Party System Fragmentation and Strategic Behavior in Diff\u000berent Contexts
Transcript of Party System Fragmentation and Strategic Behavior in Diff\u000berent Contexts
Party System Fragmentation and Strategic Behavior inDifferent Contexts
Kai-Ping Huang
Abstract
Previous studies argued that voters’ strategic action was the key to explaining
the gradual decrease in party system fragmentation, especially in an institu-
tionalized party system. This article, however, argues that a downward trend in
fragmentation is also likely to happen in a poorly institutionalized party system,
and it is mainly caused by elites’ rational entry. When parties rely on clien-
telistic networks to secure votes, they should be sensitive about the likelihood
of winning and withdraw from competition to avoid wasting resources when the
chance is slim. Testing the effects of rational entry and strategic voting on the
effective number of competitors simultaneously across nine elections in Thai-
land, the analyses confirm that strategic behavior of elites and voters asserts
different levels of importance to determine party system fragmentation before
and after the party system became institutionalized in 2001.
Keywords: rational entry, strategic voting, party system fragmentation, AV,
SMD, Thailand
1. Introduction
As is well known in the literature of electoral system effects, two mecha-
nisms from voters and elites help a restrictive electoral system put downward
pressure on the number of competitors at the district level (Cox, 1997). Vot-
ers, facing the psychological effect of electoral rules, engage in strategic voting5
as they pour votes to viable candidates to avoid wasting their votes. Parties,
in anticipation of strategic voting, withdraw from the competition when the
Preprint submitted to Elsevier October 4, 2014
chance of winning is slim. It seems that the effective number of competitors is
reduced in such a sequence: strategic voting leads to rational entry, which in
turn makes strategic voting easier in the next election. In other words, failure10
to reach the Duvergerian equilibrium is likely, foremost, due to the absence of
strategic voting.
Previous studies argued that failure to vote strategically could come from
several sources that obscured the information about the viability of parties and
candidates (Clough, 2007; Crisp et al., 2012). Voters in newly democratized15
countries face this situation more often since electoral competition is new and
little information about the viability of candidates and parties is available until
several rounds of elections have been held (Moser, 1999; Tavits & Annus, 2006;
Tavits, 2008). Thus, we would expect that the longer a country has elections,
the less fragmented a party system is over time when voters get more infor-20
mation to ascertain the likely effects of strategic voting. However, repeated
electoral competition does not guarantee that voters would vote strategically
since it also needs parties that can secure stable support and impel individual
candidates to join established parties (Moser & Scheiner, 2012). In this situa-
tion, voters are given a few options and the information needed to determine25
viable and nonviable competitors becomes easy to come by. In other words,
stable inter-party competition and institutionalized party system help convey
the information necessary for strategic voting.
Yet, this article argues that a downward trend in district-level fragmentation
is also likely to be observed in a poorly institutionalized party system, and it is30
not caused by strategic voting but elites’ rational choice. More specifically, in
an environment where party labels are insignificant for vote choice and instead
parties rely on individual candidates’ clientelistic networks to secure votes, the
assessment of the likelihood of winning should drive candidates to stay in or
withdraw from competition to avoid wasting resources. If a candidate realizes35
that the competitor’s network is more extensive than hers and, thus, the likeli-
hood to win a seat is slim, a rational actor would avoid entering the contest in
the first place. Voters, by contrast, are unlikely to engage in strategic behavior
2
when their votes are bought or cast to show allegiance to their patrons. Since
these voters do not have the liberty to cast the ballots with free will, it is unlikely40
that they would pour votes to another candidate with the intention to affect
the election outcomes. In other words, elites’ rational entry might play a larger
role in determining the number of competitors in such a context. By contrast,
once parties can structure votes and voters tend to vote for parties instead of for
candidates (i.e. a party system is relatively institutionalized), voters’ strategic45
behavior becomes the main mechanism to determine the number of competitors
at the district level.
Unlike previous studies that scrutinized the impact of only one of the strate-
gic actions on the number of competitors, this article tests the effects of rational
entry and strategic voting on district-level fragmentation simultaneously across50
nine elections in Thailand. Thailand is chosen because the country provides
different contexts conducive to testing the hypothesis. As Figure 1 shows, the
vote shares of the top two parties in each election change dramatically from
election to election before 2001. Since 2001, the top two parties are always the
Thai Rak Thai Party (Thais Love Thais, or TRT) or its successor parties1 and55
the Democrat Party. The high electoral volatility before 2001 is the result of
party switching by individual candidates. The vote shares of the TRT and the
Democrat Party stabilize after 2001, indicating that the party system has be-
come more institutionalized than before. The changing context in 2001 is the
watershed event that allows me to inspect the relative importance of rational60
entry and strategic voting in affecting district-level party system fragmentation
at different times. Moreover, a one-country study provides the opportunity to
compare the before and after effects of key factors while holding other confound-
ing variables constant, such as social heterogeneity, political institutions, and
the level of power centralization, which are found to affect party system frag-65
mentation alongside electoral systems (Shugart & Carey, 1992; Amorim Neto &
1The TRT has been disbanded and changed its party name twice. The two successorparties to the TRT are People’s Power Party (PPP) that won the 2007 election, and the PheuThai Party (PT), which won the 2011 election.
3
Cox, 1997; Clark & Golder, 2006; Singer & Stephenson, 2009; Hicken & Stoll,
2011; Chhibber & Kollman, 2004).
(Figure 1 here)
The findings suggest that rational entry does play a more important role70
in affecting district-level party system fragmentation before 2001, but strategic
voting becomes the main mechanism afterwards. In other words, party system
fragmentation might decrease over time not due to voters’ strategic behavior
but elite decisions, which are relatively neglected in previous studies. In what
follows, I first explain the relationship between electoral rules and strategic be-75
havior of elites and voters within districts. Based on the theoretical relationship
I have proposed, I provide my arguments and hypothesis. I then illustrate the
overall trend of party system fragmentation at the district level between 1986
and 2011 in Thailand. I investigate the determinants of district party system
fragmentation in each election. I discuss the findings and their relevance to the80
extant theories in the conclusion.
2. Electoral Rules and Strategic Behavior within Districts
Since the publication of Making Votes Count, the relationship between elec-
toral systems and party system fragmentation at the district level has been
tested in numerous studies. Cox (1997) argues that because electoral rules de-85
termine winners and losers in district elections, they should induce strategizing
from both candidates and voters who care about the election result. Strategic
behavior of candidates and voters within districts is the mechanism that leads to
Duverger’s Law at the district level. With regard to voters, strategic voting or
strategic defection happens when voters pour votes towards viable contenders to90
affect the current election outcome.2 Parties and candidates, in anticipation of
2Strategic voting is also called strategic defection because it refers to a phenomenon inwhich a voter’s vote choice and her preference do not match (Kawai & Watanabe, 2013; 625).That is, the voter defects from her most preferred choice in hopes of preventing her leastpreferred party/candidate from winning the election.
4
strategic voting, would calculate the likelihood of winning and decide whether to
run for established parties (strategic entry) or withdraw from the contest when
the chance is slim (rational entry). If strategic behavior occurs, the number of
viable or serious contenders under different electoral rules should be close to95
or smaller than M+1, where M is the district magnitude (the number of seats
allocated in each district). The deduced result of M+1 is called the Duvergerian
equilibrium.
While Cox argues that there are two mechanisms (strategic voting and elite
actions) that help a restrictive electoral system bring down the number of com-100
petitors at the district level, he seems to weight strategic voting more than the
strategies taken by parties and candidates to impose the upper bound on the
effective number of competitors. Elites (parties and candidates) make entry de-
cision based on the anticipation of whether they are likely to become the victim
of strategic voting. Thus, even though there are many competitors in a race (as105
those early rounds of elections in newly democratized countries), the number of
competitors will gradually decrease when voters learn to vote strategically and
elites take the message to make entry decision in the next election.
Previous studies that investigated electoral system effects across different
countries also found that failure to vote strategically usually led to more en-110
trants in elections and resulted in higher party system fragmentation even un-
der a restrictive electoral system. For instance, investigating the district-level
fragmentation in several post-communist countries, Moser & Scheiner (2012)
found that although the fragmentation level was high in these countries in the
early rounds of elections, the fragmentation level decreased in some countries115
(e.g. Hungary) faster than in others (e.g. Russia and Ukraine). The varying
dropping rate was explained by the diverging degree of party system institution-
alization across these countries. Since parties in a less institutionalized party
system do not structure votes, independent candidates usually compete side
by side with parties. As a result, competitors mushroom and it is difficult for120
voters to distinguish front-runners from hopeless losers. In this situation, vot-
ers are unlikely to vote strategically since the effect of strategic voting is less
5
predictable. When voters do not vote strategically, votes are relatively equally
distributed among candidates, which, in turn, encourages entry since everyone
has almost the equal chance of winning. Thus, we generally expect that the125
effective number of competitors at the district level would stay high in a less in-
stitutionalized party system even after elites and voters have experienced several
rounds of competition.
However, it is also possible that the effective number of competitors would
gradually decrease over time mainly due to rational entry, even when a majority130
of voters do not vote strategically. More specifically, in an environment where
candidates rely on clientelist networks to secure votes, they should be more
likely to engage in rational entry because maintaining such networks is costly.
When the likelihood of winning a seat is slim, rational parties and candidates
should choose to withdraw from the competition to avoid wasting resources. The135
evidence of rational entry should appear post hoc when the size of clientelist
networks adjusts to reflect the share of votes the winner received in the last
election. The winner’s vote share, therefore, provides information about the
likelihood of winning to the potential challengers in the next election. Thus,
the initial fragmentation of the party system might be high due to the lack140
of strategic voting, but the number of contestants should decrease with each
election cycle (Cox, 1997; 173-178).
On the other hand, if parties can secure stable support from different societal
groups, it indicates that a party system is relatively institutionalized. Since
parties now have social roots, we would expect that electoral volatility tends to145
be low as different societal groups keep supporting the parties that represent
their interests (Mainwaring, 1999; Mainwaring & Zoco, 2007). When parties can
structure votes and represent different preferences of voters, the possible winners
will be common knowledge for both voters and contenders (Cox, 1997; 77). That
is, voters are able to have ranked preferences for parties according to the interests150
they represent. Voters’ preferences, then, provide the information about the
competitiveness of parties in the contests. In this context, voters can derive the
information about parties’ competitiveness from the socioeconomic construction
6
of their districts and engage in strategic behavior (Kawai & Watanabe, 2013).
As a result, strategic voting should become more important than rational entry155
to affect fragmentation in an institutionalized party system.
The discussion above indicates that an institutionalized party system is the
precondition for electoral systems to trigger strategic voting. Different electoral
systems, district magnitudes in particular, put varying degrees of pressure on
voters to vote strategically. In a nutshell, the smaller the district magnitude is,160
the higher the pressure. Thus, a single-member district vote system (SMD) puts
the highest pressure on voters and the pressure is gradually eased as the mag-
nitude gets larger. This rule is applicable to most electoral systems nowadays
since voters are generally given one ballot. However, when voters have multiple
ballots at hand, things become more complicated.165
Thailand used two electoral systems between 1986 and 2011. Approval vot-
ing (AV) was used between 1986 and 1996, and in 2007, while mixed-member
majoritarian system (MMM) was used in 2001, 2005, and 2011. AV is an elec-
toral system that gives voters multiple ballots to select candidates in multi-seat
districts.3 Magnitude under AV elections ranged from one to three but most of170
them elected three candidates. Voters were not required to cast all ballots but
cumulative vote was not allowed. In addition, voters could split the ballots to
candidates of different parties or cast a bloc vote for all candidates representing
the same party. Parties, then, had to nominate the number of candidates equal
to the district magnitude. On the other hand, MMM has two tiers. The district175
tier uses plurality rule to elect candidates in single-member districts (called the
SMD tier), while another tier uses the proportional representation (PR) vote
system to select parties (called the PR tier). Voters have two separate ballots
to cast in each tier.
3There are three features defining the electoral system (Colomer, 2007; 263). First, thedistrict magnitude is greater than one. Second, the number of ballots each voter has is smallerthan or equal to the district magnitude. Finally, plurality or majority rule might apply. Thereare several names for the electoral system, such as multiple non-transferable vote (MNTV),block vote, or plurality-at-large voting. Here the electoral system is referred to as approvalvoting following Cox (1984).
7
Voting behavior under SMD has been scrutinized in many studies. Gener-180
ally speaking, voters are likely to engage in strategic voting under SMD due to
the strong incentive embedded in the electoral rule. Nonetheless, many studies
found that other factors, such as intense competition, recognition of wasted-
vote situation, partisanship, preferences for specific candidates, significance of
party label for vote choice, and the level of party system institutionalization185
also played a role in reducing or discouraging strategic voting (Alvarez & Na-
gler, 2000; Blais & Nadeau, 1996; Blais et al., 2001; Duch & Palmer, 2002; Karp
et al., 2002; Kawai & Watanabe, 2013; Moser & Scheiner, 2012; Myatt, 2007).
By contrast, voting behavior under AV is relatively understudied, especially re-
garding strategic voting. Therefore, it is necessary to explain the voting pattern190
under AV to which now I turn.
2.1. Strategic Voting under AV
If we only focus on district magnitude and the information load needed to
discern the possible outcomes of strategic voting, it might not be difficult for
strategic voting to take place under AV when the district magnitude is smaller195
than five (Cox & Shugart, 1996). However, AV’s multiple ballots make the effect
of district magnitude less certain. Since voters have multiple ballots at hand,
they can cast the ballots with different combinations, ranging from outright
ticket splitting to bloc votes. Depending on how voters vote, it might pose
varying degrees of difficulty for strategic voting to take place.200
I will argue that different voting patterns under AV can either increase the
information load for voters to discern the likely effect of strategic voting or
make the decision-making process simpler like that under SMD. If a majority
of voters tend to split tickets for candidates of different parties, it is difficult
to guess who the front-runners would be simply because of the large number of205
competitors in the race. Suppose that an AV district elects three candidates,
the information load is still manageable since the district magnitude is smaller
than five. But parties tend to (or they are required to) nominate the number
of candidates equal to the number of ballots (which is also equal to district
8
magnitude in most cases) in hopes of securing all ballots from the same voter210
(Carey & Shugart, 1995; Colomer, 2007).
However, voters do not necessarily cast bloc votes even though they are given
the options as blocs/parties. When voters lack strict dichotomous preferences
for these blocs or parties, ticket splitting is likely to happen (Cox, 1984). In
other words, a voter who is indifferent about the blocs/parties might split her215
tickets for candidates of different parties. If she wants to to cast a strategic
vote, she would have to know the viability of these candidates. Suppose that
there are five blocs/parties participating in the contest, there are possibly fifteen
candidates, which requires much information to ascertain the viability of these
candidates and the likely outcome of strategic voting. I suppose that in a large220
electorate, a voter might just vote for her most preferred candidates until she
exhausts her ballots. This is probably why AV is hailed as strategically proven
and is said to always elect the Condorcet winner if it exists (Brams & Fishburn,
2005; Laslier, 2009).4
By contrast, if voters tend to vote for blocs/parties, the number of competi-225
tor is greatly reduced. As voters see the competition as between blocs/parties
and cast all ballots for one bloc/party, it should be easier to ascertain the viabil-
ity of these blocs. Despite its large district magnitude, an AV electoral game is
like a grouped SMD game in this context, which should provide a strong incen-
tive for strategic voting. However, to direct voters to vote in bloc, blocs/parties230
will need to distinguish themselves from other blocs/parties so voters can de-
velop ranked preferences. I will argue that this is more likely to happen in an
institutionalized party system in which parties represent the interests of different
societal groups and voters rank their preferences for parties accordingly.
In sum, an institutionalized party system provides the precondition for re-235
strictive electoral rules to induce strategic voting. When parties cultivate sup-
4Although Thailand’s AV is slightly different from the AV these authors discussed sinceThai voters were give a certain number of ballots to cast for candidates, not signaling theirapproval for each candidate, I will argue that this is still the theoretical expectation forThailand’s AV.
9
port from a segment of the society and that party labels convey the information
about the interests being represented, competition between parties is transpar-
ent and open to observation. In this context, the competitiveness of parties can
be ascertained at least based on the socioeconomic construction of a district.240
The likely effects of strategic voting, therefore, are more predictable and strate-
gic voting is likely to take place. Moreover, voters under AV electoral system
also tend to bloc vote for parties in this context. This voting behavior makes an
AV game similar to a SMD game, which should tend to induce strategic voting.
By contrast, if parties cannot structure votes and voters tend to vote for candi-245
dates, voters should have difficulty engaging in strategic behavior because the
information load for making a strategic decision is heavy and difficult to come
by. However, if parties and candidates rely on clientelist networks to secure
votes, rational entry should lead us to see a decrease in fragmentation at the
district level over time.250
3. The Thai Context and Hypothesis
Before the rise of the Thai Rak Thai Party, major parties in Thailand relied
on individual candidates and party factions to win seats instead of attracting
voters with policies (Ockey, 1994; 2003). It is not uncommon that party leaders
“bought” strong candidates in order to win state power (Connors, 1999). Since255
party label was insignificant for vote choice and frequent party switching and
the short life spans of parties testified that established parties did not possess
advantage in the electoral arena, strategic entry is unlikely the main mechanism
determining the level of party system fragmentation. Instead, it would be more
likely that rational entry is the main mechanism determining the fragmentation260
level.
This is because party factions and candidates relied on personal clientelist
networks to secure votes. Given that they needed to invest financial resources
in upholding a viable network, a winner obtaining a high vote share conveyed
information to future contenders about the size of her electoral assets. This265
10
information signals the cost of participating in the election and the likelihood of
winning. Since factions/candidates are office seekers, they should be sensitive
about the cost and withdraw from competition when the likelihood of winning is
small. When the number of candidates entering the district contests decreases,
the effective number of competitors should decrease as well.270
On the other hand, since candidates relied on their clientelist networks to
distribute bribes to voters, this practice of vote buying deprived voters free will
when they went to the polls. One assumption of electoral system effects is voters
should have the freedom to cast their ballots (Cox, 1997; 77), and when their
votes are bought beforehand, it is less likely that they would vote strategically275
by pouring votes to another candidate. The fact that voters accepted bribes
and cast their ballots accordingly also indicates that they did not have strict
preferences for candidates/parties. Although it was found that voters took
bribes but voted for other candidates, they simply cast sincere votes for their
most preferred ones (Hicken, 2009; 103). In both cases, voters did not have the280
intention to change the election outcome, which is another assumption necessary
for strategic voting to take place.5
By contrast, urban voters and the southerners (both were the constituents
of the Democrat Party) tended to cast bloc votes (Hicken, 2009; 2013), which
indicates that they had ranked preferences for parties. However, these voters285
were in minority among the electorate before 2001. Moreover, the Democrat
Party faced challenge from several new parties in urban areas since 1992, such
as the Phalang Tham Party (Moral Force). When urban voters faced several
options that they liked equally, they also tended to split tickets for different
parties.6 The situation is similar to what Cox (1984) discussed in Britain’s290
5Voters should have a myopic goal. That is, they only care about the current electionoutcome. See Cox (1997; 76-79)
6In Bangkok, for instance, 8 of 12 districts elected only one party in the September electionin 1992. The number of one-party districts decreased to be 4 of 13 districts in the 1995 election.In the 1988 election, there were only 5 districts among 13 elected a single party. Nevertheless,one-party districts were between 7 and 10 of 12 (sometimes 13) districts in most elections,except 1988 and 1995.
11
multi-member district elections, in which liberal voters split tickets for parties
holding similar ideologies. Although Cox used “strategic voting” to describe
the voting behavior, I will argue that it is actually sincere voting. Since the
behavior does not involve defection from a voter’s preferred choice (the voter
likes the parties equally), it does not conform with the definition of strategic295
voting used in this article. Note that I do not argue that voters do not vote
strategically when they split tickets; it is possible that they can vote sincerely
and strategically by splitting their tickets. However, to split tickets with the
intention to change the election outcome, a voter has to heed which candidate
in her less preferred party she should vote for to achieve that goal. A wrong300
decision would give her the lowest payoff when her most and less preferred parties
do not win any seats. Ticket splitting is risky especially when the supporters
of her least preferred party bloc vote. Since the information to discern the
outcome of strategic voting is difficult to come by and the likely effect of such
action is less predictable, strategic voting is unlikely the main mechanism that305
brings down the effective number of competitors when voters tend to split tickets
before 2001.
However, the Thai party system began to institutionalize since 2001. As
Figure 1 illustrates, the two parties that enjoy stable support after 2001 are the
TRT and the Democrat Party (DP). The TRT mainly attracts the rural and310
urban poor, while the DP appeals to the urban middle class and the southerners
(Kitirianglarp & Hewison, 2009; Thabchumpon & Duncan, 2011). The conver-
gence between societal groups and party support provides information about
the distribution of voter preferences in a district based on the socioeconomic
characteristics of the district. Thus, even with more parties entering the con-315
tests, voters can still detect possible winners and pour votes to those parties.
In other words, I would expect that strategic voting should be the main cause
of a reduction in the number of competitors at the district level after 2001. The
testable hypothesis is as follows:
Hypothesis: Rational entry poses relative importance to determine the effective320
12
number of competitors before 2001, but strategic voting becomes the main cause
after 2001.
4. Change in the Fragmentation of the Thai Party System
Before testing the hypothesis, we shall look at the change in party system
fragmentation over time. The fragmentation of the party system is shown in325
Figure 2. Following the literature in the field, I measured party system frag-
mentation by the effective number of parties (ENP).7 Although AV has large
district magnitude, the predicted number of viable parties can be either two or
four depending on whether the majority of voters cast bloc votes or split tickets.
If the majority of voters cast bloc votes, the number of viable parties should be330
close to two since AV becomes a grouped SMD game. If the majority of voters
split tickets, the number of viable parties should be four since the majority of
districts elect three winners.
(Figure 2 here)
As discussed earlier, voters are likely to split their tickets when their votes335
are bought and when they do not have ranked preferences for parties. The
average ENPs at the district level echo this argument and are between two and
four in the period between 1986 and 1996. Moreover, the overall trend is that
party system fragmentation is in decline until the 1996 election. I will argue that
the downward trend is the result of rational entry where strong winners drive340
away potential contenders in the next election. With fewer and fewer parties
participating in elections, the average ENP decreased over time.
The fragmentation level rose in 2001 and then fell in 2005. It increased
again in 2007 and dropped in 2011. The 2001 election adopted SMD at the
district-tier, which was new to parties and voters. Thus, it is likely that the345
result is new electoral system shock as elites and voters learned to coordinate
7ENP was constructed by Laakso & Taagepera (1979), which is weighted party vote shares.The indicator is calculated by the formula ENP = 1∑n
i=1 v2i
, where vi is party i’s vote share.
13
under the new electoral rule. The 2007 election re-adopted AV and it seems
that AV’s large magnitude encouraged entry and led to higher fragmentation
at the district level. The averaged ENP dropped again in 2011 simply because
SMD was re-installed in the district-level elections. However, I will argue that350
the increasing ENPs in 2001 and 2007 should be understood by the formula
calculating the effective number of parties. The ENP index is sensitive to two
factors – the vote share of the first winning party and the number of parties
in the contest (Dunleavy & Boucek, 2003). When the vote share is more than
a majority or there are a small number of contenders, the ENP tends to be355
small. However, if the first-place winner has only a plurality and there are
many contenders, the ENP tends to inflate.
Comparing the 2001 election with the 1996, we can see these effects. The
average number of parties entering the district-level contests in 2001 was 6.85,
compared to 5.68 in 1996, and the average vote share of the first-place winning360
party was 0.51 in 2001, compared to 0.57 in 1996. It is not surprising that
the 2001 election has a higher average ENP than that of 1996. However, the
increase in the number of entrants has nothing to do with parties trying to
gain representation. There were other reasons for the burgeoning of entrants
in 2001. Since 2001, parties have received state subsidies for the purpose of365
strengthening their role as intermediaries between citizens and the state; this
policy encourages the formation of small parties. To qualify for the subsidy,
parties are required to participate in elections.8 Thus, small parties entered the
contests with no intention to win seats but they nonetheless inflated the ENPs
across districts.370
The increase in the average ENP in 2007 comes from the same source. Many
former TRT members were encouraged by the military to leave the TRT and
run against the TRT’s successor party ahead of the 2007 election (Ockey, 2007).
Thus, the increase in ENPs is not due to AV’s large magnitude, but the military’s
intervention. Nevertheless, the election results proved that the intervention was375
8See Organic Law on Political Parties, mandated by the 1997 Constitution.
14
futile. The People’s Power Party, the successor party to the TRT, still won
nearly half of the district seats in the 2007 election (233 of 400 district seats).
The results, therefore, indicate that most parties were not viable parties in the
2007 election and voters have poured votes to only a few parties.
As I argued earlier, when parties can structure votes and voters have devel-380
oped ranked preferences for parties; party label provides the crucial information
for voters to discern the prospect of winning for the party they support given
the distribution of voter preferences in a district. As voters are able to distin-
guish possible winners and hopeless losers through party labels, they direct their
votes to viable parties. Strategic voting is likely the main force constraining the385
effective number of competitors after 2001, even in the 2007 election that used
AV (although we do not see this by simply comparing ENPs). By contrast, it
seems likely that the gradual decrease in fragmentation before 2001 is related
to rational entry, when candidates withdraw from contests in districts where
winners have strong support. The next section tests this hypothesis.390
5. Data and Methods
To test the hypothesis, I used the election data from 1986 to 2011 at the dis-
trict level. However, since two variables have a one-election lag, I only analyzed
nine election results. To make sure that each district covered exactly the same
territory as in the previous election, information about districting was refereed395
to ensure that the same sub-districts were analyzed even when they were in
different electoral districts across elections. In other words, the unit of obser-
vation was sub-districts and they might belong to different electoral districts
across elections.9 The purpose of this is to investigate the effects of rational
entry and strategic voting based on the results of the previous election. Note400
that the districting information is only available since the 1996 election. For
the elections between 1986 and 1995, I assigned the sub-districts based on the
9These sub-districts are khet in Bangkok and amphoe in other provinces.
15
districting information of the 1996 election.
More specifically, sub-districts of the electoral district 1 in 1996 were as-
sumed to belong to the same electoral district (District 1) in the previous elec-405
tions. Since the 1996 election had more electoral districts than the previous
elections, it is less risky that a sub-district would be wrongly assigned. That
is, a province might be divided into two electoral districts in 1996 due to the
increase in population, but the two electoral districts were one district in 1995
and the elections before it. Thus, it is less troublesome to assign sub-districts to410
an old electoral district than to a new one when sub-districts were re-assigned to
different electoral districts. The latter happens when the electoral system was
changed between AV and SMD. But these elections have districting information
that I can refer to. Electoral data between 1986 and 1992 are from Kollman
et al. (2010), and I collected the electoral data between 1995 and 2011 from De-415
partment of Administration (1995; 1996) and Office of the Election Commission
of Thailand (2001; 2005; 2007; 2011).
My dependent variable was the level of party system fragmentation at dis-
tricts. Although in the previous section, I used ENP to measure party system
fragmentation, here, to investigate the system fragmentation under AV, I in-420
stead used the effective number of candidates (ENC) as my dependent variable.
I calculated ENC with the same formula used to calculate ENP, but using can-
didates’ vote shares instead of parties’ vote shares. The original argument made
by Cox (1997) is that electoral rules affect the number of candidates entering
the contests at the district level. In a one-seat district, every party only gets425
to nominate one candidate, so the votes a candidate receives are her party’s
receiving votes. Thus, the ENC is equivalent to the ENP. In an electoral system
that has multi-seat districts (e.g. AV, open list PR, SNTV, and STV, etc.),
each party is likely to nominate multiple candidates. In Thailand’s AV, for in-
stance, each party was required to nominate a number of candidates equal to430
the district magnitude. When there are six parties competing in a three-seat
district, the number of contenders is eighteen. With such a large number, the
ENC is always greater than the ENP. Although what we concern is party system
16
fragmentation at the district level, it is more accurate to use ENC instead of
ENP in this analysis because voters cast ballots for individual candidates not435
parties.
Under AV, voters can cast their ballots in different combinations, ranging
from bloc votes to ticket splitting. Using ENP as the dependent variable would
have required an assumption that voters cast bloc votes and would have ignored
the dynamics of ticket splitting. Using ENC, on the other hand, allowed me440
to capture both. For instance, consider the scenario where party A has one
strong candidate that wins first place while the party’s two other candidates lag
behind. In total, party A’s vote share is high because of the strong candidate
(which is the only one who gets elected). This will result in a small ENP. ENC,
however, will be a large number, reflecting the poor performance of the other445
two candidates. On the other hand, if all three candidates of party A win seats,
both ENP and ENC will be small. In short, ENC captures different voting
combinations and reflects the level of fragmentation better than ENP under the
electoral rule of AV. For the elections using SMD, ENC and ENP are equivalent
since each party only nominates one candidate per district.450
Rational entry and strategic voting are the two explanatory variables. Ra-
tional entry is measured by the vote share of the winner in the previous election.
The rationale is that if the winner’s vote share is high in the previous election,
it indicates the size of the clientelist network of the winner and it should deter
potential contenders from entering in the next election. When the number of455
candidates decreases in the next election, ENC should tend to be smaller. In
other words, I expect to see a negative effect of rational entry on ENC. Note
that in the elections that use AV, rational entry is defined by the last-place
winner’s vote share. That is, in a three-seat district, rational entry is measured
by the third-place winner’s vote share.460
Strategic voting is measured through the SF-ratio provided by Cox (1997).
The SF-ratio is the ratio of the votes between the second-place loser and the
first-place loser. For instance, in a three-seat district, the first-place loser is
the fourth candidate ranked by the receiving votes in descending order. The
17
rationale is that if there is strategic voting, we shall see voters pour votes to465
the first-place loser in order to make him/her win the last slot. When this
happens, votes are relatively concentrated in the front-runners, thus creating a
lower ENC. The SF-ratio is between 0 and 1, with 1 indicating no vote pooling
to viable contenders. I calculated the SF-ratio under AV using candidates’
receiving votes instead of parties’ receiving votes. Since the dependent variable470
is ENC in the elections using AV, SF-ratio is calculated based on the same unit
(candidates). For the elections using SMD, a candidate’s receiving votes are
his/her party’s receiving votes.
The analyses controlled for four variables in both electoral systems. The first
control variable is the number of parties participating in the district election.475
Although most analyses controlled for district magnitude as it directly related
to the number of candidates in the contest, I controlled for the number of
parties instead due to a Thai regulation. The regulation required parties to
field candidates in up to at least one-fourth to one-half of the total number of
parliamentary seats, depending on the election year (Hicken, 2009; 96, footnote480
20). That is, even though parties might like to field candidates in only a few
districts where they had a good chance of winning seats, they had to participate
in other districts to comply with the regulation. This regulation led to more
parties than necessary entering races in districts with small magnitudes since it
was easier there than in larger districts to run the required number of candidates485
(which, as stated earlier, must be equal to the number of seats in contention).10
Moreover, state subsidies to political parties mentioned earlier also encouraged
entrants based on a reason irrelevant to the concern over election results. Thus,
the number of competing parties must be controlled for, since it also affects
the dependent variable. It is reasonable to omit district magnitude because the490
effect of the number of parties overlaps with that of district magnitude to some
extent. Omitting district magnitude also makes the statistical results under
10For instance, the average ENP of two-seat districts is 3.10 across all AV elections. Theaverage ENP of three-seat districts is only slightly higher, to be 3.29.
18
different electoral systems comparable.
The second control variable is the margin of victory in the last election. The
margin of victory is the vote gap between the last winner and the first loser.495
That is, in a three-seat district under AV, the margin of victory is the vote
difference between the candidate who won the last seat and the candidate who
was right behind, divided by the former’s votes. Thus, a large value means
the last winner garners far more votes than the first loser. By contrast, a
small value indicates that the competition between the last winner and the first500
loser is pretty tight. The purpose is to investigate the effect of strategic voting
with different conditions. In a district with a large margin in the last election,
voters might think that strategic voting is unlikely to change the outcome of the
current election given the overwhelming support the winner garnered last time.
Even though voters pour votes to the first loser, the first loser would still lose.505
Since defection from the most preferred choice would not change the outcome,
the incentive for strategic voting is reduced in those districts. By contrast, the
effect of strategic voting should be more evident in districts with a small margin.
Contests in these districts are competitive, which is thought a reason to trigger
strategic voting (Moser & Scheiner, 2012; 153-154).510
The index of margin of victory is between 0 and 1, and I recoded the index
into a dichotomous variable, with 1 meaning the index greater than 0.2 and 0
otherwise. Following Cox (1997), we shall see that the effect of strategic voting
is more salient in districts with a small margin than in districts with a large
margin. The third control variable is, therefore, the interaction term between515
strategic voting and the margin of victory in the last election. Finally, I also
controlled for the region of South. The South is the stronghold of the Democrat
Party and, thus, the party tended to enjoy stable support even before 2001
(Askew, 2006). Moreover, voters there tended to bloc vote for the Democrat
Party, resulting in lower level of party system fragmentation in the region.520
The purpose of the analysis is to inspect the relative importance of rational
entry and strategic voting to determine ENC at different times. Thus, in order
to compare the relative importance of these two variables across elections, I plot
19
the absolute values of the estimated coefficients and inspect their changes over
time. I expect to see the values of rational entry are larger than the values of525
strategic voting before 2001, and the values of strategic voting become larger
by contrast afterwards. Since the dependent variable, ENC, is continuous, all
the models are estimated using OLS.
6. Findings and Discussion
Table 1 shows the results. Winner’s vote share in the last election has a530
negative effect on ENC. That is, the higher the vote share the winner garnered
in the last election, the lower the ENC was observed in the current election,
other things being equal. By contrast, SF-ratio has a positive effect, meaning
that the larger the ratio (no vote pooling to the front-runners), the higher the
ENC was the result. Although I argued that strategic voting should have small535
effects before 2001, it does not mean it should not have any effect at all. Urban
voters were found to be more likely to cast bloc votes, which indicates that they
were capable of ranking parties. This means voters were also capable of strategic
voting although they were in a small number among the electorate. Note that
in the districts where the margin of victory in the previous election was greater540
than 0.2, the coefficient of the SF-ratio tended to be smaller, although most of
them are not statistically significant (see the interaction term of SF-ratio and
the margin of victory). The results indicate that a close race in the last election
did tend to intensify the impact of the SF-ratio on the current election. The
number of parties, as expected, has a positive effect on the dependent variable,545
except in the 2011 election. The South tended to have lower ENC on average
compared to other regions, although in some elections the region had higher
ENC on average.
(Table 1 about here)
The purpose of this article, however, is to inspect the importance of ratio-550
nal entry and strategic voting in different contexts. Specially, I am interested
20
in knowing the impacts these two actions have on district-level fragmentation
before and after 2001, when the party system began to institutionalize. From
the table, it is clear that winner’ vote share in the last election constantly had a
larger coefficient than that of the SF-ratio to determine ENC before 2001. The555
situation, however, is reversed since 2001 as the SF-ratio had a larger coefficient
than winner’s vote share to affect the dependent variable. To better illustrate
the relationship, I plot the absolute values of the coefficients across elections
and scrutinize the changes in relative importance of these two variables over
time.560
Figure 3 illustrates the changes of rational entry and strategic voting on
both types of districts. As expected, the values of rational entry are constantly
larger than the values of strategic voting before 2001. However, the coefficients
of strategic voting are larger than the coefficients of rational entry in both types
of districts since 2001, except the 2005 election. In 2005, the TRT was highly565
popular among the electorate and the party won 308 of 400 district seats. As
the party was expected to win the election handily, the competition was not
close enough for voters to think that strategic voting would make a difference.
This is probably why the effect of strategic voting was not salient in this election
like the other ones’.570
(Figure 3 here)
The graph clearly demonstrates that both actions assert different levels of
importance to determine the district-level party system before and after 2001.
The continuing decrease in system fragmentation at the district level before
2001 was mainly due to rational entry. The effect of rational entry was not575
statistically significant in the 2001 election perhaps due to re-districting under
different electoral rules. Nevertheless, the coefficients of rational entry became
much smaller compared to those before 2001. With regard to strategic voting,
this variable has asserted more influence in determining ENC since 2001. Strate-
gic voting is based on the premise that voters have ranked preferences for parties.580
However, such preferences mainly developed after 2001 due to the TRT’s social
21
policies. Thus, strategic voting’s larger effect in the 2001 election was a surprise.
This result might have been due to the electoral rules of SMD, which can trigger
strategic voting more easily than AV. That is, given the one-ballot structure and
a small district magnitude, it is easier to distinguish front-runners and losers585
under SMD than it is under AV. For the 2001 election, the electoral system
might have been the reason for the significant effect of strategic voting. How-
ever, the 2007 election used AV and strategic voting was still the main cause in
determining the level of fragmentation. Therefore, electoral system alone can-
not explain why strategic voting became the main determinant after 2001. As590
argued in this article, change in party system institutionalization accounts for
the outcome.
Nevertheless, to rule out the effect of electoral systems on the effective num-
ber of candidates and to take advantage of the Thailand case, I ran another
analysis comparing the effects of rational entry and strategic voting across all595
AV elections. The 2007 election re-adopted AV but the context was very dif-
ferent from the previous AV elections. I ran three models with different control
variables and compared the effects of strategic actions before and after 2001. I
omitted district magnitude in the previous analysis in order to compare the re-
sults under different electoral systems. Here, I put the district magnitude back600
to the model and Table 2 shows the results. Since I am interested in knowing
the change in the effects of the explanatory variables across time, I plot the
results in Figure 4.
(Table 3 and Figure 4 here)
It is clear that the pattern I found in the previous analysis re-appears in this605
analysis, except Model 2. Model 2 controlled for district magnitude and the
effect of rational entry is larger than strategic voting in the 2007 election, which
defies my argument. However, as I mentioned earlier, a lot of nonviable parties
participating in the contests because of two regulations; therefore, the number of
parties should also be controlled for. After adding the number of parties, we see610
the usual pattern in Model 3. That is, rational entry asserts more importance
22
than rational entry to determine the district-level party system fragmentation
before 2001, but the effects of these two strategic actions are reversed afterwards.
Note that the interpretation of the effects of strategic voting is different in this
analysis compared to the previous one. The models here emphasize effects at615
different times. Although the effect of strategic voting is larger in districts
with a large margin of victory than in districts with a small margin in the 2007
election in all three models, it does not mean that the way the margin of victory
conditions strategic voting is reversed in 2007. The correct interpretation is that
the effect of strategic voting in districts with a large margin is greater than the620
effect with the same condition in the pre-2001 elections. Given that politics
has been polarized after the 2006 military coup and voters were concerned with
the election outcome, it is not surprising to see the greater impact of strategic
voting on the dependent variable in 2007, even in districts where the incentive
is low.625
Taken together, it is clear that party system fragmentation at the district
level is determined by different strategic actions at different times. Between
1988 and 1996, elites’ rational choice is the main mechanism responsible for the
downward trend. After 2001, voters’ strategic action kept the fragmentation
level low even when AV was re-adopted and new parties entered the contests630
with the attempt at inflating the number of parties in the system. Rational entry
and strategic voting asserts different levels of importance to determine system
fragmentation due to the change in party system institutionalization in 2001.
The extant literature tends to emphasize the role of strategic voting in bringing
down fragmentation over time, but this article shows that rational entry is also635
an important factor leading to the same outcome in the environment of poorly
institutionalized party system. Although this article did not inspect the effect of
rational entry under SMD in the context of poorly institutionalized party system
since no such observation was available in the Thailand case, I speculate that
rational entry should be responsible for bringing down the number of parties640
and probably even more so in the same context. Because there is only one seat
to be won under SMD, candidates who own strong clientelist networks would
23
become apparent in just a few elections. Therefore, it is probably more likely
that the effect of rational entry is stronger under SMD. Nevertheless, future
research should examine this speculation.645
7. Conclusion
This article argues that the fragmentation of the district party system is de-
termined by different mechanisms in different contexts. Testing the hypothesis
on Thailand, the statistical analyses confirmed that before 2001 rational en-
try asserted relative importance to determine the level of fragmentation. After650
2001, strategic voting took the leading role in determining the effective num-
ber of competitors. The different effects of strategic behavior by candidates
and parties versus that of voters were closely related to the political context
at different times. When parties are the coalitions of candidates who rely on
clientelist networks to secure votes, rational calculation of the costs and the like-655
lihood of winning determines the number of the contenders in the next election,
which in turn determines the fragmentation level in districts. By contrast, when
voters cast ballots for parties according to the interests parties represent, party
label signals the information about viable contenders. In this context, strate-
gic behavior from voters became the main mechanism determining the level of660
fragmentation. This article also argues that in such a context, an AV electoral
game is similar to a SMD game and both tend to induce strategic voting that
constrains party system fragmentation.
Compared with strategic voting, elite decision is relatively understudied in
the field of electoral system effects. The Thailand case demonstrates that elite665
action helped bring down party system fragmentation even when the role of vot-
ers was marginalized. In other words, a poorly institutionalized party system
is not necessarily associated with high party system fragmentation in the long
run. As I argue in this article, rational entry should be an important factor
determining party system fragmentation over time. It is because elites who rely670
on clientelism to secure votes are rational calculators and would only enter the
24
contests when there is a good chance of winning. Although financial consider-
ation is a common factor affecting entry decision, future work should explore
other factors that lead elites to stay in or withdraw from competition. Given
that elites play an important role in less institutionalized party systems, they675
deserve much attention in such an environment when studying the causes of
party system fragmentation.
25
Appendix
Summary Statistics of Variables
680
Election 1988Minimum Mean Maximum S.D. No. of Obs.
ENC 2.228 7.257 16.78 3.069 928Winner’s vote share in 1986 0.083 0.158 0.631 0.087 928Sf-ratio 0.047 0.684 0.997 0.259 928Margin of victory < .2 in 1986 0 0.305 1 0.46 928No. of parties 3 9.985 15 2.672 928South 0 0.164 1 0.371 928
Election 1992.1Minimum Mean Maximum S.D. No. of Obs.
ENC 1.891 7.059 15.11 2.581 928Winner’s vote share in 1988 0.086 0.197 0.603 0.094 928Sf-ratio 0.037 0.775 0.999 0.228 920Margin of victory < .2 in 1988 0 0.361 1 0.481 928No. of parties 2 7.417 14 2.604 928South 0 0.164 1 0.371 928
Election 1992.2Minimum Mean Maximum S.D. No. of Obs.
ENC 1.735 6.929 14.23 2.525 928Winner’s vote share in 1992.1 0.08 0.186 0.619 0.086 928Sf-ratio 0.041 0.774 0.997 0.211 928Margin of victory < .2 in 1992.1 0 0.335 1 0.472 928No. of parties 2 6.489 12 2.336 928South 0 0.164 1 0.371 928
Election 1995Minimum Mean Maximum S.D. No. of Obs.
ENC 1.829 5.639 11.65 2.093 928Winner’s vote share 1992.2 0.085 0.197 0.718 0.094 928Sf-ratio 0.028 0.649 0.998 0.291 928Margin of victory < .2 in 1992.2 0 0.353 1 0.478 928No. of parties 2 5.669 11 2.17 928South 0 0.164 1 0.371 928
26
Election 1996Minimum Mean Maximum S.D. No. of Obs.
ENC 0.562 4.806 10.93 1.662 928Winner’s vote share in 1995 0.096 0.23 0.662 0.101 928Sf-ratio 0.025 0.607 0.995 0.305 917Margin of victory < .2 in 1995 0 0.417 1 0.493 928No. of parties 2 5.536 12 2.517 928South 0 0.164 1 0.371 928
Election 2001Minimum Mean Maximum S.D. No. of Obs.
ENC 1.114 2.782 5.963 0.843 928Winner’s vote share in 1996 0.099 0.257 0.887 0.115 928Sf-ratio 0.014 0.475 0.999 0.307 918Margin of victory < .2 in 1996 0 0.514 1 0.5 928No. of parties 2 6.739 12 1.923 928South 0 0.164 1 0.371 928
Election 2005Minimum Mean Maximum S.D. No. of Obs.
ENC 1.152 2.057 3.459 0.445 928Winner’s vote share in 2001 0.251 0.514 0.946 0.131 928Sf-ratio 0.003 0.311 0.998 0.272 853Margin of victory < .2 in 2001 0 0.764 1 0.424 928No. of parties 2 4.224 11 1.376 928South 0 0.164 1 0.371 928
Election 2007Minimum Mean Maximum S.D. No. of Obs.
ENC 2.063 6.816 12.47 2.134 928Winner’s vote share in 2005 0.334 0.627 0.93 0.119 928Sf-ratio 0.068 0.791 0.999 0.187 928Margin of victory < .2 in 2005 0 0.876 1 0.329 928No. of parties 4 9.666 14 2.156 928South 0 0.164 1 0.371 928
Election 2011Minimum Mean Maximum S.D. No. of Obs.
ENC 1.14 2.198 4.305 0.625 928Winner’s vote share in 2007 0.094 0.203 0.618 0.082 928Sf-ratio 0.004 0.346 0.999 0.302 916Margin of victory < .2 in 2007 0 0.483 1 0.5 928No. of parties 2 6.221 13 1.904 928South 0 0.164 1 0.371 928
27
References
Alvarez, R. M., & Nagler, J. (2000). A new approach for modelling strategic685
voting in multiparty elections. British Journal of Political Science, 30 , 57–75.
Amorim Neto, O., & Cox, G. W. (1997). Electoral institutions, cleavage struc-
tures, and the number of parties. American Journal of Political Science, 41 ,
149–174.
Askew, M. (2006). Culture and Electoral Politics in Southern Thailand. Election690
Campaigning, Group Formation and the Symbolic Construction of Political
Allegiances in Songkhla Province.. Bangkok, Thailand: King Prajadhipok’s
Institute.
Blais, A., & Nadeau, R. (1996). Measuring strategic voting: A two-step proce-
dure. Electoral Studies, 15 , 39–52.695
Blais, A., Nadeau, R., Gidengil, E., & Nevitte, N. (2001). Measuring strategic
voting in multiparty plurality elections. Electoral Studies, 20 , 343–352.
Brams, S., & Fishburn, P. C. (2005). Going from theory to pratice: The mixed
success of approval voting. Social Choice Welfare, 25 , 457–574.
Carey, J. M., & Shugart, M. S. (1995). Incentives to cultivate a personal vote:700
A rank ordering of electoral formulas. Electoral Studies, 14 , 417–440.
Chhibber, P., & Kollman, K. (2004). The Formation of National Party Systems:
Federalism and Party Competition in Canada, Great Britain, India, and the
United States. Princeton: Princeton University Press.
Clark, W., & Golder, M. (2006). Rehabilitating duverger’s theory. Comparative705
Political Studies, 39 , 679–708.
Clough, E. (2007). Strategic voting under conditions of uncertainty: A re-
evaluation of duverger’s law. British Journal of Political Science, 37 , 313–
332.
28
Colomer, J. M. (2007). On the origins of electoral systems and political parties:710
The role of elections in multi-member districts. Electoral Studies, 26 , 262–
273.
Connors, M. K. (1999). Political reform and the state in thailand. Journal of
Contemporary Asia, 29 , 202–226.
Cox, G. W. (1984). Strategic electoral choice in multi-member districts: Ap-715
proval voting in practice? American Journal of Political Science, 28 , 722–738.
Cox, G. W. (1997). Making Votes Count: Strategic Coordination in the World’s
Electoral Systems. New York: Cambridge University Press.
Cox, G. W., & Shugart, M. S. (1996). Strategic voting under proportional
representation. Journal of Law, Economics, and Organization, 12 , 299–324.720
Crisp, B. F., Olivella, S., & Potter, J. D. (2012). Electoral contexts that impede
voter coordination. Electoral Studies, 31 , 143–158.
Department of Administration (1995). Statistics and Results of the 1995 General
Election. Bangkok: Ministry of Interior.
Department of Administration (1996). Statistics and Results of the 1996 General725
Election. Bangkok: Ministry of Interior.
Duch, R. M., & Palmer, H. D. (2002). Strategic voting in post-communist
democracy? British Journal of Political Science, 32 , 63–92.
Dunleavy, P., & Boucek, F. (2003). Constructing the number of parties. Party
Politics, 9 , 291–315.730
Hicken, A. (2009). Building Party Systems in Developing Democracies. New
York: Cambridge University Press.
Hicken, A. (2013). Late to the party: The development of partisanship in
thailand. TRaNS: Trans-Regional and-National Studies of Southeast Asia, 1 ,
199–213.735
29
Hicken, A., & Stoll, H. (2011). Presidents and parties: How presidential elections
shape coordination in legislative elections. Comparative Political Studies, 44 ,
854–883.
Karp, J. A., Vowles, J., Banducci, S. A., & Donovan, T. (2002). Strategic voting,
party activity, and candidate effects: Testing explanations for split voting in740
new zealand’s new mixed system. Electoral Studies, 21 , 1–22.
Kawai, K., & Watanabe, Y. (2013). Inferring strategic voting. The American
Economic Review , 103 , 624–662.
Kitirianglarp, K., & Hewison, K. (2009). Social movements and political oppo-
sition in contemporary thailand. The Pacific Review , 22 , 451–477.745
Kollman, K., Hicken, A., Caramani, D., & Backer, D. (2010). Constituency-level
elections archive. URL: www.electiondataarchive.org.
Laakso, M., & Taagepera, R. (1979). Effective number of parties: A measure
with application to west europe. Comparative Political Studies, 12 , 3–27.
Laslier, J. (2009). The leader rule. Journal of Theoretical Politics, 21 , 113–136.750
Mainwaring, S. (1999). Rethinking Party Systems in the Third Wave of Democ-
ratization: The Case of Brazil . Stanford: Stanford University Press.
Mainwaring, S., & Zoco, E. (2007). Political sequences and the stabilization of
interparty competition. Party Politics, 13 , 155–178.
Moser, R., & Scheiner, E. (2012). Electoral Systems and Political Context: How755
the Effects of Rules Vary Across New and Established Democracies. New
York: Cambridge University Press.
Moser, R. G. (1999). Electoral systems and the number of parties in postcom-
munist states. World Politics, 51 , 359–384.
Myatt, D. P. (2007). On the theory of strategic voting. The Review of Economic760
Studies, 74 , 255–281.
30
Ockey, J. (1994). Political parties, factions, and corruption in thailand. Modern
Asian Studies, 28 , 251–277.
Ockey, J. (2003). Change and continuity in the thai political party system.
Asian Survey , 43 , 663–680.765
Ockey, J. (2007). Thailand in 2006: Retreat to military rule. Asian Survey , 47 ,
133–140.
Office of the Election Commission of Thailand (2001). Statistics and Results of
the 2001 General Election. Bangkok: Office of the Election Commission of
Thailand.770
Office of the Election Commission of Thailand (2005). Statistics and Results of
the 2005 General Election. Bangkok: Office of the Election Commission of
Thailand.
Office of the Election Commission of Thailand (2007). Statistics and Results of
the 2007 General Election. Bangkok: Office of the Election Commission of775
Thailand.
Office of the Election Commission of Thailand (2011). Statistics and Results of
the 2011 General Election. Bangkok: Office of the Election Commission of
Thailand.
Shugart, M. S., & Carey, J. M. (1992). Presidents and Assemblies: Constitu-780
tional Design and Electoral Dynamics. Cambridge University Press.
Singer, M. M., & Stephenson, L. B. (2009). The political context and duverger’s
theory: Evidence at the district level. Electoral Studies, 28 , 480–491.
Tavits, M. (2008). Party systems in the making: The emergence and success
of new parties in new democracies. British Journal of Political Science, 38 ,785
113–133.
Tavits, M., & Annus, T. (2006). Learning to make votes count: The role of
democratic experience. Electoral Studies, 25 , 72–90.
31
Thabchumpon, N., & Duncan, M. (2011). Urbanized villagers in the 2010 thai
redshirt protests. Asian Survey , 51 , 993–1018.790
32
Table 1. Determinants of ENC across Elections
1988 1992.1 1992.2 1995 1996(Intercept) .892* 4.467*** 3.631*** 3.213*** 5.027***
(.394) (.310) (.447) (.193) (.179)Winner’s vote share* -12.931*** -15.041*** -13.021*** -6.882*** -9.040***
(.774) (.687) (.889) (.481) (.450)SF-ratio (SF) 3.832*** 2.965*** 4.279*** 3.526*** 1.909***
(.286) (.281) (.371) (.179) (.135)Margin of victory > .2 (MV)** 1.004** .711* 2.226*** 1.069*** .418**
(.379) (.352) (.406) (.196) (.144)SF*MV -.043 -.202 -2.760*** -1.726*** -.831***
(.522) (.417) (.500) (.265) (.207)No. of parties .546*** .401*** .371*** .293*** .134***
(.027) (.020) (.025) (.020) (.015)South 1.959*** .386** -.690*** -1.060*** -.215*
(.185) (.136) (.143) (.118) (.105)
R2 0.6483 0.6954 0.6777 0.7033 0.6901No. of Obs. 928 920 928 928 917
800
2001 2005 2007 2011(Intercept) 1.277*** 2.535*** 1.886*** 2.160***
(.102) (.090) (.589) (.082)Winner’s vote share* -.275 -1.400*** -2.437*** -.807***
(.205) (.126) (.525) (.233)SF-ratio (SF) 1.185*** .330*** 4.149*** 1.261***
(.092) (.097) (.627) (.073)Margin of victory > .2 (MV)** -.194** -.024 .762 -.095
(.073) (.045) (.579) (.051)SF*MV .191 .164 -.264 -.494***
(.127) (.111) (.721) (.105)No. of parties .169*** .023* .305*** -.008
(.011) (.011) (.026) (.008)South -.429*** .182*** -1.552*** -.320***
(.059) (.038) (.153) (.047)
R2 0.5150 0.2711 0.386 0.4138No. of Obs. 918 853 928 916Note: * Last winner’s vote share in the last election** Margin of victory in the last electionSignificance level * < .05, ** < .01, *** < .001
35
Table 2. Determinants of ENC under AV Elections805
Model 1 Model 2 Model 3(Intercept) 3.596*** 1.881*** -1.103***
(.132) (.211) (.192)Winner’s vote share (WV) -11.249*** -7.675*** -4.972***
(.319) (.408) (.355)SF-ratio (SF) 3.529*** 3.767*** 3.374***
(.116) (.125) (.107)Margin of victory > .2 (MV) 1.105*** .510*** .761***
(.136) (.147) (.126)SF*MV -1.161*** -.909*** -1.071***
(.179) (.192) (.165)Election 2007 -1.918*** .767 1.257*
(.539) (.591) (.506)WV*Election 2007 8.823*** 4.256*** 1.929***
(.588) (.678) (.583)SF*Election 2007 .096 -.764 -1.871***
(.595) (.643) (.551)SF*MV*Election 2007 2.073** 1.135 3.211***
(.689) (.740) (.635)MV*Election 2007 -1.258* -.070 -2.016***
(.550) (.590) (.507)No. of parties .342*** – .347***
(.008) – (.007)District magnitude – 1.361*** 1.404***
– (.051) (.044)South -.119 -.788*** -.050
(.061) (.063) (.057)
R2 0.6102 0.5499 0.6696No. of Obs. 5549 5549 5549Significance level * < .05, ** < .01, *** < .001
37