
/
-1.24***
(.148)
-1.32***
(.219)
-1.14***
(.158)
1.62***
(.167)
Informal Party
Endorsement
-/
-.278***
(.114)
-1.02*
(.445)
.278
(.144)
-.732***
(.098)
Partisan Balance
-/
.024***
(.005)
.016
(.010)
-.033***
(.008)
-.036***
(.005)
South
-.411 ***
(.099)
-.993***
(.159)
-.732***
(.120)
-.732***
(.098)
State Gained/Lost Seats
-.289
(.209)
.219
(.299)
-.550*
(.251)
-.569**
(.215)
Constant
-1.99***
(.332)
-2.12***
(.529)
-.623
(.387)
1.32***
(.314)
N
3,001
2,085
2,088
2,989
Log likelihood
-1,815
-920
-1,280
-1,698
.105
.200
.1103
.133
Pseudo-R2
Note: Year dummies not shown.
***p < .001; **p < .01; *p < .05.
-
UnintendedConsequences
447
The resultsare presentedin Table2. All told, thereare 36 tests of
18 predictions generatedby the game-theoreticmodel. Of these, 26
bear out as the model predicts, indicating that rationalentry plays a
very importantrole in determiningwhere multiple entry occurs. One
unexpectedfinding amongthe controlvariablesis that Southernstates
witnessed less competitionin their House races than did other states,
a reversalfrom Key's (1949) depiction of Southernpolitics.1 Among
othervariables,some of the most interestingfindings are the contrasts
between groups. For instance, if an incumbent has done poorly in
the previouselection, multipleentryis more likely in the out-partybut
not in the incumbent'sparty,indicatingthatthis signalof vulnerabilityis
received exclusively by the members of that other party.Also, the
incumbent'svoluntarydeparturefromthe districtresultsin an increase
in the likelihood of multiple entry across the board, but involuntary
departuresaffect only in-partyDemocrats.Among the tests that did
not result as predicted, there are eight in which the model predicts
a significant result and the coefficient is not statisticallydiscernable
from 0, and only two coefficients are significant and are in the
wrong direction.11
Despite the model's success at predicting the occurrence of
multiple-entryprimaries,"multipleentry"is not the same as "divisive."
Nevertheless,the two arevery closely entwinedtheoretically:multiple
entry is the first necessary condition for a divisive primary,and the
factorsleadingto multipleentry(such as Hypothesis 1's valuableparty
nomination)should also lead candidatesto care more about winning
and to create a divisive primary.Additionally, the dummy variable
MultipleEntry and Margin of Primary Victorycorrelateat .89. As a
result, the same logic that predicts multiple entry should also predict
divisive primaries.To check this hypothesis,I estimatedordinaryleast
squares(OLS) regressionmodels of Margin of Primary Victorywith
the same set of independentvariables.Results appearin Table3. Note
thatthe dependentvariableis inverselyrelatedto primarydivisiveness
in theseestimations,sinceas a primarygrowsmoredivisive,thewinner's
margin of victory diminishes. Thus, the predicted signs for each
coefficient are reversed from those in Table 2. The results indicate
that the rational-entrymodel fares just as well in predicting a close
primaryas it did in predictinga contested primary;in some cases, it
did better.
Jeffrey Lazarus
448
TABLE 3
of
Estimates
PrimaryElection VictoryMargins,
Regression
1976-98
(standard errors in parentheses)
Variable
Predicted
In-Party
Out-Party Out-Party
Relationship: In-Party
Democrats Republicans Democrats Republicans
in-party/out-party
-/
Incumbent Voluntarily
Leaves
Incumbent Leaves due
to Death or Arrest
-/0
- / -
Lagged Incumbent Vote
Incumbent Age
Incumbent Tenure
Majority Party
-
/0
-/0
-/-
-44.9***
(1.71)
-51.6***
(1.75)
-19.2**
(2.86)
-20.4***
(2.56)
-27.5***
(5.04)
-15.6**
(5.98)
9.89
(10.1)
.679
(7.99)
-.002
(.03)
.076*
(.035)
.271***
(.062)
.184***
(.049)
-.255***
(.064)
-.254**'
(.067)
.004
(.115)
-.154
(.097)
-.080
(.079)
-.039
(.097)
-.006
(.172)
.058
(.121)
-12.9***
(3.25)
-2.94
(2.56)
-25.4***
(5.54)
6.13
(5.01)
Legal Party
Endorsement
+/+
9.49***
(1.58)
8.68***
(1.56)
18.6***
(2.65)
23.4***
(2.32)
Informal Party
Endorsement
++ +
2.99*
(1.45)
11.81 ***
(3.16)
-8.02**
(2.75)
12.3***
(3.4)
Partisan Balance
+/+
-.235***
.057
-.132
(.090)
.401 *
(.154)
.488***
(.089)
South
1.77
(1.21)
7.43***
(1.26)
6.86**
(2.32)
7.08***
(1.84)
State Gained/Lost Seats
1.61
(2.60)
-4.12
(2.73)
9.48*
(4.59)
9.73*
(3.86)
105.2 **
(3.96)
101.6***
(3.89)
52.3***
(7.07)
48.65***
6.26
2,883
2,000
1,771
2,433
.258
.351
.099
.113
Constant
N
Adj R2
Note: Year dummies not shown.
***p < .001; **p < .01; *p < .05.
UnintendedConsequences
449
III. Contested Primaries and General Election Results
Section II indicatesthatthe attractivenessof the primaryelection
to non-incumbentsis a majordeterminantof its level of divisiveness.
Table2 demonstratesthat candidateperceptionsof theirability to win
explain the occurrenceof multiple-entryprimaries;Table 3 demonstrates that the same perceptions explain primary divisiveness. In
particular,the variablesgeneratedby Hypothesis 1 measurethe degree
to which the seat is winnable for non-incumbents. Most of these
variablesstronglycorrelatewith both multipleentryand primarydivisiveness. These results supportthe idea that candidates'perceptions
of the likelihood of winning the seat cause primariesto be divisive. I
now turnto the next question:What effect does divisiveness have on
general election outcomes?
Many studies have asked this questionbefore, and all have used
a similarmethodology:estimatethegeneralelectionoutcomeandinclude
a measure of divisiveness in the model. Although the literaturehas
never settled on the best way to operationalizedivisiveness, all studies
measurethe closeness of the primaryelection, but they do not account
for candidateperceptionsof how attractivethe primaryis to run in.12
This methodology does not account for the spurious nature of the
relationshipbetweenprimarydivisiveness andgeneralelection results,
even in studies that attemptto control for reciprocalcausality (Born
1981).
In this section, I discuss how I replicatedthe same analysis as
previousstudiesbutintroducedtwo new operationalizationsof primary
divisiveness to accountfor candidateperceptionsof the attractiveness
of running for office. One variable is Total Number of Candidates
who appearedin eitherparty'sprimaryelection ballot in that district.
The results of the previous section (that more candidatesenter races
that are more attractive)indicate that this variableaccuratelyreflects
elections' relative attractiveness.The second measure of attractiveness is Total Amount of Money spent by losing candidates in the
district'sprimaryelections. 13 High spendingin the primaryelection is
an indicationthatat least one groupof strategicactors-donors-feels
thatthe primaryelection'sprize (the rightto runin a generalelection) is
worthwinning.14Additionally,I followed earlierstudiesby includinga
measure that reflects the closeness of the primaryoutcome, Margin
of Primary Victory.
I tested two sets of predictionsfor this section. First,divisiveness
shouldhave a differentrelationshipwith generalelection outcomesfor
incumbent-andout-partyprimaries.The previoussection revealedthat
450
Jeffrey Lazarus
divisive primariesoccur where incumbentsare vulnerable;vulnerable
incumbentsdo worse in the general election than do strong incumbents.If vulnerableincumbentswin primariesdespitetheirvulnerability
(and thus go on to compete in the general election), then measuresof
primary divisiveness should correlate with challengers doing better
in the general election and incumbents doing worse. Vulnerable
incumbentsdo win primaries.In the 12 election cycles includedin the
dataset,4,713 incumbentsranin a primaryelection andonly 57 of them
lost. Of those, 10 lost to otherincumbentsin a redistrictingyear,so only
47 incumbents-about 1%of those who ran-lost a primaryelection
to a challenger.15Despite incumbents'relative safety in primaryelections, however,theirvulnerabilityis a real concernin generalelections
since challengersuse this factor to determinewhetheror not to enter
races; this relationshipbetween incumbentvulnerabilityand primary
election challenges is borne out in the estimates of the likelihood of
incumbent-partychallenger entry in Table 2. The end result is that
most vulnerableincumbentssurvive close primariesbut then receive
vigorous attackfrom their generalelection opponents.
The second proposition is that candidates' perceptions of the
likelihood of winning office if they enter a primaryelection-not the
closeness of the primary election itself-are the root cause of any
relationshipbetweenprimarydivisivenessandgeneralelectionoutcomes.
Thus, the coefficients associated with variablesthat reflect candidate
perceptionsof race attractiveness(numberof primarycandidatesand
amountof primaryspending)shouldbe significant.The coefficient on
the variablemeasuringprimarycloseness (Marginof Primary Victory)
shouldnot be significant.If the conventionalwisdomregardingdivisive
primariesis trueanddivisivenessdirectlyaffectsgeneralelectionresults,
thenthe oppositewill hold:the closenessvariablewill be significant,but
the variablesthat reflect candidateperceptionswill not be. Finally, if
the underlyingrelationshipis one of reciprocalcausality,as suggested
by Born (1981), then both sets of variableswill be significant.
The data include all House general elections between 1984 and
1998. In addition to the key independentvariables, I also included
standardHouse electionresultspredictors:a challenger-qualitydummy
(coded as 1 if the challengerhad won electoraloffice previously,as 0
otherwise),'6the incumbent'sshareof the two-partyvote in the previous
election,the log of candidatespending(adjustedfor inflation),the district
normal vote (represented as the percentage of the two-party vote
received by the party's presidentialcandidatethat year or two years
prior), and a dummy variable for each year included (excluding
the most recent). I also includeda controlvariableto separatecaucus-
UnintendedConsequences
451
TABLE 4
The Effect of PrimaryElection Divisiveness
on GeneralElection Vote Share
(standarderrorsin parentheses)
Variable
Predicted
Democratic Republican Democratic Republican
Relationship:
Challengers Challengers Incumbents Incumbents
challengers/incumbents
-.042**
(.015)
-.007
(.017)
Primary Election
Vote Margin
0/0
.004
(.007)
-.009
(.006)
Total Candidates in
Both Primaries
+/-
.623***
(.188)
.469***
(.132)
-3.01**
(.202)
-3.19***
(.237)
+/Total Spending by
Losing Candidates in
Both Primaries ($10,000's)
.479***
(.137)
.406***
(.118)
.312
(.209)
.297
(.227)
Challenger Experience
3.83***
(.495)
3.22***
(.472)
Incumbent's Vote in
Previous Election
-.092***
(.012)
-.038***
(.011)
Log of Candidate Spending
.830***
(.053)
.855***
(.047)
District's Presidential Vote
.295***
(.028)
.379**
(.016)
-5.85***
(.824)
.102***
(.017)
-5.33***
(.399)
.396***
(.024)
-5.12***
(.812)
.108***
(.019)
-6.74***
(.484)
.306***
(.043)
-2.23*
(.874)
Caucus Primary
-1.62**
(.553)
-.321
(.531)
-1.46
(.874)
Constant
15.6***
(1.99)
11.02**
(1.45)
124.83***
(6.14)
1,048
1,357
1,633
1,233
.520
.651
.505
.498
N
Adj R2
149***
(7.58)
Note. Year dummies not shown.
***p < .001; **p < .01; *p < .05.
mechanisms.Again,
nominatingmechanismsfromprimary-nominating
I divided observationsinto Democrats and Republicans,incumbents
and challengers.
I excluded open seats from the analysis because the theory
presented in this article makes no predictions about them. That is, I
arguethatthe relationshipbetweengeneralelectionresultsandprimary
divisivenessis drivenby whatnon-incumbentsthinkabouttheirchances
of beatingthe incumbent.In open-seatraces, thereare no incumbents,
452
Jeffrey Lazarus
so non-incumbent perceptions of incumbent vulnerability do not
contributeto candidates'decisionsas to whetherornotto entertherace.'7
The resultsof these regressionsappearin Table4, again with the
predictedsigns of coefficients indicated.The first groupof predictions
bears out entirely. Challengers who come out of highly contested
primariesfare betterin the generalelection thando otherchallengers,
andincumbentswho come out of highly contestedprimariesfareworse
thanotherincumbents.For challengers,boththe numberof candidates
andthe amountof money spentin the primaryelection has a significant
effect on general election vote returns.Each additionalentrantin the
primaryelection is associatedwith a Democraticchallengerreceiving
almost two-thirdsof a percentagepoint more in the general election,
and a Republicanchallengerreceiving almost half a percentagepoint.
Between one and eight candidatescompetedin 99% of all challengerpartyprimaryelections; the differencebetween being the only candidaterunningfor office andrunningin a field of eight is 4.4 points in the
generalelection for Democratsand 3.3 for Republicans.Furthermore,
spendingby primarycontestantsis associatedwith challengersdoing
betterin the general election: for each additional$10,000 spent in the
primaryelection,challengersreceive slightlyless thanhalf a percentage
point in the generalelection. This is a substantialeffect (almost seven
points), since losing candidateshave cumulatively spent as much as
$1.6 million duringa primarycampaign.
Incumbents,on the other hand, do worse when coming out of
highly contested primaries.The amount of money spent by primary
election losers has no significant influence on incumbents' general
election fortunes.But the numberof candidatesrunningto unseat the
incumbentcan indicatereal trouble.Each additionalprimaryelection
candidateresultsin the averageincumbentlosing approximatelythree
full percentagepoints in the general election. The differencebetween
having no competition and runningin a "full"primaryfield of eight
candidatesis 21 percentagepoints for the averageDemocrat,22.5 for
the averageRepublican.Contraryto the predictions,MarginofPrimary
Victoryis associatedwith doing somewhatbetterin the generalelection
among Democratic incumbents.The same is not true for Republican
incumbents,and the effect is relatively small. The differencebetween
winninga primaryelection by 1 point andwinningit by 100 (havingno
competition)is aboutfourpoints at the polls in the generalelection.
The second set of predictions,thatcandidateperceptionvariables
have a significant relationshipwith general election vote returnsbut
primarycloseness variablesdo not, is also confirmed.The results for
challengers supportthe unintended-consequencestheory unambigu-
UnintendedConsequences
453
ously. Forchallengersof bothparties,the relationshipbetweenprimary
election closeness and general election outcome is not significant,
whereas the relationshipbetween both measuresreflecting candidate
perceptionsandgeneralelection outcomeis significant.The resultsfor
incumbentsalso supportthe unintended-consequences
theory,although
not quite as forcefully.One of the two candidateperceptionvariables
has a significant relationshipwith incumbents'general election vote
shares,andthatrelationshipis very strong.Additionally,primaryelection
closeness, as measuredby marginof victory,has a significantrelationship with general election vote only among Democratic incumbents.
The substantive impact of that variable is dwarfed by that of the
relationshipbetween generalelection resultsand the numberof candidatesrunningin the primary.Whereasprimaryelection victory margin
can move incumbents' general election vote shares 4.3 percentage
points, the numberof candidatesin the primaryelection can move vote
shares 21 points. The effect of the perceptions-reflectingvariable is
about five times the effect of the closeness variable.
IV. Conclusion
The unintended-consequencestheory is both the first theory to
explain why divisive primariesoccur and the first to offer a falsifiable
explanationof the relationshipbetweenprimarydivisivenessandgeneral
electionoutcome.18Divisive primaryelectionsoccurbecausecandidates
generallyenterracesthey thinkthey can win. The most-winnableraces
involve vulnerableincumbentsand a disproportionatenumberof nonincumbentswho enter those races to create divisive primaries.This
behaviorresultsin primariesin which incumbentsaremost likely to do
poorly in the general election, and it creates a relationshipbetween
primarydivisiveness and generalelection results:where primariesare
divisive, incumbentsare vulnerableand thus likely to do poorly in the
general election. Thus, divisive primariesoccur where, on average,
incumbentsdo worse in general elections and challengersdo better.It
is in this way thatdivisive primariesare an unintendedconsequenceof
behaviordirectedtowardthe goal of winning a generalelection.
The literaturestudyingdivisiveprimaryelectionsgoes backalmost
40 years, and much of it is devoted to searching for evidence of a
conventional wisdom-that a divisive primary harms the primary
winner's chances in the general election. Despite the conventional
wisdom's wide acceptance, the search for supportingevidence often
producednull or inconclusive findings. The most recent scholarship
even indicatesthatthe conventionalwisdom may be wrong:challengers
454
Jeffrey Lazarus
tendto do betterwhen comingoutof divisiveprimaries(Alvarez,Canon,
and Sellers 1995;Arbourand McKenzie 2002; Herrnson2000; Hogan
2003). Threedistinctlimitationsin the studyof divisiveprimaryelections
have producedthis patternof results.First, authorshave not offered a
theory of why a divisive primaryhurts in the general election. Some
offer conjectures, but they are usually ad hoc and do not contain
falsifiablehypotheses.Second,the conventionalwisdomandthe studies
searchingfor its evidence have been unableto accountfor many of the
empiricalresults the literaturehas produced,including frequentnull
findings andresultsin which divisiveness helps non-incumbentcandidates.Third,the literaturehas lookedfor causalityin only one direction:
scholarshave looked for evidence thatprimarydivisiveness influences
general elections, and so that is what they have found. For example,
Alvarez,Canon,andSellers(1995) "conclusively"findthat"incumbents
arehurtand challengersarehelped by hotly contestedprimaries"(13).
This articleaddressesall threeshortcomings.First,theunintendedconsequences theory offers a formal model of primaryelections in
whichthelevel of divisivenessis explainedby non-incumbents'
prospects
of winning the general election. It also offers falsifiable hypotheses
that relate this model to general election outcomes. Empirical tests
supportthemodel.Second,thetheoryprovidesa theoreticalunderpinning
to the literature'spreviouslyunexplainedpatternof findingsby offering
independentpredictionsfor challengersandincumbents.Thesepredictions, that challengers do well and incumbentsdo poorly in general
elections where a divisive primaryhas occurred,accuratelyreflect the
literature'sresults.Third,the theoryadvancesa more refinedperspective on causalityin divisiveprimaryelections.Primarydivisivenessdoes
not cause generalelectionresults;rather,expectationof generalelection
results causes primariesto be divisive.
Divisive primariesdo not hurtthe partynominees who come out
of them, as the common wisdom assumes;the victors do not enterthe
fall campaign hobbled from a debilitatingstruggle. For that matter,
divisiveness does not help nominees either.The strongestcandidatesin
the field may survive a brutallyDarwinianprocess to win the primary
election, but the primary election is not what makes the surviving
candidatesstrong.Althoughprimarydivisivenesscorrelateswith general
election success for challengers, and with general election struggles
for incumbents,the directionof causality is the opposite of what the
literaturehas long held.
Jeffrey Lazarus is Assistant Professor of Political Science,
Georgia State University, PO. Box 4069, Atlanta, GA 30302-4069.
UnintendedConsequences
455
APPENDIX
PrimaryElection EntryGame
This is a one-shotgame in which two playerssimultaneouslydecide whetheror
not to entera primaryelection, which I treatas a lottery.Both playerschoose fromtwo
strategies:Enterthe raceor Don'tEnterthe race.The prize for the winnerof this lottery
is entry into the general election, which I treat as another lottery. There are two
nonstrategicplayers, Dummy and OtherParty.The game is built upon the following
assumptions:Player 1 and Player 2 are members of the same party.Both are strong
candidatesand the only strong candidatesavailable to run in the district. Both are
strategic,andeach will enterthe race only if the expectedutility of doing so is positive.
If both candidatesenter,then they will run againsteach otherin a primaryelection. If
only one candidateruns, then the primaryopponentis Dummy, who is nonstrategic
(i.e., will run regardlessof the probabilityof winning), not likely to win the primary
election againstPlayer 1 or Player 2, and not expected to win the generalelection. If
neithercandidateruns, then Dummy runs in, and wins, the primaryelection. Finally,
the winner of the primaryelection runs in the generalelection againstOtherParty.
Given these parameters,the value of enteringthe race to Player 1 is
P(2)U(runl2)+ (1 - P(2))U(runlD),
where P(2) = probabilityof Player2 entering,U(runl2)= the utility of runningagainst
Player 2, and U(runlD)= the utility of runningagainst Dummy. Players'utilities are
determinedby the function
Uj(seec officei) = PJ [HIB - rc + (1 - n J)rJ]+ (1 - PJ)qJ- C,
where PJ = probabilityof candidatej winning the primaryelection of office i, q = the
value of losing the primaryelection, C = the cost of runningin the primaryelection,
l = the probabilityof winningthe generalelection given a win in the primaryelection,
B = the value of winningthe generalelection,r = the value of losing the generalelection,
c = the cost of runningin the general election, and n = the discountfactorassociated
with costs c. For exposition, assume q = r = 0. The utility functionnow reads
- itc]- CJ.
Uj(seekofficei)= P,-[HdpB
Substitutingthis functioninto the first equationand eliminatingsub- and superscripts
from the utility function,we can express the value of runningto Player 1 as
P(2)[P2(IB - rc) - C] + (1 - P(2))[PD(HB- nrc)- C],
wherePi representsthe probabilityof defeatingPlayeri in a primaryelection. Thus,the
conditionsunderwhich Player 1 chooses Enter can be fully expressedas
P(2)[P2(HB- rc) - C] + (1 - P(2))[PD(nB - nc) - C] > b,
where b representsPlayer 1's next-best option, should he or she decide not to run.
Includingb allows us to specify payoffs, shown in TableAl. Each player who enters
receives some probability(P1 or P2) of winning the primaryelection and of thereby
being allowed to enterthe generalelection. If only one playerenters,then the entering
playerreceives the generalelection with certainty.Thatis, Dummy is assumedto lose
the primaryelection.The generalelection is itself a lottery,andthe enteringplayerwins
the benefits of office B with probabilityH, less the discountedcosts of the futurerace
456
Jeffrey Lazarus
TABLE Al
to
in
the
Payoffs
Players
Primary Election Entry Game
Player 1
Enter
Not Enter
P2(nB - rc) - C2
b
Enter
P,(nB - Irc)- C,
Player 2
IB-
HB - c - CD
c-CD
b,
Not Enter
b,
b,
mk(where k is a discountfactor).I name the outcomes for exposition:if both players
enterthe race,theneach mustruna HardRace againstthe other;if one playerentersand
the other does not, then the enteringplayer runs an Easy Race againstDummy.Any
playerwho does not enterthe race receives the Next-Bestcareeroption.
By assumption,C, < CD and C2< CD.That is, the cost of runninga primary
election againstDummy (CD)is less thanthe cost of runningagainsteitherone of the
strongcontenders.Additionally,0 < P1 < 1 and 0 < P2 < 1. Given these conditions,the
value of Easy Race is always higherthanthe value of HardRace. Thus,the solutionto
the game dependson the value of the players' next-bestoptions in relationto the two
primaryraces. The possible payoff orderingsare designatedin TableAl.
Since the player designationsare arbitrary,there are six possible preferenceorderingpairs(andthus six differentgames for the playersto play): (a, a), (a, b), (a, c),
(b, b), (b, c), and(c, c). Five of the games aredominancesolvablewith uniquesolutions.
In the threegames in which players'preferencesdiffer,the dominant-strategyequilibrium is (Don't Enter,Enter),in which the player who most prefersNext-BestOption
chooses Don't Enter.In (a, a), in which both playersmost preferNext-BestOption,the
dominant-strategyequilibriumis (Don't Enter,Don't Enter).In (c, c), in which both
players least prefer Next-Best Option, the dominant-strategyequilibriumis (Enter,
Enter).The only game thatis not dominancesolvable is (b, b). Thereare,however,two
weak Nash equilibria:(In, Out) and (Out, In).
NOTES
I thankGaryCox, GaryJacobson,Neal Beck, Ben Nyblade, WalterStone, and
anonymousreviewersfortheirhelpfulcomments,andGaryJacobsononce againforhis
assistancein obtainingdata.
1. I excludedfromthe categoryof"serious"candidatesthose who runin orderto
publicizetheirbusiness,to have fun, or to exaltbeer,for example.I includedthose who
approximatethe traditional"rationalentry"models discussedmore fully in the article.
2. Kenney and Rice (1984) discuss reciprocalcausalitybut do not controlfor it
in their empiricalanalysis and do not separateincumbentsand challengers.Alvarez,
UnintendedConsequences
457
Canon,andSellers(1995) divide observationsfor incumbentsandchallengers,butthey
do not controlfor reciprocalcausality.All otherstudiesoverlookthe issues altogether.
3. In major-partyHouse primariesbetween 1976 and 1998, the top two votewithin 20 percentagepoints of each other 13%of the time, and within
finished
getters
10 pointsof each other8%of the time. If we limit the discussionto contestedprimaries
(i.e., those contests in which at least two candidatesactually run), then we find the
proportionsjump to 35% and 21%, respectively.
4. An ampleamountof researchconfirmsthis assumption.For instance,Maisel
and Stone (1997) find that"factorsrelatedto potentialcandidates'chancesof winning
the seat are [the] most influential"in theirdecisions aboutwhetheror not to run(1997,
85). For additionalevidence, see Bianco 1984; Bond, Covington,and Fleisher 1985;
Robeck 1982;and Squire 1989.
5. Otherformalmodels of candidateentrythatalso assume each candidatehas
an equal chance of winning also predictthat the numberof entrantsdepends on the
costs andbenefitsof entering.See Besley andCoate 1997 (91); Fedderson,Sened, and
Wright1990 (1014); and Weber1992 (6); for a summary,see Cox 1997 (153-58).
6. One hypothesis not directlyrelatedto this article's thesis is that Multipleentryprimaries occur rarely. Only two of the six games allow for the possibility of
both players enteringthe primaryelection. In (c, c), double entry is the equilibrium
outcome, and in (b, b) double entryis possible if the playerscannotcoordinateon one
of the two Nash equilibria.Assumingthatthe game situationsaredistributedacrossreal
primaryelections relativelyevenly-or at least in such a way that(c, c) and (b, b) are
not severely overrepresented-we expect multiple entry in primaryelections to be
more the exception than the rule. This is the case in the House of Representatives.
Among the 10,440 major-partyHouse primariesconductedbetween 1976 and 1998,
63% were not contested:eitherno candidateran (resultingin an uncontestedgeneral
election for the other party's nominee) or, more prevalent, a single candidate ran
unopposed.Multiple entry occurredonly in the remaining37% of primaries.
7. I do not claim thatthe incumbentdoes not influenceentryinto the out-party
primaryelection.Indeed,this incumbentinfluenceis relatedto the core of my argument:
vulnerableincumbentsmean more candidatesenter the out-partyprimaryand, as a
result,winningthatprimarybecomesharder.Once entryhas concludedandthe primary
election itself is conducted,however, the probabilitywith which any single candidate
defeatsthe othersdoes not dependon incumbentattributes-at least,not those attributes
traditionallymeasuredby political scientists.
8. The time periodof my datasetcovers the era in which Democraticcontrolof
the Southwas eroding.The Republicanpartywas still weak in the regionuntilthe early
to mid-1990s,however.Some elementsof "classic"Southernpoliticsmay thereforebe in
place throughoutthe dataset,andI controlledfor those elementswith the variableSouth.
9. Note that,because of the variable'soperationalization,as Partisan Balance
risesin value,thebalanceof the districtfalls.As a result,thevariablecorrelatesnegatively
with multiple entry.The same is true for Lagged IncumbentVote.
10. Key (1949) showed that the South saw a much higher degree of intraparty
electoral competitionunder one-partyrule, at least within the dominantDemocratic
party.The reversalof this trendin recent decades suggests that Southernincumbents
are less vulnerablethanothers.By one measurethey are: Southernincumbents'mean
vote shareis higherthanotherincumbents'for every yearbetween 1974 and 1998. This
458
Jeffrey Lazarus
finding implies that the South is still politically distinct from the rest of the nation;
is no longerone-partyrule,however,butsafeincumbents.
thedistinguishingcharacteristic
11. Some of these results are explainable,post hoc. The correlationbetween
marginalprevious election and multiple entry appearsonly in the out-party,which
makes sense if we assume that the incumbent'smarginalvote share relates only to
voters' willingness to replacethe incumbentwith a memberof the out-party.Second,
candidateswho challenge incumbentswho are growing distant from their districts
seem to zero in on incumbentage ratherthanlengthof service.Thisfindingis consistent
with the idea that the true underlyingvariable is "degreeto which incumbentis no
longerresponsiveto the district"andincumbentswhose retirementsaremost imminent
are less responsive. Third, there is no relationshipbetween involuntaryincumbent
retirementand multiple entryfor in-partyRepublicans,but this finding is determined
by only 30 cases out of the universeof 10,440.Fourth,majoritypartystatusis associated
with a higherprobabilityof multipleentryonly amongDemocrats;amongRepublicans,
majoritystatusdecreases the probabilityof multipleentry.This result could reflect a
recenttrendof multiple-entryprimariesoccurringless frequentlyamong both parties,
reflectedin Figure 1. Finally,for out-partyDemocrats,informalstateendorsementsare
associated with a higher probability of multiple entry, contrary to the predicted
relationship.But the relationshipis not discerniblefrom0 at significancelevelp < .05.
12. Alvarez, Canon, and Sellers (1995) use primaryelection spending to account for these preferences, but they do not acknowledge that spending relates as
much to candidateexpectations as it does to primarydivisiveness.
13. I included only losing candidates' spending because the Federal Election
Commissiondoes not separatepre-primaryandpost-primaryspendingin any election
prior to 2000. Thus, including primarywinners' spending would include money the
winnerspentin the generalelection. By using only losers' spending,I can assumethat
all money was spent prior to the primaryelection.
14.A thirdway to measurecandidateperceptionsof theirprospectsfor winning
might take into accountthe qualityof the primaryelection candidatepool, in addition
to the quantityof challengers.Well-qualifiedchallengerstendto runwhen theirchances
of winning arehighest;one could thereforeinferthatthe more qualifiedthe candidate
pool is in aggregate,the higherthe prospectsof winning are for any given candidate.
I attemptedto accountfor challengerqualityusing three differentoperationalizations
of the quality of the candidatepool. First, I simply added a dummy variable to the
regressionpresentedin Table4, indicatingthe presenceof an experiencedcandidatein
the primarycandidatepool. Second, I replaced the dummy with a cardinalvariable
that counted the numberof experienced candidates.This method was methodologically unsatisfying,however, because it double-countedexperiencedcandidates-i.e.,
they were counted in both the "totalcandidates"and "qualitycandidates"variables.
Lastly,I includedthe count variablebut changedtotal numberof candidatesto reflect
only those candidatesnot coded as quality challengers.In all three cases, the quality
of the primary election pool often had an insignificant relationship with general
election outcomes, with little change in the coefficients and standarderrorsof the
other key variables.
15. The 1982 reapportionment
victimswere Grisham(CA 33), Derwinski(IL 4),
Evans (IN 6), Lee (NY 33), Smith (PA 3), and Bailey (PA 21); in 1992, they were
Russo (IL 3), Bruce (IL 19), Miller (OH 10), and Staggers(WV 2).
UnintendedConsequences
459
16. I used a dichotomouscodingfor candidatequalityprimarilybecauseapplying
a more detailed coding would have requiredan overwhelming amount of work and
added very little value. Extant detailed codings have all been applied to general
election challengers to the Senate and House of Representatives,for whom data is
relatively easy to obtain.Obtainingthe same data forprimary election challengersto
the House-who outnumbergeneral election challengersby about3 to 1 and many of
whom are very obscure-would be onerous at best and impossible at worst. To the
best of my knowledge, data for the more elaboratemeasure employed in Green and
Krasno1988, for one, arenot availableat all for some of the time periodin my dataset.
Furthermore,the dichotomous measure is correlated with the Green and Krasno
measure at .8 (Jacobson 1990), meaning that any gains from the use of the more
detailedmeasurewould be marginalat best.
17. In incumbentraces,the correlationbetweenprimarydivisivenessandgeneral
election outcomes is drivenby a confoundingvariable,incumbentvulnerability.This
confounderis not present in open-seat races because there are no incumbents.As a
result,to ask if the same correlationexists for open-seatraces is to ask if an equivalent
confoundingvariableexists to cause it. If the answeris no, then we would expect there
to be no relationshipbetween primarydivisiveness and general election outcomes. I
repeatedthe Table 4 analysis on open-seat races, dividing observations into candidates of the same party as the incumbent who departedthe seat, and those of the
opposite party.For same-partycandidates,therewas no relationshipbetween primary
divisiveness and general election outcomes. For other-partycandidates,there was a
negative correlationbetween the number of candidates in the primaryand general
election outcomes, indicatingthe presence of a lurkingvariable.
18. Two methodshave been employedto identifythe causeof divisive primaries,
but neither is satisfactory.One line of research involves surveying party activists
abouttheir activities in the primaryand general elections (Comer 1976; Johnsonand
Gibson 1974; Stone 1986). The researchshows that activists working for a primary
loser tend to defect from the partyduringthe generalelection. Such activist defection
does not predict divisive primaries, however, because there is no indication that
activists workingfor candidateswho lost a divisive primaryelection defect more than
activists working for candidates who lost by a wide margin. The second line of
research looks at voters' tendencies to defect from the party after a close primary
election. These studies have a more directbearingon the relationship,but so far they
have only been conducted in the context of presidentialelections (Southwell 1986;
Sullivan 1977-78).
REFERENCES
Abramowitz,Alan I. 1988. "ExplainingSenateElectionOutcomes."AmericanPolitical
Science Review 82: 385-03.
Alvarez, R. Michael, David T. Canon, and Patrick Sellers. 1995. "The Impact of
Primarieson GeneralElectionOutcomesin theU.S. HouseandSenate."California
Instituteof Technology.Social Science WorkingPaper932.
Arbour, Brian K., and Mark JonathanMcKenzie. 2002. "The Effects of Divisive
Primarieson GeneralElectionFund-Raising."Presentedatthe annualmeetingof
the Midwest Political Science Association, Chicago.
460
Jeffrey Lazarus
Bernstein,RobertA. 1977. "Divisive PrimariesDo Hurt:U.S. Senate Races, 19561972."AmericanPolitical ScienceReview71: 540-5.
Besley, Timothy,and StephenCoate. 1997. "An Economic Model of Representative
Democracy."QuarterlyJournal of Economics 112: 85-114.
Bianco, William T. 1984. "StrategicDecisions on Candidacyin U.S. Congressional
Districts."LegislativeStudies Quarterly9: 351-65.
Binder, Sarah A. 1997. "Minority Rights, Majority Rule: Partisanship and the
Developmentof Congress."New York:CambridgeUniversityPress.
Bond, Jon R., Cary Covington, and RichardFleisher. 1985. "ExplainingChallenger
Qualityin CongressionalElections."Journalof Politics 47: 510-29.
Born,Richard.1981."TheInfluenceof HousePrimaryElectionDivisivenesson General
ElectionMargins,1962-76." Journalof Politics 43: 641-61.
Comer,John. 1976. "AnotherLook at the Effects of the Divisive Primary."American
Politics Quarterly4: 121-28.
Cox, GaryW. 1997. Making VotesCount.New York:CambridgeUniversityPress.
Cox, GaryW., and JonathanN. Katz. 1996. "WhyDid the IncumbencyAdvantagein
U.S. House ElectionsGrow?"AmericanJournalofPolitical Science40: 478-97.
Cox, GaryW.,andJonathanN. Katz.2002. ElbridgeGerrys Salamander:TheElectoral
Consequences of the ReapportionmentRevolution. New York: Cambridge
University Press.
Cox, Gary W., and Mathew D. McCubbins. 1993. Legislative Leviathan. Party
Governmentin the House. Los Angeles: Universityof CaliforniaPress.
Feddersen,TimothyJ., Itai Sened, and StephenG. Wright.1990. "RationalVotingand
CandidateEntryunderPluralityRule."AmericanJournalofPolitical Science34:
1005-16.
Fenno, RichardF. Jr. 1978. Home Style. House Membersin their Districts. Boston:
Little, Brown.
Gelman,Andrew,and Gary King. 1990. "EstimatingIncumbencyAdvantagewithout
Bias."AmericanJournalof Political Science 34: 1142-64.
Green,Donald,and JonathanKrasno.1988. "Salvationfor the SpendthriftIncumbent:
Reestimatingthe Effects of CampaignSpendingin House Elections."American
Journalof Political Science 32: 884-907.
Hacker, Andrew. 1965. "Does a 'Divisive' PrimaryHarm a Candidate's Election
Chances?"AmericanPolitical ScienceReview59: 105-10.
Herrnson,Paul. 2000. CongressionalElections. 3d ed. Washington:CQ Press.
Herrnson,Paul, and James G. Gimpel. 1995. "DistrictConditionsand PrimaryDivisiveness in CongressionalElections."Political ResearchQuarterly48: 117-34.
Hogan, Robert E. 2003. "The Effects of PrimaryDivisiveness on General Election
Outcomesin StateLegislativeElections."AmericanPolitics Research31: 27-47.
Jacobson,GaryC. 1983. "RunningScared:Electionsand CongressionalPolitics in the
1980s." In Congress. Structureand Policy, ed. Mathew McCubbinsand Terry
Sullivan.New York:CambridgeUniversityPress.
Jacobson,GaryC. 1990. "TheEffects of CampaignSpendingin House Elections:New
EvidenceforOldArguments."AmericanJournalofPolitical Science34: 334-62.
Jacobson,Gary C., and Samuel Kernell. 1981. Strategyand Choice in Congressional
Elections. New Haven, CT:Yale UniversityPress.
UnintendedConsequences
461
Jewell, Malcolm E. 1984. Parties and Primaries. NominatingState Governors.New
York:Praeger.
Jewell, MalcolmE., andSarahM. Moorehouse.2001. Political Parties andElectionsin
AmericanStates.Washington:CQ Press.
Johnson,DonaldBruce,andJamesR. Gibson. 1974. "TheDivisive PrimaryRevisited:
PartyActivists in Iowa."AmericanPolitical Science Review68: 67-77.
Kenney,PatrickJ. 1988. "SortingOutthe Effects of PrimaryDivisiveness in Congressional and SenatorialElections."WesternPolitical Quarterly40: 765-77.
Kenney, PatrickJ., and Tom W. Rice. 1984. "The Effect of PrimaryDivisiveness in
Gubernatorialand SenatorialElections."Journalof Politics 46: 904-15.
Kenney, Patrick J., and Tom W. Rice. 1987. "The Relationship between Divisive
PrimariesandGeneralElectionOutcomes."AmericanJournalofPolitical Science
31: 31-44.
Key, V.O.Jr. 1949. SouthernPolitics in Stateand Nation. Knoxville:The Universityof
Tennessee Press.
Piereson, James E., and TerryB. Smith. 1975. "PrimaryDivisiveness and General
ElectionSuccess:A Reexamination."JournalofPolitics 37: 555-62.
Robeck, Bruce W. 1982. "StateLegislatorCandidaciesfor the U.S. House: Prospects
for Success."LegislativeStudiesQuarterly7: 507-14.
Schlesinger,JosephA. 1966. Ambitionand Politics. Political Careers in the United
States. Chicago,IL: RandMcNally.
Segura, Gary M., and StephenP. Nicholson. 1995. "SequentialChoices and Partisan
Transitionsin U.S. SenateDelegations:1972-1988."JournalofPolitics 57: 86100.
Southwell, Priscilla L. 1986. "The Politics of Disgruntlement:Nonvoting and Defection among Supportersof NominationLosers, 1968-1984." Political Behavior
8: 81-95.
Squire, Peverill. 1989. "Challengersin U.S. Senate Elections." Legislative Studies
Quarterly14: 531-45.
Stone, WalterJ. 1986. "The CarryoverEffect in PresidentialElections."American
Political ScienceReview80: 271-80.
Stone,WalterJ., andL. SandyMaisel. 2003. "TheNot-So SimpleCalculusof Winning:
PotentialU.S. House Candidates'NominationandGeneralElectionProspects."
Journalof Politics 65: 951-77.
Sullivan,Denis G. 1977-1978. "PartyUnity,AppearanceandReality."Political Science
Quarterly91: 635-45.
Ware,Alan. 1979. "'Divisive' Primaries:The ImportantQuestions."BritishJournalof
Political Science 9: 381-84.
Weber,Shlomo. 1992. "An Equilibriumin ElectoralCompetitionwith EntryCosts."
York(Canada)Departmentof Economics. Unpublishedmanuscript.