Over-nominating candidates,undermining the party

Abstract

Any political party has a profound interest in maximizing seats, which in turn requires running the optimum number of candidates. However, to do this presumes solving a collective action problem among self-interested party members or leaders, and is deeply conditioned by the electoral system. The case of Japan’s Liberal Democratic Party under the Single Non-Transferable Vote electoral system provides a superb illustration of how party leaders, even in a famously electorally successful party, will be unable to solve these dilemmas because of key facilitating institutions: first, party president selection rules; second, prime ministerial control over allocation of positions; third, a weak party label. Contrary to existing literature, we find ambitious factions consistently nominated too many candidates – deliberately risking the party’s losing seats. We draw attention to the sources of party strength in a novel way, and to how party rules interact with electoral systems to shape both parties and politics.

Keywords

Electoral systems factionalism Japan Liberal Democratic Party

Introduction

The fundamental problem of acting together for collective benefit bedevils all political parties, whether in elections, policy-making, legislative organization, leadership selection or elsewhere. Probably the most basic level of cooperation as a party is to be found in the electoral arena, where the collective interest of the party demands running the number of candidates that will maximize won seats, ceteris paribus. Yet, in many venues and fora, party leaders confront potential or actual challenges from fellow party members. In a classic ‘problem of the commons’, factions or subgroups operating within the party may face contradictory incentives: a desire to win more seats for the party versus a stronger and potentially conflicting desire to win more seats for the faction itself. Although scholars have long known the importance of factions in selecting and supporting individual candidates, we have lacked a clear understanding of how and under what conditions the party is unable to coordinate the factions and winds up running the wrong number of candidates. Party leadership needs to coordinate factional leaders’ actions to optimize the number of winnable seats, while factions face their own incentives and may or may not cooperate to achieve the party’s collective goals. Who will coordinate and how will the coordination be achieved (if at all)? What institutional arrangements at the party organizational level increase the chances of achieving coordination?

We argue here that the electoral system profoundly affects intra-party factions’ strategies and thus political parties’ ability to solve this universal collective action dilemma. In this sense, electoral systems ‘make‘ parties, by structuring the conditions under which parties – which even under the loosest of definitions would include some form of electoral cooperation – cohere. Under most electoral systems, party leaders have sufficient resources to effectively enforce such cooperation. However, we show that under a single non-transferable voting (SNTV) system, even the most successful parties are unable to restrain the self-interested ambitions of faction leaders – at the expense of imperiling the party’s lifeblood: its ability to win seats in the legislature.¹ This SNTV system provides an optimal case to explore the aforementioned questions. First and foremost, the nature of the system forces a larger party to send multiple candidates to the same district in order to win a majority in the assembly. This, in turn, forces party leadership to correctly estimate the number of winners in a district and send the optimal number of candidates. Under-nominations leave votes on the table, while over-nominations could result in vote dispersion among candidates and therefore a party may get fewer seats than it could have won (Browne and Patterson, 1999; Cox and Niou, 1994; Swindle, 2002).

Political science usually operates with a strong assumption of a unitary actor in analysing how parties contest elections. However, we show that relaxing this assumption to study how parties solve the collective action problem, successfully or not, can improve our understanding of the parties and electoral results (Maeda, 2012). The Liberal Democratic Party (LDP) was notorious for its strong factions. Despite its long string of electoral success, the party leadership never completely overcame the problem of over-nomination under SNTV. We explain why the LDP could never solve this problem, show under what conditions over-nominations especially occurred and detail the specific obstacles a party may face in surmounting ‘the problem of the commons’ in such systems. Under the highly personalistic SNTV system, party label had become a relatively unimportant resource to candidates and factions so that the party leadership could not effectively constrain factional leaders’ selfish behaviour. This failure was abetted by a significant intervening factor at the party organizational level: the intra-party mechanism to determine the party president encouraged ambitious factions to make every effort to expand their size. The third piece of the puzzle is that gaining the presidency of the LDP effectively equated (due to near constant LDP majorities under SNTV) with ascending to the prime ministership, with exclusive power to allocate important governmental privileges and positions. The leadership of the LDP could not effectively restrain such factional misbehaviour, because the president of the LDP himself was the head of a faction that had the incentives to maintain or increase its size in preparation for the next presidential race.

The next section is a review of how Japan’s SNTV system worked and what the LDP’s nomination strategies looked like. In doing so, we show that, although these strategies may have helped produce a downward trend in the number of candidates, the LDP could not effectively overcome the over-nomination problem throughout the life of SNTV. In the third section we discuss the institutional settings that encouraged factions to ambitiously compete for office. The fourth section is our data analysis, followed by our conclusions.

Japan’s SNTV system

Japan's lower house used an SNTV system from 1947 to 1993. Almost all the districts in the approximately 500-seat lower house had magnitude of three to five, inclusive. The inability to pool votes has two important implications. First, in order for a party to win a majority it needs to send more than one candidate to all or most districts. Party leadership needs to correctly estimate the number of winnable seats in a district so that it can optimize the number of candidates. Sending more than the optimal number of candidates can lead to the worst outcome, as over-nominations could result in vote dispersion among candidates and therefore a party may get fewer seats than it could have won (tomodaore in Japanese) (Browne and Patterson, 1999; Cox and Niou, 1994; Swindle, 2002). Second, the fact that a party must send multiple candidates to win a majority means that an SNTV system reinforces intra-party fragmentation (Carey and Shugart, 1995). Each of the candidates even from the same party is in zero-sum competition, as votes cannot be pooled. Individual candidates have to differentiate themselves, by resorting to their own personal attributes, individual vote mobilization strategies (Tatebayashi, 2004) and/or intra-party factions, while party label becomes a relatively unreliable resource and cannot by itself guarantee victory: Thayer (1969: 120) estimates an LDP endorsement to be worth about 10,000 votes, compared to 79,806 votes that an average candidate would need to win a seat. The conventional wisdom of the literature on Japanese party politics implicitly or explicitly suggests that this factionalism was largely reinforced and stimulated by Japan’s SNTV system (Cox et al., 1999; Kohno, 1992; Ramseyer and Rosenbluth, 1993).²

Nevertheless, it is unclear how party leadership tried to achieve the collectively desired outcome under these circumstances. Given the intra-party fragmentation, achieving a collective action should have required extensive coordination among individual candidates as well as factions. In essence, there are three players at work: party headquarters, factions and candidates. The existing literature points to the LDP’s Policy Affairs Research Council and other organs to divide votes in the same district (McCubbins and Rosenbluth, 1995; Ramseyer and Rosenbluth, 1993); members’ strategies to regionally divide the district (Mizusaki and Mori, 2007); and the interactions of these (Tatebayashi, 2004). Other work (Cox and Rosenbluth, 1994, 1996) analyses nominations but only in terms of which kind of factions (mainstream) may gain an advantage over others (anti-mainstream) in endorsements. Although informative, these accounts are not sufficient for our purpose, because their primary focus is on how elected incumbents, not candidates, utilized the LDP’s institutions to coordinate vote-dividing strategies. The exception is Reed (1990, 2009), who examines the trial-and-error process among candidates. This is a promising approach, especially in analysis of change over time. However, unlike Reed, our analysis emphasizes the importance of factions as an intervening variable of party organization, and we explain the conditions that can lead to factions preventing the party from solving the collective action dilemma.

Others point to the LDP’s Electoral Strategy Committee (Senkyo taisaku iinkai) to forge ‘a party-wide agreement to reduce the costs of tomodaore by reducing the number of non-incumbents endorsed’ (Cox and Rosenbluth, 1994: 13), but the committee’s roles were inconclusive at best. As Cox and Rosenbluth (1994: 13) admit, such an agreement ‘was considerably tempered in practice by factional competition’. Indeed, their data suggest that over-nominations were almost constantly observed in more than 10 percent of the districts. By comparing Japan with Ireland’s STV, Swindle (2002: 288) argues that ‘the collective incentives of a majority-seeking party under SNTV compel party leadership to actively control the electoral environment through strict nomination control and sophisticated vote management strategies’. But his evidence only shows that the LDP was less likely to over-nominate candidates than Irish parties, while it does not explain the mechanism inside the LDP to over-nominate candidates in more than 30 percent of the cases. According to a new method suggested by Patterson (2009) to correctly estimate the LDP’s nomination errors, the LDP on average made errors in 12.0 percent of the districts from 1958 to 1990; and this figure would only be larger if we include factionally supported conservative independents.

In sum, the existing research suggests only that the LDP was not able to avoid the collective action problem under SNTV (Christensen, 2000). The LDP persistently supported more candidates than the number of seats it could have won with certainty throughout the SNTV period. The LDP could not completely solve its collective action dilemma despite powerful electoral incentives to do so. The next section explores why.

Argument: The presidential race, party presidential power and an unreliable party label

We have shown that the party leadership could not completely overcome the problem throughout the SNTV period. We point to three factors that made collective action difficult: the intra-party rule used to determine the party president; exclusive access to a wide range of office benefits given to the president of the LDP; and the weak and insufficient party label conditioned by the extremely personalistic SNTV system. Let us examine each in detail.

Presidential election

The LDP’s presidential election system has experienced several changes since 1955 (Jiyu Minshu Henshubu, 1995), but the bottom line is that Diet members were given significant positions in electing the party president during the SNTV period. From 1955 to 1976, the system was a relatively simple majority run-off, where only Diet members and two delegates from each of the 47 prefectural branches participated in the party caucus. Thus, given that the LDP consistently had more than 400 Diet members when SNTV was in place, victory in the presidential race crucially depended on Diet members’ votes. Factions made electoral alliances of convenience with the other factions before the race, but such coalitions were highly unstable (Leiserson, 1968). Factional leaders dished out huge sums of money for support from their members and other factions (Thayer, 1969). Aiichiro Fujiyama, a well-to-do businessman and one of the ambitious leaders, even drove his companies into bankruptcy. The size of a faction played a dominant role in its fortunes in the presidential race, thus encouraging ambitious factional leaders to expand ever more.

In 1978, the LDP abolished the majority run-off and instead introduced a primary (yobi senkyo) where ordinary party members could vote for one candidate, but party members were disenfranchised in 1989, with each prefectural branch instead casting one to four ballots in the first round. Moreover, there was the emergency clause stipulating that the party could skip the primary and simply hold a majority run-off election in Tokyo to elect the president. Often, factional leaders used this provision to exclude prefectural branches and instead engaged in behind-the-door negotiations to choose the next president. Furthermore, perversely, the primary came to drive the penetration of factional competition from the national level to the local level (Sato and Matsuzaki, 1986). Factional leaders simply mobilized koenkai, or the Diet members’ personal vote-mobilization organizational network, to expand the number of their supporters who registered as party members (Krauss and Pekkanen, 2010).

The party president’s power

Competition for the party presidency was intense and costly because being the president of the LDP always meant being the prime minister of Japan from 1955 to 1993. Diet members ardently campaigned for precious seats in the cabinet, exclusively controlled by the prime minister (Article 68 of the Japanese Constitution). A post in the cabinet translated to an increase of votes in the next election, new opportunities to bring benefits to districts and interest groups, augmented policy influence, new channels to raise funds and the chance to ascend the career promotion ladder all the way to prime minister (Thayer, 1969).

Because of the nature of the party presidential elections described above, the larger a faction, the more likely it was to be in a winning coalition, and therefore the bargaining power for posts. The size of a faction was thus important to not only win the prime minister’s office, but also distribute more rewards for loyal factional members. This further strengthened ambitious factional leaders’ incentives to run for the party president, as well as individual members’ incentives to get affiliated with factions. It is true that the LDP gradually institutionalized the rules so that factions should have fair access to posts: posts were more or less proportionally allocated to factions (habatsu kinko), and cabinets were frequently reshuffled (Kawato, 1996). But these rules continued to be only implicit and informal in the sense that some prime ministers did not follow them. Some of the prime ministers shuffled cabinets somewhat less frequently or distributed posts less strictly proportionally.³ This ability, combined with the prestige of the position, the ability to have more influence on policy than other faction leaders (Hayao, 1993), and the constant pressure from followers to expand the faction to give it more influence in negotiations over sub-cabinet posts and the prime minister’s faction being better able to attract new members, all made the prime minister a coveted position among faction leaders.

Nominating candidates

The third element that made collective action difficult to achieve was the comparatively weak and insufficient party label under the extremely personalistic SNTV system. Because the LDP’s party label was only worth about 12–13 percent of the votes required for a seat on average (Thayer, 1969) and candidates from the same party label had to differentiate themselves from the other co-partisans, they had to emphasize their own personal attributes, such as their abilities to bring selective benefits to local districts, or they had to create and maintain personal vote-mobilizing networks, koenkai, on their own.

Thus factions were critical in the elections. On the demand side, individual candidates had to resort to factions for campaigning and organizational support. Of all the 4,410 LDP candidates from 1958 to 1993, 4,064 (or 92.2 percent) had factional backing, while only 346 (or 7.8 percent) ran for election without any factional support. And of all the LDP 3,528 winners in the same period, only 170 (4.8 percent) had no factional support, suggesting that factional support was nearly a required condition for winning a seat. Typically, factional support included not only supportive campaign speeches delivered by factional leaders and/or celebrities and popular athletes who had connections with the leaders (Thayer, 1969), but also money to launch koenkai and finance campaigns (Asahi Shinbun Seijibu, 1968). On the supply side, on the other hand, factions were willing to support quality candidates, for the reason explained above: ambitious factions had the strong incentives to support as many candidates as possible to win the prestigious party presidency. As Thayer (1969) notes: ‘A faction leader preparing to make a bid for the prime minister’s job is often very eager to add more followers to his stable, and the more actively he searches out new candidates the less he can be concerned with ideological affinity’ (p. 48). So-called ‘conservative independents (hoshukei mushozoku)’ – what Reed (2009) calls ‘Liberal Democratic Independents’, or LDIs – exemplify the frail party label and vital factional support in the elections quite well. Because ‘failure to obtain a nomination is not fatal to a candidate’s chances’ (Thayer, 1969: 38), quite a few candidates ran as independents without any party label, but with support from a faction.

The leadership could not completely prevent factions from supporting candidates even when over-nomination threatened incumbents; even withholding the official LDP label was insufficient. The president of the LDP himself was the head of a faction that had the incentives to maintain or increase its size in preparation for the next presidential race. This ‘who guards the guardian’ problem made commitment to the Pareto-optimal cooperation unsustainable. All the leadership could do in order to control over-nomination was to coordinate official nominations through the party’s Election Strategy Committee (Senkyo Taisaku Iinkai) (Cox and Rosenbluth, 1994, 1996), and the committee indeed somewhat reduced the number of over-nominations over time, but imperfectly.

In sum, the LDP’s rules for choosing the party president encouraged ambitious factional leaders to expand ever more. The LDP party presidency inevitably brought the prime ministership and so was irresistible to ambitious politicians. In an attempt to expand, ambitious factions in the times of the elections supported as many candidates as possible both financially and organizationally, even as LDIs. Factions oftentimes ignored the leadership’s efforts to reduce candidates as the LDP president himself was the head of a faction who had a stake in expanding the size of his own faction.

Hypotheses, data and analysis

Hypotheses

Thus far we have seen that, because in the presidential race factional size mattered, the presidency was the key to exclusive office benefits, and the extremely personalistic SNTV system made it difficult for the party leadership to control over-nominations, a faction would provide financial and organizational support to candidates regardless of their nomination status; and this intra-party, inter-faction electoral competition would lead to over-nominations. In general, faction leaders have to balance the chance of gaining a seat for their faction versus the chance of over-nomination costing the party a seat. Any faction leader would accept a 100 percent chance of increase in faction without risk of the party losing a member. No faction leader would accept a zero chance of extra faction member win with a certain loss of the party seat. Each faction leader would have a trade-off point where the chance of an extra faction seat outweighs the chance of losing a party seat. However, when the faction leader is planning to run for party leader, the extra faction seats are even more valuable than usual (and, the loss of a party seat that is held by a rival faction is even less important than usual).

So, we hypothesize:

Hypothesis 1. If the LDP’s collective interest in securing a majority is stronger, then a faction is more likely to pursue mutual cooperation and less likely to support candidates.

Hypothesis 2. If a faction leader has the ambition to run in the next presidential race, then it is more likely to support candidates.

Hypothesis 3. If a faction leader has the ambition to run in the next presidential race, then it is more likely to support candidates in a district where rivals have incumbents.

Hypothesis 4. If a district contains non-incumbent candidates from ambitious factions, then the district is more likely to have an over-nomination error (tomodaore).

Data

We test these hypotheses with a dataset that covers all the candidates from the LDP and factionally supported independents between 1960 and 1993. We use Steven R. Reed’s data on factional affiliations and electoral results; we supplement these data with newspaper articles on nominations (Asahi and Yomiuri), political yearbooks (Kokkai Binran and Seiji Handobukku), which contain factional affiliations of LDP incumbents, new and non-incumbent LDP candidates, and even LDIs from 1969 on. For the pre-1969 data, we intensively consult various articles and monographs on factions (Leiserson, 1968; Sato and Matsuzaki, 1986; Thayer, 1969; Watanabe, 1958).

We assume the following equation for the first three hypotheses:

L o g i t (Y_{i, j, t}^{1}) = a_{0} + a_{1} X_{i, j, t} + a_{2} F_{i, t} + a_{3} D_{j, t} + e_{i, j, t},

where i refers to factions (i = 1, …, 36), j districts (j = 1,…, 130) and t elections (t = 1960, …, 1993). There were 8 to 9 factions on average, so there are about 13,000 observations. The dependent variable (Y¹ _i,j,t ) is a dichotomous dummy variable coded 1 if faction i supported a non-incumbent candidate⁴ to district j at election t, so logit models are used to test the hypotheses. X is faction i’s characteristics in district j at election t; F faction i’s characteristics at election t; and D a set of the characteristics of district j at t.⁵ The error term is represented by e.

To test the fourth hypothesis, we assume the following equation:

L o g i t (Y_{j, t}^{2}) = a_{0} + a_{1} X_{j, t}^{2} + a_{2} D_{j, t} + e_{j, t},

where the dependent variable (Y² _j,t ) is whether the LDP made the over-nomination error in district j at t and the independent variable of our interest in this equation (X² _j,t ) is the number of non-incumbent candidates from ambitious factions. There is the extensive literature on the over-nomination error (Browne and Kim, 2003; Browne and Patterson, 1999; Christensen and Johnson, 1995; Cox and Niou, 1994; Cox and Rosenbluth, 1994). Based on this literature, we define over-nomination error as the situation in which the conservative camp could have won one (or more) additional seat by hypothetically reducing the number of candidates. The estimation was conducted under two different assumptions: (1) the ‘invariable vote assumption’ that a candidate’s votes would be perfectly transferred to the other conservative candidates if he/she did not run (Browne and Patterson, 1999; Christensen and Johnson, 1995; Cox and Niou, 1994); and (2) the ‘variable vote assumption’ that part of a candidate’s votes may be lost (Patterson, 2009).

The main independent variables we use are: (1) the LDP’s seat-share in the Diet at t; (2) the ambition of faction i; (3) the presence of rival factions that were also ambitious for the next presidential race for the first set of models; and (4) the number of non-incumbent candidates supported by ambitious factions. A relatively straightforward measure, the LDP’s seat-share in the Diet at t is meant to capture T_t , or the LDP’s collective interest in securing a majority. If the LDP’s seat-share is high, then it may give a faction some leeway to support candidates in marginal cases, while if the LDP’s majority status is shaky, selfish behaviour may be discouraged.

The ambition of faction i is difficult to measure. It could be simply a dichotomous dummy coded 1 if a faction ran for the next presidential race. This is problematic, because it can be biased in two ways: (1) although a faction was ambitious before the election, it gave up running for the party presidency, because it could not win enough legislators (e.g. Bamboku Ohno was well known for his ambition to become Prime Minister, but he always had to abandon his aspirations); or (2) although a faction was not ambitious before the election, it eventually ran for the race or won the race, because of some unexpected events after the election (e.g. Zenko Suzuki’s assuming Prime Minister in 1980). So we consulted pre-election articles from Asahi and Yomiuri, two major newspapers, about factions’ intentions to run for the next presidential race. We searched for any articles in which factional leaders openly announced their intentions to run for the race, as well as any articles in which newspaper journalists mentioned specific factional leaders’ intentions to run for the race. We also consulted other available sources on factional competition for the presidential race to check if there were any inconsistencies (Leiserson, 1968; Sato and Matsuzaki, 1986; Thayer, 1969). What these two rounds of newspaper search and bibliographical checks produced is reported in Table 1. We coded Ambitious Factions as 1 if a faction, implicitly or explicitly, intended to run for the next race and 0 otherwise.⁶

Table 1.

Ambitious factions.

Election year	Ambitious factions	Sources
1960	Ikeda, Sato, Fujiyama, Kohno, Ohno	Yomiuri, July 11, 1960: 2; Asahi, October 28, 1960: 2
1963	Ikeda, Sato, Fujiyama, Kohno, Ohno	Asahi, May 31, 1963: 2
1967	Fujiyama, Sato, Miki, Fukuda	Asahi, November 2 (Evening), 1966: 1; Asahi, December 31, 1966: 2
1969	Miki, Sato	Asahi, September 13, 1969: 1
1972	Fukuda, Tanaka, Miki, Ohira, Nakasone	Asahi, August 5, 1972: 2; Asahi, July 9, 1972: 2; Asahi, October 14, 1972: 2; Asahi, October 30, 1972: 2
1976	Fukuda, Miki	Yomiuri, June 11, 1976: 2
1979	Ohira, Fukuda, Nakasone	Asahi, March 24, 1979: 4; Asahi, September 22, 1979: 2
1980	Kohmoto, Nakasone, Ohira	Asahi, February 7, 1980: 2; Asahi, January 26, 1980: 2
1983	Fukuda, Nakasone	Yomiuri, November 4, 1983: 2; Asahi, November 16, 1983: 2
1986	Suzuki, Fukuda, Tanaka	Yomiuri, February 10, 1986: 2; Yomiuri, May 15, 1986: 3; Asahi, July 6, 1986: 3
1990	Abe, Miyazawa	Asahi, November 26, 1989: 2
1993	Watanabe	Asahi, July 8, 1993: 7

Table 2.

Logit estimations for a faction’s supporting a non-incumbent candidate.

	Model 1	Model 2
LDP size	2.415*** (0.817)	2.114*** (0.816)
Ambitious faction	0.288*** (0.072)	0.030 (0.096)
Rivalry	0.173*** (0.047)	0.003 (0.065)
Rivalry * Ambitious faction		0.327*** (0.082)
Controls
Electoral performance	0.795*** (0.072)	0.787*** (0.072)
LDP incumbent retired	0.888*** (0.101)	0.884*** (0.101)
LDP incumbent retired * Electoral performance	0.880*** (0.135)	0.897*** (0.135)
Presidential faction	0.027 (0.103)	0.037 (0.103)
Secretary-General faction	–0.011 (0.096)	–0.025 (0.096)
Mainstream faction	–0.114 (0.080)	–0.124 (0.080)
Opposition incumbents	1.341*** (0.217)	1.314*** (0.217)
Faction size	7.928*** (0.648)	7.812*** (0.644)
Time until next race	–0.009** (0.004)	–0.009** (0.004)
Magnitude	0.015 (0.040)	0.013 (0.040)
M increased	0.677*** (0.225)	0.660*** (0.224)
Urban–rural	–0.070* (0.038)	–0.065* (0.038)
Effective no. of candidates	4.804*** (0.517)	4.709*** (0.517)
Vote division	0.460*** (0.066)	0.483*** (0.066)
Constant	–6.210*** (0.498)	–5.900*** (0.500)
Obs	13093	13093
Log likelihood	–3387.940	–3379.857
Pseudo R²	1195.459	1211.625

Standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1.

We also create a variable to represent the presence of a rival faction: Rivalry is the number of incumbent candidates from other ambitious factions in district j at t. When ambitious, faction i might want to support a candidate in a district where a rival faction had a stronghold; if a candidate could defeat the rival incumbent, then the benefits would be maximal; and even if a candidate could not win a seat but the rival incumbent also lost due to the vote dispersion, still faction i would get some benefits. Ambitious Factions’ Presence is coded as the number of non-incumbent candidates, including candidates endorsed by the LDP and conservative LDIs, supported by ambitious factions.

A range of controls is included in our models. For X_i,j,t , we control for faction i’s Electoral Performance in district j at t – 1 as the sum of votes that faction i’s candidates got in j at t – 1, divided by the Droop quota. For example, if a candidate from faction i got 135 percent of the Droop quota at t – 1, then faction i may want to support another candidate in the same district, because gaining 60 percent of the quota may suffice a seat with the surplus 35 percent being transferred from the existing incumbent. If faction i did not support any candidates at t – 1, then this Electoral Performance variable takes the value of zero. We also control for the influence of an open seat created by natural termination (retirement or death) with a dichotomous dummy LDP Incumbent Retired: this variable is coded 1 if an incumbent from the LDP or an LDP-affiliated independent incumbent did not run in district j at t, and 0 otherwise. If an open seat was born in a district, then factions may tend to take advantage of it. We also include an interactive term between Incumbent Retired and Electoral Performance, as factions tend to hire and run a retiree’s son or daughter, relative or secretary to inherit the seat (Ishibashi and Reed, 1992) thus a faction whose candidate showed a sufficiently strong performance at t – 1 would have the incentives to hire a second-generation candidate to inherit the seat.

For F_i,t , we control for faction i’s characteristics at t with Presidential Faction, Secretary-General Faction, Mainstream Faction and Faction Size. The literature suggests that the presidential, Secretary-General and mainstream factions were advantaged in getting LDP nominations (Cox and Rosenbluth, 1996), which could affect factional leaders’ strategies to support candidates. The size would also matter, as the bigger and therefore financially more powerful factions would be more likely to support more candidates. Faction Size is the number of faction i’s members relative to the size of the LDP at t. We also expect that the sooner the next presidential race would be, the more likely factions are to support candidates. So we calculate the anticipated date for the subsequent presidential race – the LDP president’s term was two (1955–1972 and 1978–1993) or three years (1972–1978) – and code Time until Next Race as how many months were left until the next anticipated race.

For D_j,t , we control for district j’s characteristics at t. Opposition Incumbents is included to control for the strength of the opposition parties in a district, which might discourage factional leaders from supporting candidate(s). The variable is the number of opposition incumbents (the Japan Socialist Party, the Democratic Socialist Party, the Clean Government Party, the Japan Communist Party, and others) present in district j at t, divided by district magnitude. Magnitude is district magnitude; a faction would be more likely to support candidate(s) in a larger district. To respond to demographic changes and malapportionment, magnitude was changed for some districts from 1955 to 1993; M Increased is coded 1 if district j experienced an increase in magnitude before t and 0 otherwise. Urban-Rural is a four-scale urban-rural measure, with 1 being the most rural and 4 the most urban. Rural districts would have more over-nominations, because rural districts tend to be geographically larger and therefore candidates could geographically divide their votes more easily within the district (Mizusaki and Mori, 2007; Tatebayashi, 2004). Effective Number of Candidates measures how crowded district j was at t – 1. It is calculated as the natural log of [the effective number of candidates, divided by M + 1]. If a district was in the Duvergerian equilibrium, it should be 0, but we have variations; relatively crowded districts may make the electoral results uncertain and encourage factions to support more candidates. Vote Division measures how effectively (or ineffectively) LDP candidates divided votes at t – 1. This is calculated as [the total number of votes divided by the Droop quota at t – 1] minus [the total number of seats the LDP and LDP-affiliated conservative independents won at t – 1]. In other words, it measures the differential between the number of seats the LDP could have won at t – 1 and the number of seats the LDP actually won at t – 1. If it is positive, then it means that if the LDP could have divided votes effectively there would have been an additional seat for the LDP, giving factional leaders the opportunity to support another candidate.

For equation (2), we control for Opposition Fragmentation and Malapportionment. Scholars have pointed out several mechanisms that helped the LDP stay in power despite the coordination difficulties posed by the electoral system: the opposition force’s fragmentation (Cox and Niou, 1994; Kohno, 1997) and malapportionment (Baker and Scheiner, 2007; Christensen and Johnson, 1995). Opposition Fragmentation is measured by the normalized Hirshman-Herfindahl index and Malapportionment is a district’s standard score for the number of voters per seat. We also control for LDP Incumbent Retired, Opposition Incumbents, Magnitude, Urban–Rural, Effective Number of Candidates and Vote Division, factors that can affect factions’ incentives to support non-incumbent candidates.

Analysis

Our results for the equation (1) are presented in Table 2. To explain our controls first, the variables Electoral Performance, LDP Incumbent Retired and the interactive term of these significantly matter. This suggests first that an open seat generated by natural termination of a legislator encourages factions to support new candidates in the district and, second, that if a faction owned such an open seat it wants to recruit a successor to maintain the seat. This confirms the existing literature on second-generation politicians in Japan (Ishibashi and Reed, 1992). Time until Next Race shows negative signs, meaning that factions became less enthusiastic about increasing their members when the next presidential race was distant. Slightly puzzling is that the presence of opposition party incumbents matters negatively: if the opposition force’s seat-share was high in a district, then a faction was more likely to support a new candidate, although the coefficient suggests that the likelihood increased only moderately (an increase in the opposition force’s seat-share by two standard deviations (0.38) would increase the likelihood by 1.6 percent, on average). It could be the case that, when choosing to support a new candidate in a district, a faction generally aimed to take an incumbent seat from the opposition, on average.

District magnitude also matters, although Magnitude per se did not encourage a faction to support a candidate. Rather, it is whether or not a district increased its magnitude that matters: when the size of a district was enlarged, the electoral results become less predictable, and this encouraged factions to run more candidates in general. The effective number of candidates (relative to magnitude) at t – 1 also matters. A more crowded district in previous elections heightened uncertainty over the electoral results and provided chances for factions to support candidates. The positive and significant coefficient of Vote Division implies that if LDP candidates and LDP-affiliated independents worked efficiently in dividing votes in the previous elections, then this did not allow factional leaders any leeway to encroach on incumbents’ strongholds.

Our primary independent variables show the expected signs with significance. Table 3 shows how the likelihood of a faction’s supporting a new candidate would change using the results attained in Model 2.⁷ First of all, the incentives to cultivate a collective action – the LDP’s seat-share – matter moderately but significantly. If the size of the LDP relative to the assembly size was small, factional leaders would need to pay some heed to the worst-case scenario – the LDP’s losing a majority – coming about due to over-nomination errors. On the other hand, a stable LDP majority provided relatively fewer incentives for factions to collaborate. Our hypothesis 1 is confirmed.

Table 3.

How the likelihood of a faction’s supporting a non-incumbent candidate would change.

Independent variables	Change	Likelihood change
LDP size	From minimum (0.43) to maximum (0.60)	+1.3% (+0.3% – +2.3%)
Ambitious faction Rivalry	From 0 to 1 From 0 to 3	+6.4% (+3.9% – +9.1%)
Ambitious faction Rivalry	From 0 to 1 From 0 to 4	+9.9% (+5.6% – +14.9%)

Simulations based on Model 2. All the other variables set at the median value. 95% confidence intervals in parentheses.

Table 4.

Logit estimations for the over-nomination error.

	Model 3 ‘Invariable vote’ assumption	Model 4 ‘Variable vote’ assumption
LDP size	4.85*** (1.83)	4.35** (1.69)
Ambitious factions’ presence	0.45*** (0.10)	0.55*** (0.10)
Controls
Opposition fragmentation	0.11 (0.50)	0.97** (0.45)
Malapportionment	0.026 (0.12)	–0.066 (0.11)
LDP incumbent retired	0.21 (0.21)	0.19 (0.20)
Opposition incumbents	–1.04* (0.55)	–0.79 (0.52)
Magnitude	0.18 (0.11)	0.35*** (0.10)
Urban–rural	0.033 (0.12)	0.0026 (0.11)
Effective no. of candidates	8.48*** (1.30)	8.83*** (1.23)
Vote division	–0.19 (0.13)	–0.13 (0.13)
Constant	–5.98*** (1.15)	–6.43*** (1.08)
Obs	1,596	1,596
Log likelihood	–521.88	–582.02
Pseudo R²	0.1103	0.1297

Standard errors in parentheses. ***p < 0.01, **p < 0.05, *p < 0.1. On the ‘invariable vote’ and ‘variable vote’ assumptions, Patterson (2009).

Second, ambition in the presidential race matters for over-nomination. The models suggest that a faction gunning for the party presidency in the next race was more likely to support a new candidate. Furthermore, this is conditional on the presence of incumbents from rival ambitious factions: the presence of ambitious rivals further encouraged ambitious factions to support candidates. In other words, a faction positioning itself to be a serious contender for the next president was more likely to support candidates in districts that were the stronghold of a rival ambitious faction. The loss of a rival faction member cost less, because it could reduce the size of those rival factions and give an edge in the presidential election. For example, an ambitious faction was 9.9 percent (on average) more likely to support a new candidate in a district with four incumbents from ambitious rivals, compared to the case in which a faction was not ambitious and a district did not have any incumbents from ambitious factions. Therefore our hypotheses 2 and 3 are confirmed.

We also argue that the conservative camp could never solve the collective action problem because of the LDP’s presidential elections, Prime Minister’s constitutional powers and the highly personalistic nature of SNTV. To test this claim, we offered the second equation to test the fourth hypothesis that ambitious factions indeed cost some seats that the LDP could have won if there were a strict coordination mechanism. Our tests for the fourth hypothesis are shown in Table 4. The significant and positive signs of Effective Number of Candidates suggest that the over-nomination error of the LDP is path-dependent to some extent. Like our results in Table 2, the crowdedness of a district and therefore uncertainty over the electoral results affected factions’ calculations about whether or not they should support candidates. Although not consistently significant throughout the two models, the positive signs of Opposition Fragmentation and Magnitude are also confirmatory with this electoral uncertainty view. With the opposition force more fragmented and district magnitude higher, the elections would become more uncertain and factions would take advantage of this uncertainty.

What is important for our purpose is that we can find again that the collective good for the LDP – reducing the over-nomination error – could be achieved when the costs of losing a seat for the party were felt to be greater, while ambitious factions tended to spoil the collective good. The results do not significantly change when we use different dependent variables based on different assumptions. Much the same as what we have already shown in Table 3, Table 5 shows that when the LDP had some extra seats, the conservative camp was, on average, 7.4 percent more likely to engage in self-destructive electoral competition. In addition, ambitious factions were indeed the main drivers of the over-nomination error. When a district had one or more non-incumbent candidates supported by ambitious factions, the conservative camp was significantly more likely to lose some seats that it could have won. Our hypothesis 4 is confirmed.

Table 5.

How the likelihood of nomination errors would change.

Independent variables	Change	Likelihood change
LDP size	From minimum (0.43) to maximum (0.60)	+7.4% (+1.7% – +12.7%)
Ambitious factions’ presence	From 0 to 1	+5.8% (+3.7% – +7.9%)
Ambitious factions’ presence	From 0 to 2	+14.3% (+8.5% – +20.7%)

Simulations based on Model 4. All the other variables set at the median value. 95% confidence intervals in parentheses.

All in all, each of our hypotheses was confirmed. When the need to protect the LDP’s collective interest in securing a majority became more urgent, factions responded by becoming more collaborative. However, if a faction leader harboured an ambition for the next presidential race, then this eroded the incentives to achieve the collective action of optimizing the number of candidates in order to eliminate over-nomination and maximize expected seats won by the party. This tendency was accentuated in a district where rivals had incumbents, as ambitious factions would gain office-seeking benefits by defeating these incumbents – two birds with one stone. This was enabled by the extremely personalistic SNTV system where personal vote-mobilizing organizational networks called koenkai dominated campaigning.

Conclusion

Because they structure the tool sets available to both party leaders and the (possibly defecting) rank and file, electoral systems affect the likelihood of parties’ solving the fundamental collective action problems they face. In our analysis here, we found that Japan’s LDP was unable to solve its collective action problem in agreeing on the number of candidates to run in a district – a basic issue at the core of the function of any party. Without effective control over nominations by the party head, ambitious figures within the party risked losing the party seats in the legislature in order to self-interestedly advance their own designs. The nature of the electoral system contributed mightily to the weakness of the LDP as a party, even in the most fundamental of arenas.

We point to three factors making collective action difficult. The first is the intra-party rule to determine the party president. The LDP used a majority run-off system whereby each of the Diet members can cast one vote to choose the president. This encouraged ambitious factions to make every effort to expand their sizes. The second factor was exclusive access to a wide range of office benefits given to the president of the LDP. During the period of 1955 to 1993, the LDP stayed in power, and the president of the party automatically became the prime minister, with the constitutional prerogative to allocate important governmental positions. The third factor was a relatively unimportant party label under the extremely personalistic SNTV system. Leaders in the party – in fact they were also the bosses of factions that needed expansion – could not tame factional leaders because they would not permit any effective sanction (e.g. giving no party nomination) to restraining their own behaviour.

We utilized data covering all the elections from 1960 to 1993 to test these arguments. The results were confirmatory. First, if the size of the LDP relative to the assembly size was small, fears of losing an LDP majority due to over-nomination errors restrained factional leaders. Second, if a faction aimed for the party presidency in the next race, then it was more likely to support candidate(s). Third, this was further encouraged by the presence of rival factions in a district: factions were more likely to support candidates especially in those districts in which similarly ambitious factions had strongholds. Fourth, the unsolved collective action dilemma and ambitious factional leaders’ selfish behaviour indeed cost some seats for the LDP by dispersing votes among the conservative camp. Further research is necessary to examine how factional party politics is affected by other types of electoral systems, and how political party leaders are able, or not, to overcome the coordination and ‘problems of the commons’ obstacles these electoral system incentives provide.

Afghanistan, Japan's upper house, Indonesia's upper house, Jordan and Vanuatu use SNTV now, but our findings are of general interest. We seek to answer these key theoretical questions: how the incentives of a particular electoral system shape the organization of parties and their behaviour; how and whether political parties can overcome their collective action dilemmas; and how the intervening variable of party organization affects party behaviour.

Footnotes

Acknowledgements

We acknowledge helpful comments from Steven R. Reed and Matthew Søberg Shugart, the journal editor and the anonymous reviewers, and research assistance from Itaru Yanagi.

Funding

This research received support from the National Science Foundation for grants no. SES - 0751662 (University of California, San Diego) and SES- 0751436 (University of Washington) which funded the completion of the database used in this article.

Notes

References

Asahi Shinbun

Seijibu

(1968) Seito to Habatsu (Parties and factions). Tokyo: Asahi Shinbunsha (Japanese).

Baker

Scheiner

(2007) Electoral system effects and ruling party dominance in Japan: A counterfactual simulation based on adaptive parties. Electoral Studies 26(2): 477–491.

Browne

Sunwoong

Kim

(2003) Factional rivals and electoral competition in a dominant party: Inside Japan’s Liberal Democratic Party, 1958–1990. European Journal of Political Research 42(1): 107–134.

Browne

Patterson

(1999) An empirical theory of rational nominating behaviour applied to Japanese district elections. British Journal of Political Science 29(2): 259–289.

Carey

Shugart

(1995) Incentives to cultivate a personal vote: A rank ordering of electoral formulas. Electoral Studies 14(4): 417–439.

Christensen

(2000) Ending the LDP Hegemony: Party Cooperation in Japan. Honolulu: University of Hawaii Press.

Christensen

Johnson

(1995) Toward a context-rich analysis of electoral systems: The Japanese example. American Journal of Political Science 39(3): 575–598.

Cox

Niou

(1994) Seat bonuses under the single nontransferable vote system: Evidence from Japan and Taiwan. Comparative Politics 26(2): 221–236.

Cox

Rosenbluth

(1994) Reducing nomination errors: Factional competition and party strategy in Japan. Electoral Studies 13(1): 4–16.

10.

Cox

Rosenbluth

(1996) Factional competition for the party endorsement: The case of Japan’s Liberal Democratic Party. British Journal of Political Science 26(2): 259–269.

11.

Cox

McCall Rosenbluth

Thies

(1999) Electoral reform and the fate of factions: The case of Japan’s Liberal Democratic Party. British Journal of Political Science 29(1): 33–56.

12.

Hayao

Kenji

(1993) The Japanese Prime Minister and Public Policy. Pittsburgh, PA: University of Pittsburgh Press.

13.

Ishibashi

Reed

(1992) Second-generation Diet members and democracy in Japan: Hereditary seats. Asian Survey 32(4): 366–379.

14.

Jiyu Minshu

Henshubu

(1995) Jiyu Minshuto Sosai Senkyo no Ayumi (The history of the LDP’s presidential elections). Jiyu Minshu 1996(9): 56–81 (Japanese).

15.

Kawato

(1996) Sinioriti Rulu to Habatsu: Jiminto ni okeru Jinji Haibun no Henka (Seniority rule and factions: The LDP’s posts allocation and its change). Revaiasan Rinji Zokan: 111–145 (Japanese).

16.

Kohno

(1992) Rational foundations for the organization of the liberal democratic party in Japan. World Politics 44(3): 369–397.

17.

Kohno

(1997) Electoral origins of Japanese socialists’ stagnation. Comparative Political Studies 30(1): 55–77.

18.

Krauss

Pekkanen

(2010) The Rise and Fall of Japan’s LDP: Political Party Organizations as Historical Institutions. Ithaca, NY: Cornell University Press.

19.

Leiserson

(1968) Factions and coalitions in one-party Japan: An interpretation based on the theory of games. American Political Science Review 62(3): 770–787.

20.

Maeda

(2012) An irrational party of rational members: The collision of legislators’ reelection quest with party success in the Japan Socialist Party. Comparative Political Studies 45(3): 341–365.

21.

McCubbins

Frances

(1995) ‘Party Provision for Personal Politics: Dividing the Vote in Japan’, in Peter F. Cowhey and Mathew D. McCubbins (eds) Structure and Policy in Japan and the United States, pp. 35–55. Cambridge: Cambridge University Press.

22.

Mizusaki

Mori

(2007) Sosenkyo no Tokuhyo Bunseki: 1958–2005. (Japanese general elections: 1958–2005). Tokyo: Bokutakusha (Japanese).

23.

Patterson

(2009) Candidates, votes, outcomes: A method for evaluating nomination strategies in MMD/SNTV electoral systems. Legislative Studies Quarterly 34(2): 273–285.

24.

Ramseyer

McCall Rosenbluth

(1993) Japan’s Political Marketplace. Cambridge, MA: Harvard University Press.

25.

Reed

(1990) Structure and behavior: Extending Duverger’s law to the Japanese case. British Journal of Political Science 20(3): 335–356.

26.

Reed

(2009) Party strategy or candidate strategy: How does the LDP run the right number of candidates in Japan’s multi-member districts? Party Politics 15(3): 295–314.

27.

Sato

Matsuzaki

(1986) Jiminto Seiken (The LDP Administration). Tokyo: Chuo Koron Sha (Japanese).

28.

Swindle

(2002) The supply and demand of the personal vote. Party Politics 8(3): 279–300.

29.

Tatebayashi

(2004) Giin Kodo no Seiji Keizaigaku: Jiminto Shihai no Seido Bunseki (The logic of legislators’ activities: Institutional analysis of LDP dominance in Japan). Tokyo: Yuhikaku (Japanese).

30.

Thayer

(1969) How the Conservatives Rule Japan. Princeton, NJ: Princeton University Press.

31.

Tomz

Wittenberg

King

(2001) CLARIFY: Software for Interpreting and Presenting Statistical Results 2.0. Cambridge, MA: Harvard University,

32.

Watanabe

(1958) Habatsu: Hoshuto no Kaibo (Factions: Analyzing the Conservative Party). Tokyo: Kobundo (Japanese).