Where Have All the Zoku Gone? LDP DM Policy Specialization and Expertise. Robert Pekkanen University of Washington

Transcription

1 Where Have All the Zoku Gone? LDP DM Policy Specialization and Expertise Robert Pekkanen University of Washington Benjamin Nyblade University of British Columbia Ellis S. Krauss University of California, San Diego Paper prepared for delivery at the Stanford Conference on Legislative and Electoral Politics in Japan, June 11-13, Please note that all results and analysis are more preliminary than usual with this paper. Comments are especially welcome. If you do wish to cite this paper, please contact the authors to make sure you have the most recent version as we are planning to revise and update this paper in the very near future. All data and analyses are based on Krauss and Pekkanen s J-LOD Database (Japanese Legislative Organization Database) unless otherwise specified. 1

2 Where Have All the Zoku Gone? LDP DM Policy Specialization and Expertise This paper distinguishes between politicians policy expertise and policy specialization and explores changes in DM policy expertise and policy specialization in Japan related to changes in the electoral system. Traditional measures of policy expertise for Liberal Democratic Party House of Representatives DMs (membership in seisaku zoku policy tribes ) in Japan show a decline in the number of policy experts since electoral reform. However, this paper shows that the decline in policy experts is attributable to the nature of the measure of expertise and the decline in the overall seniority of the LDP party caucus, rather than changes in the electoral rules. However, political career path analysis of LDP House of Representative DMs in Japan shows that policy specialization, particularly by junior DMs, has declined since electoral reform. We attribute this decline in policy specialization to the need to represent a broader range of constituents as most DMs moved from multi-member districts to single-member districts following electoral reform. 2

3 Introduction Why might politicians specialize in certain policy areas? There is a vast literature on this in the context of the study of the U.S. Congress, and a somewhat less than vast literature on this in the comparative context. Common to most explanations of policy specialization is that the electoral connection drives specialization, be it due to the individual vote-garnering incentives facing legislators or to the electoral incentives that underpin collective incentives for the development of institutional policy-making (and credit-claiming) capacity and/or a valuable party label. Expertise and Specialization In this paper we distinguish between expertise and specialization. These concepts have often been conflated in the extant literature, but distinguishing them is conceptually and empirically fruitful. Specialization could be a means to creating expertise, but might also be valued for its own sake. For example, specialization could prove useful in signaling to voters, perhaps for the purpose of dividing the vote as was widely argued under Japan s SNTV electoral system. Expertise, and not specialization per se, leads to improved policy-making and thus has greater value to the party label. Differentiating expertise and specialization allows us to formulate more precise hypotheses in investigating, say the impact of electoral change in committee structure or legislators career paths. The literature on policy specialization in Japan originally focused on the development of expertise by Japanese politicians, and frequently placed this in the context of the long-running bureaucrats vs. politicians: who rules? debate (e.g. Sato and Matsuzaki 1986, Inoguchi and Iwai 1987). However, some of the literature moved to consider specialization and its electoral 3

4 purposes, rather than simply the development of expertise by DMs (e.g. Ramseyer and Rosenbluth 1993, McCubbins and Rosenbluth 1995, Tatebayashi 2004). In other words, policy expertise is the extent to which a particular representative is attributed knowledge and influence in a specific policy area due to her having occupied certain key party, parliamentary, or government positions in that sector; policy specialization, on the other hand, refers to the extent to which a legislator tends to concentrate in certain policy areas. One may acquire policy expertise through specialization or not. Most who specialize may become experts but not all experts may specialize. In this paper we seek to consider both the development of expertise over the course of MPs careers, and how that relates to specialization and electoral incentives. If it is indeed electoral incentives that create incentives for legislators to specialize and develop expertise in certain policy areas over the course of their careers, different electoral incentives should lead to differences in the nature and extent of policy specialization by legislators. In this paper we rely on the natural experiment provided by an electoral reform in Japan s House of Representatives (HoR) that was enacted in 1994 to analyze the differences in policy expertise and policy specialization by HoR DMs from Japan s ruling Liberal Democratic Party (LDP). Unlike most existing analyses of specialization, we consider specialization over the course of a legislator s career, and find that the post-reform decline in policy experts in Japan s LDP was primarily not due to electoral reform, but that the decrease in policy specialization by (particularly junior) DMs is likely to be due to the incentives of the new electoral system. 4

5 Policy Specialization: Theoretic and Empirical Background The literature on policy specialization has been dominated by studies of the U.S. Congress and committee assignments, although there is disagreement on whether specialization in committee roles is simply due to individual legislators responding to constituents by ensuring durable institutionalized gains from trade, enhancing their collective interest in their common party label, or attempting to better use governmental power (Shepsle 1978; Weingast and Marshall 1988; Cox and McCubbins 1993; Krehbiel 1991). There is a dearth of comparative literature on specialization and the electoral connection. Those that exist are primarily on Germany and Japan and also differ on both whether such a connection is important and if so, how. Some studies on Germany show no connection (see Nohlen 1990, Ismayr 1992) but other studies argue for a connection in which local district representatives are more likely to specialize in assignments that allow them to bring particularistic benefits to their districts than PR candidates who are more likely to be assigned to public goods or higher policy posts (Lancaster and Patterson 1990; Stratmann and Baur 2002; see also Pekkanen et. al. 2006). Almost all the above studies, however, focus on electoral incentives to specialize via service on committees or other sub-sets of organizations in terms of cross-sectional or committee-level data, rather than individual-level longitudinal career data. There are a number of reasons why this might be problematic, particularly in the case of Japan, where senior LDP politicians may no longer a formal role on a PARC or parliamentary committee or in cabinet, but may still be able to claim credit as a senior member of a certain policy tribe. The comparative literature on individuals career specialization and its connection to electoral incentives is probably most developed in studies of Japanese politics. Influential 5

6 analysts of Japan s pre-1994 Single Non-Transferable Vote (SNTV) electoral system saw policy specialization as a rational response to the dilemma of the ruling Liberal Democratic Party (LDP) to dividing the vote among multiple candidates in the multi-member, single vote SNTV electoral system. (McCubbins and Rosenbluth 1995: 48-52; Ramseyer and Rosenbluth 1993: 33-34), although others (Fukui 1970: 87) have implied it is a simple response of individual Diet members interests due to pressures from constituents, interest groups, or ministries, something that might occur for example even in SMD systems. In the former conception, however, policy specialization is particularly a means by which individual LDP Diet Members from the same district differentiate themselves from their fellow representatives in that district to help gather their personal vote. Policy specialization takes place by frequent assignment of LDP Diet Members to the same divisions of the party s Policy Affairs Research Council (PARC) and their moving up the career ladder through various executive positions within those divisions. To test this hypothesis, McCubbins and Ramseyer used a random sample of data from twenty-six districts in 1990, and find that the results indicate that the process of distribution of Diet Members to PARC divisions was indeed nonrandom (McCubbins and Ramseyer: 52) and therefore this kind policy specialization particularly to deliver pork barrel to districts was practiced among LDP members from the same district.. Subsequently, there has been some analyses that have revised this view of the SNTV electoral system-policy specialization connection. Tatebayashi and McKean (2002) and later Tatebayashi with a more extensive analysis (Tatebayashi 2004) assume two different specialization strategies of vote-division, one based on geography and the other on policy specialization: In districts with equilibrium based on geographical division of the vote, politicians tended to focus on delivery of specific pork-like services to their own sub-region in the 6

7 electoral district. On the other hand, in districts with an equilibrium based on policy issues, LDP politicians who each became policy specialists in their (different) respective policy fields could, taken together, actually cover a wide range of policy issues on behalf of their constituents. (Tatebayashi and McKean 2002: 4) This is then validated with a comprehensive data analysis based on five elections. Their findings indicated that those who concentrated on bringing benefits to a specific part of their district (geographic specialization) tended to attain the pork barrel division assignments, whereas those who specialized in policy sector differentiation for electoral reasons were not those who were on the pork barrel divisions, and there was a lack of specialization among those in the same districts. In other words, policy specialization for electoral differentiation and pork barrel for constituents were separate and different types of specialization. 2 Another study comparing Japan s SNTV and Ireland s STV systems (Swindle 2002) further found a major difference in candidate-driven incentives to provide personalist policy outputs as in Japan and collective party-driven incentives to do so in Ireland. By implication, when voters cannot transfer votes as was the case in Japan, there are even greater candidatedriven incentives to attempt specialist careers. Before electoral reform, DMs in Japan s Lower House were primarily elected under a system of medium-sized districts in which three to five DMs were elected in each district with voters casting a single, non-transferable vote. Under this system DMs could appeal to a narrower subset of the population than in single-member districts (SMDs), as they frequently needed little more than 15% of the vote to be elected, rather than the plurality of votes required in most SMD systems. The existing literature on elections and politics in Japan under the old electoral system has emphasized that Japanese politicians had generally cultivated a narrower 2 For reviews and critques of Tatebayashi s 2004 book, see Saito 2005 and Masuyama 2005) 7

8 constituency (jiban) than would be expected in SMDs (Reed 1990, Ramseyer and Rosenbluth 1993, Tatebayashi 2004). After electoral reform, DMs in Japan s House of Representatives have been elected from a parallel mixed electoral system with 300 DMs elected from SMDs, and 180 (initially 200) DMs elected from PR lists. Most of those elected from PR lists also ran in SMDs, as the system allows dual-listing of candidates. 3 In general then, we expect that HoR DM policy specialization should decline following electoral reform as there are greater incentives under an SMD system for representatives to broaden the policy areas through which they respond to a more diverse constituency to get reelected, as there is now only one representative per district, rather than several and it requires a larger proportion of votes to win than in SNTV. Our first empirical section of this paper focuses on DM policy expertise, drawing on traditional measures of policy expertise for LDP DMs: membership in seisaku zoku policy tribes. By this measure there is a marked decline in the number of policy experts since electoral reform, and a slight decrease in the emphasis on specialization in distributional politics. Thus the question: where have all the zoku gone? Despite our theoretic predictions suggesting that we should see less narrow policy specialization by DMs in the post-reform period, the data suggests that the primary cause of the decline in zoku giin is more prosaic: the LDP HoR party caucus is simply less experienced post-reform, thus there are fewer DMs who qualify as experts by traditional measures. However, when we conduct analyses on new measures of policy specialization, we find that policy specialization has decreased as DMs have moved from multi-member districts to single-member districts, and that this result is statistically significant and robust to multiple 3 Candidates who were dual-listed were frequently not only dependent on their rank on the party list for election, but their rank could be influenced by their electoral performance in their district, thus causing dual-listed DMs elected in PR to behave in many ways more similarly to SMD DMs (e.g. Pekkanen et al 2006). 8

9 operationalizations. We also find that during the transition period, when DMs were elected under the old electoral system but could act in anticipation of the new system they increased the breadth of issue areas they focused on, and that there has been an increase in specialization over the first ten years under the new electoral system, perhaps as DMs have increasingly adapted to the new system. We find the effects of electoral reform on policy specialization to be particularly important for more junior DMs, while they are more vulnerable electorally and before they achieve status as policy experts according to the traditional Japanese conceptualization of policy expertise in the LDP. Expertise: Where Have all the Zoku Gone? We begin our examination of the empirical record of DM policy expertise and specialization by looking at the standard categorization of LDP DM policy experts in the literature: membership in a policy zoku (tribe). Ever since 1976, journalists and academics have dubbed those who so specialize and attain various important positions within the party, Diet, and government related to the same policy area, as part of that policy area s policy tribe (seisaku zoku), and attributed to them a major influence over policy through these positions and their developed connections to related interest groups and bureaucrats (Sat and Matsuaki 1986; Inoguchi and Iwai 1987; Yuasa 1986; Nihon Keizai Shimbunsha 1985). 4 The definition of who actually comprise zoku, however, vary depending on the criteria adopted. Sato and Matsuzaki identify four differerent definitions but reject three of them on the grounds of being either too broad or not reflecting zoku relationships to bureaucratic 4 The first reference to zoku Diet Members seems to have been in a newspaper article in 1976 discussing the Lockheed scandal at that time and referring to those LDP Diet Members who specialized in airline transportation. See ("K k zoku giin, Asahi Shimbun 1976) 9

10 jurisdictions. 5 So they adopt a fourth definition that is Persons who are prior to their first entry into the Cabinet or first term ministerial experience (you can say that these are largely mid-level representatives) among those representatives who are exercising a strong daily influence concerning policy areas that are compartmentalized with the bureaucratic agency as the basic unit (Sato and Matsuzaki : ). Inoguchi and Iwai (1987) follow a somewhat different approach in identifying zoku. They find Sato and Matsuzaki s approach a bit too narrow, and therefore incorporate those they identify, along with those identified a zoku representatives by journalists and in other books. Compiling a list of such representatives in each policy area from these sources they then construct charts of the posts they held in their careers and the number of PARC divisions and House committees they served on. Key in these charts and in the careers of the zoku are the posts of PARC Chair and Vice-Chair and secondarily the posts of House Committee Chair and Director (riji), the equivalent of a Vice-Chair (Inoguchi and Iwai: ; ). Inoguchi and Iwai also develop a points system for ascribing points to the various levels of posts, which they use to confirm their identification of zoku membership. For Inoguchi and Iwai, the key factors in creating the phenomenon of zoku within the LDP were the adoption of the seniority rule for post acquisition, a change that deprived former bureaucrats of their policy expertise advantages in gaining top posts, and the LDP s vote decline that began in the 1960 that drove LDP politicians to pay greater attention to constituency maintenance and thus to responding to the demands of interest groups and increasing the 5 The three they reject but which are used variously in the media are 1)simply a group of representatives who have an interest in a policy area, but they consider this definition is too wide; 2)representatives who have a strong influence on a certain policy area, but this too is too broad because it could include party executives also; 3) those groups of middle-level representatives who not only have a strong influence in a specific policy area, but are in a position to exercise that on a daily basis. This definition too they find wanting because it does not necessarily reflect the importance of the connections of zoku to specific bureaucratic agencies. 10

11 importance of the policy process within the LDP. It was then an easy step for representatives to establish connections to bureaucratic agencies (24-28). Clearly too, the long-term rule of the LDP made possible the seniority rule s effect of LDP politicians working their way up the ladder of PARC and Diet and sub-cabinet posts to become zoku (Sat and Matsuaki 1986). Despite differences in the definitions and methods for identifying zoku giin, as Table 1 shows, Inoguchi and Iwai and Sato and Matsuzaki s codings of zoku giin are broadly similar. One exception is that Sato and Matsuzaki s identify groups of DMs who are policy experts but did not necessarily self-identify or have formal recognition as zoku in certain areas (the five groups beginning with Cabinet in Table 1). However, when looking at the same zoku, they disagree only on the membership of a few individuals. Moreover, their general assessment that LDP DMs are most likely to specialize in areas of distributive politics (most notably the Big Three of Construction, Commerce and Agriculture) is quite consistent with either coding scheme. How have things changed? The KNP column in Table 1 reports zoku giin membership for 2004 using rules almost identical to that of Inoguchi and Iwai. There are several differences worth noting. First of all, the greatest difference in zoku giin is in overall numbers. In 2004, there were 25% fewer zoku giin (thus the title of the paper). This cannot simply be attributed to a decrease in the strength of the LDP: there were 254 LDP HoR DMs in 1985 and in 2004 there were 246. The greatest declines in zoku giin are in construction (-56%), foreign affairs (-50%), commerce (-46%) and labor (-45%). The only four zoku to see an increase in membership are fisheries, finance, legal affairs and environment, and the first of these may be due to coding 11

12 issues. 6 Overall, although there is a slightly greater decrease in the numbers of zoku giin in distributional policy areas compared to the decreases in high policy and public goods areas, this difference is small enough that it may not be robust to minor coding changes. So what caused this decrease in the expertise of LDP DMs in specific policy areas? Is it that DMs are becoming generalists under the new electoral system to a greater degree as a consequence of the incentives to cultivate broader constituencies in single-member districts? Sadly, the evidence suggests that this is not the cause of the decline in zoku giin. Rather the answer is somewhat more prosaic: the decline in zoku giin is due to changes in the aggregate experience of the LDP. Table 2 reports the seniority profiles of the LDP in 1985 and <Table 2 Approximately Here> House of Representatives DMs in 2004 as a whole are much more inexperienced politicians than those in Half of DMs in 2004 had served 3 terms of fewer, whereas nearly half of DMs in 1985 had served 5 or more terms. There were three times as many superexperienced DMs with 10+ terms in 1985 as there were in In Inoguchi and Iwai s data 98% of all DMs who qualify as zoku giin had served four or more terms in parliament there were only 5 exceptions. And although there were roughly the same number of LDP HoR DMs in 2004 as there in 1985 (254 vs. 246), there was in fact a 27% decline in the number of DMs with 4+ terms, almost identical to the decrease in numbers of zoku giin. Ultimately, although the traditional measure of Japanese LDP DM policy expertise shows a decline in the number of policy experts, as we might expect given the greater importance for DMs in single member districts to develop policy breadth in order to cultivate a broader constituency, ultimately this difference is more likely attributable not the change in electoral 6 We also set aside the decline in Cabinet zoku, which is clearly due to coding differences. In subsequent versions of this paper we may trace the year-by-year zoku membership and also resolve such coding differences, but have not yet had the time and resources to do so. 12

13 system, but to the overall decrease in experience in the LDP House of Representatives caucus, which in turn is primarily due to the greater electoral volatility of the last fifteen years. This seems to be quite indirectly related to electoral reform, if at all. However, as we suggested early on in this paper, expertise and specialization are distinct concepts, and since most of the predictions about DM adaptation to the new electoral system we have are about policy specialization rather than expertise, we forge ahead in the next section to see whether we can find any differences in DM policy specialization over the course of a career using more complete data, new measures and statistical tests. Policy Specialization: New Measures and Tests There appears to be no standard measure in the literature for capturing policy specialization by legislators, nor any standard way of statistically analyzing it over the course of a career. In this section, we present a few fairly simple measures that we use both to capture policy specialization descriptively over the course of LDP DMs careers (both before and after reform) and analyze the differences. All of the analyses reported here are preliminary, but we believe that they provide strong support for our expectation that policy specialization by DMs in Japan is influenced by the rules under which those DMs are elected. We were unable to consider all of the traditional issue areas used in the analysis of zoku giin for these analyses as the administrative reforms in 2001 led to the merger of a number of posts (for example, transportation and construction committees and posts were merged). Instead we limit ourselves to twelve issue areas, including five areas we considered to be high policy (Cabinet, Finance, Legal Affairs, Foreign Affairs, Defense), four areas we considered to be distributional policy (Construction & Transportation, Commerce, Agriculture, Posts & Local 13

14 Affairs) and three that we considered to public goods policy (Education & Science, Labor & Social Welfare & Health, Environment). We began by coding all major posts in the LDP policy-making process in Japan (including posts in parliament, the executive and the party) by issue area. The posts we considered were PARC committee Chairs and Directors; Parliamentary Committee Members, Directors and Chairs; as well as Deputy or Vice Ministers and Ministers. 7 As DMs achieve greater seniority, the type of post they are likely to hold changes, a fact particularly true in the early stages of an DMs career (cf. Kohno 1992, Ramseyer and Rosenbluth 1993). <Figure 1 About Here> In general, the pre- and post-reform patterns of post acquisition by LDP DMs in the early stages of there career are similar, as shown in Figure 1. The most significant difference is that serving as a Chair of an LDP PARC Committee is generally reserved for somewhat more senior DMs in the post-reform period. Of course, simply looking at the types of post DMs attain over the course of their career does not get us at policy specialization. To measure any changes in policy specialization, we instead focus on the extent to which DMs concentrate in particular issue areas or types of issues in the various levels of posts they hold. Specifically, we look at the effective number of issue areas or issue types that an DM has held posts in over the course of their career. The effective number is the inverse of the Herfindahl index that has been frequently used to measure industry concentration (Laakso and Taagepera 1979). Of course, over the course of a career, the number 7 Committee membership rolls in LDP s PARC are no longer fixed after the reforms (DMs may attend committee meetings more freely), making pre- and post-reform comparison of PARC committee membership difficult. However, leaders of the committees (Chairs, Vice-Chairs) are still regularly appointed. See Krauss and Pekkanen

15 of potential opportunities to have different posts increases, thus necessitating that as we consider the specialization of DMs we control for the number of years in office. 8 <Figure 2 Approximately Here> Figure 2 shows the effective number of policy issues and effective number of issue types based on the posts that LDP DMs hold over the first dozen years of their career 9, showing the different patterns for those elected under SNTV, in SMDs and those elected based on PR lists. 10 Figure 2 provides the first preliminary support for our contention that the rules by which DMs are elected affect their policy specialization. The first half-dozen years of an DM is in office sees a gradual increase in the effective number of policy areas and policy types in which an DM holds party, parliamentary or executive posts, after which there is a general stabilization in the degree to which they specialize. However, DMs elected since electoral reform regardless of whether they are elected from SMD or PR, are more likely to hold posts in a greater number of different policy areas, particularly during the first half-dozen years in office. We expect that DMs elected under proportional representation rules should be more likely to specialize than those elected from SMDs, and in general that is what we see. However, this difference is dwarfed by the degree to which LDP DMs elected from SNTV Districts are likely to specialize more narrowly, particularly during an DMs first two terms in office (the first 8 For example, an DM in their first year might hold a single parliamentary committee membership on the foreign affairs committee, and thus they would be coded as having one effective issue area and one effective issue type. In their second year, they might hold a post in agriculture. This would lead to two effective issue areas and two effective issue types. If in the third year, that DM held a post in education, the DM would have three effective issue areas and issue types, but if they were to specialize more post were instead have a post again in agriculture, the effective number of issue areas and issue types would decrease to approximately Due to data collection and coding issues we are limited to those DMs who begin their careers in 1980 or later. Future versions of this paper will likely be able to extend this back to 1975 or so, but given data on post allocation before that is not consistently available, we will most likely not be able to go further back than Due to the dual-listing of candidates in the Japanese electoral system, our PR category here does not include all DMs elected under PR rules, only those who were not double listed (and thus also ran in SMDs) and also not in a Costa Rican arrangement to alternate with another DM in an SMD. As both of those types of DMs are effectively SMD-based, we include them with the SMD DMs (cf. Pekkanen et al 2006). 15

16 six or seven years). An average third-term DM elected under SNTV held posts in (effectively) 2.0 different policy areas, whereas an average third-term DM elected from an SMD held posts in 2.9 different policy areas. Third-term DMs elected in PR on average have held posts in 2.7 different policy areas. On average, DMs elected in PR effectively specialize in just 0.1 different posts than those elected from SMDs, not a particularly sizable difference, suggesting that perhaps LDP DMs elected in PR are not specializing in policy areas the way one might usually expect under PR rules. 11 To what extent are these differences significant and robust to various controls? We consider this question with a few statistical analyses reported in Table 3. Table 3 reports random effects GLS models with the effective number of issue types (the issue areas broken down into high policy, distributional and public goods types) and effective number of issue areas (the twelve categorizations discussed before). In order to capture the non-linear effects of experience on our measures of policy specialization, we include a set of dummy variables for each year of seniority an DM has. In order to be able to compare pre- and post-reform DMs completely, we limit our sample to DMs first sixteen years in the HoR. 12 Models 1a and 1b (on effective number of issue types and issue areas respectively) report our most basic results. In these models we have simply interacted seniority and the type of electoral system under which DMs were elected. We use a simplified dichotomous coding of the 11 There are a number of reasons to expect LDP PR DMs to be more district-oriented than most theoretic perspectives on PR would suggest, given the dual-listing provisions and Costa Rican arrangements that some DMs elected on the PR list take advantage (see Bawn and Thies 2001). For simplicity, we will exclude PR Only DMs from our statistical analyses in the remainder of the paper (they represent less than 10% of our observations and the std errors tend to dwarf any differences we found in preliminary statistical analyses), although in future versions of the paper we will likely address the issue in our reported statistical analyses. 12 Including DMs with greater seniority actually does not disrupt our results, but contributes nothing since we cannot do the comparison across periods for those DMs since our post dataset at the moment goes back only to

17 electoral system, separating out those elected in the old SNTV 3-5 seat districts from those elected in SMDs. 13 As Model 1a and 1b show, there is a statistically significant difference in the degree of specialization of DMs elected under the two systems, but this seems to be only true in the first few years in office (the first four years for issue types, the first eight for issue areas). This suggests that DMs early in their careers in SMDs seem to be acquiring a broader range of posts and experience in various issue areas in their first few terms in office the terms in which they are most vulnerable. However, the difference between senior DMs under the two systems is not particularly strong in this basic model. Models 2a and 2b replicate Models 1a and 1b, except they have been run with the inclusion of year dummies to attempt to control for the vagaries of post allocation in any given year (coefficients for the year dummies are not reported). The general findings of Models 1a and 1b are confirmed, with the year dummies slightly weakening the results of issue types, but strengthening the results for issue areas. Most of the year dummies were statistically insignificant, although there was a general positive trend over time, and the years showing a highly significant positive coefficients quite distinct from this trend. Electoral reform was passed in 1994, but the first election under the new rules occurred in Thus although, DMs in the period are coded as being elected from their SNTV constituencies, those intending to continue on politically could anticipate being elected from an SMD, and thus already faced incentives to broaden their constituency. 14 The electoral logic 13 There was one DM elected in an SMD under the old system for most of the period we examine, and some elected from 2 member districts, we do not include these under the SNTV old system coding however consistent with our theory, those elected in those districts behave more like DMs in SMDs under the new system. 14 Most LDP HoR DMs anticipated running in SMDs in the first election, although some in the end did run solely in PR. However, most of those who ran solely in PR either were so senior that they drop out of our analyses, or were engaged in a Costa Rican arrangement and intended to alternate with another LDP DM in an SMD. 17

18 underlying our theoretic expectations is anticipatory, so this empirical finding concerning the year dummies is not surprising. Similarly, to the extent that we see gradual adaptation to electoral rules (or perhaps generational effects, with those elected for the first time under the new system being particularly sensitive to its incentives), a trend over time towards decreasing specialization should not be surprising, at least not in the post-reform period. Models 3a and 3b incorporate both a dummy variable for the transition period (which affects only those elected from SNTV) and a year-trend variable (1 for 1980, 2 for 1981, etc.), the latter interacted with SNTV also. All of these variables are significant in the manner we might expect, and these model shows an increased difference between the two systems as well, particularly in Model 3a. DMs elected under SNTV during the transition period are much more likely to take posts in issue areas and of issue types which they had not previously had posts in, presumably in anticipation of running under SMD rules in the subsequent election. There also is a strong time trend in the post-reform era: DMs in 2005 are much more likely than those in 2000 or 1997 to be less specialized (hold posts in a broader range of issue areas and types). The net coefficient for the time-trend is close to zero for issue type in the pre-reform period for issue types. Perhaps more interestingly, there is still a significant, positive time-trend in the prereform period for the effective number of issue areas, although the trend is significantly weaker than in the post-reform period. Discussion This paper reports some preliminary theorizing and analysis concerning DM policy specialization and the development of DM policy expertise in Japan, and how this is influenced by electoral rules. Our primary empirical findings suggest that electoral reform in Japan does not 18

19 seem to be directly responsible for changes in the traditional measures of DM policy expertise in Japan. Membership as a policy specialist in a policy tribe (seisaku giin) has declined, but almost directly in proportion to the decline in the number of senior LDP HoR DMs. There does seem to be a significant difference in policy specialization by junior DMs. DMs elected from SMDs are more likely to have posts in a wide variety of policy areas, particularly in their most vulnerable first or second terms (first six or seven years in office). All of these results of course are quite preliminary, and there is substantial room for improvement in this work, both theoretically and empirically. On the theoretic side, we feel it is important to think through more systematically the relationship (and differences) between policy specialization and policy expertise, and how this distinction plays out in the existing literature on legislative organization and party politics. We have included a few of our preliminary thoughts on this matter in this paper, but will need to develop this further as we continue to work on this project. Empirically, there is at least as much work to be done. The models included are quite preliminary, and as is well-understood there are a large number of issues regarding model specification for this sort of time-series cross-sectional statistical modeling. We have neither kept up with the current developments on the best TSCS estimators and specification techniques (the most issue in Political Analysis includes several papers on TSCS for example), nor do we feel we have had the time to get to know the data challenges specific to our analyses either. We know that this is one of the major empirical tasks remaining for this paper. Furthermore, there are a large number of additional sets of variables we would like to test for consideration. One set that might be of particular interest is the number of co-partisans in the same district under the old system. It seems likely to us that LDP DMs who are the sole LDP 19

20 representative from their district will be more likely to be policy generalists than those with copartisans. District characteristics also seem likely to matter. Preliminary runs for example suggest that DMs from urban districts are more likely to be generalists and that this difference outweighs the fact that most urban districts have a greater number of seats and thus DMs theoretically could be elected by slightly narrower constituencies in terms of share of votes (theoretically 25% of the vote for a 3-seat district vs. 17% for a 5-seat district). Also, as has been well-demonstrated in the literature, DMs adopted various strategies in running under SNTV (e.g. Tatebayashi 2004). Preliminary analyses not reported in the text of this paper did not find a strong relationship in policy specialization (using our measures) and vote concentration within a district, although this is something that we will need to pursue further. Although the results and write-up of this paper are much more preliminary than is usual for us, we hope that readers will take it as a sign of our eagerness for feedback and look forward to discussions of the issues addressed in this paper. 20

21 Table 1. House of Representative LDP Zoku Giin Comparison: Mid 1980s vs Zoku Sato-Matsuzaki (Mid 1980s) Inoguchi-Iwai (Mid 1980s) KNP (2004) Change Commerce Construction Agriculture Transportation Postal Education Health/Welfare Labour Finance Defense Foreign Affairs 16 n.a Fisheries Cabinet 25 n.a Local Affairs 14 n.a Science 13 n.a Legal Affairs 9 n.a Environment 5 n.a All High Policy 61 (21%) 35 (16%) 49 (24%) (0.12) All Distributional 154 (53%) 136 (62%) 100 (49%) (-0.10) All Public Goods 75 (26%) 50 (23%) 57 (28%) (0.10) Notes: In 1985 there were 254 LDP HoR LDP DMs, in 2004 there were 246 DMs. The KNP Zoku codings rely on the Zoku point system reported in the main text, any DM with 3+ points is counted as a Zoku member. 21

22 Table 2. LDP Seniority (Terms in Office) in 1985 and Term Number Cum % Number Cum % Total

23 Table 3a. Random Effects GLS Regression: SMD vs. SNTV Policy Specialization Model 1a. Effective Number of Types Model 1b. Effective Number of Areas New System SNTV Interaction New System SNTV Interacted Base/1 st Yr 1.32 (0.03)** (0.04)* Base/1 st Yr 1.56 (0.07)** (0.09)* 2 nd Year 0.31 (0.04)** (0.06)** 2 nd Year 0.58 (0.08)** (0.11)** 3 rd Year 0.47 (0.04)** (0.06)** 3 rd Year 1.12 (0.08)** (0.11)** 4 th Year 0.56 (0.04)** (0.06)** 4 th Year 1.43 (0.09)** (0.12)** 5 th Year 0.66 (0.05)** (0.06) 5 th Year 1.80 (0.09)** (0.12)** 6 th Year 0.77 (0.05)** (0.06) 6 th Year 2.10 (0.09)** (0.12)** 7 th Year 0.81 (0.05)** 0.07 (0.07) 7 th Year 2.25 (0.09)** (0.13)** 8 th Year 0.84 (0.05)** 0.03 (0.07) 8 th Year 2.34 (0.10)** (0.14)* 9 th Year 0.86 (0.05)** 0.04 (0.07) 9 th Year 2.41 (0.10)** (0.14) 10 th Year 0.88 (0.05)** 0.05 (0.08) 10 th Year 2.47 (0.10)** (0.15) 11 th Year 0.86 (0.06)** 0.08 (0.09) 11 th Year 2.42 (0.11)** 0.11 (0.17) 12 th Year 0.86 (0.06)** 0.07 (0.09) 12 th Year 2.53 (0.11)** 0.07 (0.17) 13 th Year 0.88 (0.06)** 0.09 (0.11) 13 th Year 2.57 (0.11)** 0.13 (0.20) 14 th Year 0.89 (0.06)** 0.05 (0.11) 14 th Year 2.63 (0.12)** 0.14 (0.22) 15 th Year 0.93 (0.06)** 0.05 (0.13) 15 th Year 2.76 (0.12)** 0.05 (0.25) 16 th Year 0.94 (0.06)** 0.14 (0.14) 16 th Year 2.81 (0.12)** 0.18 (0.26) N: 2830 N: 2830 R 2 :.26 R 2 :.39 Model 2a. Effective Number of Types Model 2b. Effective Number of Areas New System SNTV Interaction New System SNTV Interaction Base/1 st Yr 1.35 (0.17)** (0.14) Base/1 st Yr 1.76 (0.32) (0.26) 2 nd Year 0.36 (0.05)** (0.07)** 2 nd Year 0.70 (0.09) (0.13)** 3 rd Year 0.51 (0.05)** (0.07)** 3 rd Year 1.23 (0.09) (0.13)** 4 th Year 0.58 (0.05)** (0.07)* 4 th Year 1.46 (0.09) (0.14)** 5 th Year 0.69 (0.05)** (0.07) 5 th Year 1.83 (0.10) (0.13)** 6 th Year 0.78 (0.05)** (0.08) 6 th Year 2.11 (0.10) (0.15)** 7 th Year 0.80 (0.05)** (0.08) 7 th Year 2.18 (0.10) (0.15)** 8 th Year 0.85 (0.05)** (0.08) 8 th Year 2.28 (0.10) (0.16)** 9 th Year 0.86 (0.05)** (0.09) 9 th Year 2.35 (0.11) (0.17)** 10 th Year 0.86 (0.05)** (0.09) 10 th Year 2.35 (0.11) (0.18) 11 th Year 0.81 (0.06)** (0.1) 11 th Year 2.22 (0.11) (0.19) 12 th Year 0.82 (0.06)** (0.11) 12 th Year 2.33 (0.12) (0.20) 13 th Year 0.82 (0.06)** (0.12) 13 th Year 2.36 (0.11) (0.23) 14 th Year 0.84 (0.07)** (0.13)** 14 th Year 2.43 (0.13) (0.25) 15 th Year 0.84 (0.07)** (0.15) 15 th Year 2.51 (0.13) (0.28) 16 th Year 0.85 (0.06)** (0.15) 16 th Year 2.53 (0.13) (0.30) N: 2830 N: 2830 R 2 :.31 R 2 :.44 23

24 Table 3b. Random Effects GLS Regression: SMD vs. SNTV Policy Specialization Model 3a. Effective Number of Types Model 3b. Effective Number of Areas New System SNTV Interaction New System SNTV Interaction Transition n.a (0.03)** Transition n.a (0.06)** Year-Trend 0.02 (0.004)** (0.01)** Year-Trend 0.05 (0.01)** (0.01)* Base/1 st Yr 1.14 (0.09)** (0.10)* Base/1 st Yr 0.65 (0.17)** 0.40 (0.18)** 2 nd Year 0.32 (0.04)** (0.06)** 2 nd Year 0.62 (0.08)** (0.11)** 3 rd Year 0.48 (0.04)** (0.06)** 3 rd Year 1.10 (0.08)** (0.11)** 4 th Year 0.58 (0.04)** (0.06)** 4 th Year 1.45 (0.09)** (0.12)** 5 th Year 0.67 (0.05)** (0.07)** 5 th Year 1.77 (0.09)** (0.13)** 6 th Year 0.77 (0.05)** (0.07)** 6 th Year 2.03 (0.09)** (0.13)** 7 th Year 0.81 (0.05)** (0.07) 7 th Year 2.19 (0.09)** (0.14)** 8 th Year 0.83 (0.05)** (0.08)** 8 th Year 2.25 (0.10)** (0.15)** 9 th Year 0.85 (0.05)** (0.08)** 9 th Year 2.28 (0.10)** (0.16)** 10 th Year 0.86 (0.05)** (0.09)** 10 th Year 2.30 (0.10)** (0.17)* 11 th Year 0.84 (0.06)** (0.10)* 11 th Year 2.26 (0.11)** (0.19) 12 th Year 0.83 (0.06)** (0.10)** 12 th Year 2.33 (0.11)** (0.20) 13 th Year 0.83 (0.06)** (0.12)* 13 th Year 2.37 (0.11)** (0.23) 14 th Year 0.83 (0.06)** (0.13)** 14 th Year 2.40 (0.12)** (0.24) 15 th Year 0.86 (0.06)** (0.14)** 15 th Year 2.49 (0.12)** (0.28) 16 th Year 0.85 (0.06)** (0.15) 16 th Year 2.50 (0.13)** (0.29) N: 2830 N: 2830 R 2 :.29 R 2 :.43 24

25 Figure 1. LDP Seniority and Post Allocation Pre-Reform LDP HoR Post Allocation by Years in Office LDP Comm Vice-Chair Vice-Minister LDP Comm Chair HoR Comm Chair Post-Reform LDP HoR Post Allocation by Years in Office LDP Comm Vice-Chair Vice-Minister LDP Comm Chair HoR Comm Chair

26 Figure 2. Policy Specialization over LDP DMs First Twelve Years Effective Number of Policy Areas SNTV SMD PRONLY Effective Number of Policy Types SNTV SMD PRONLY