Does reapportionment in a legislature affect policy

Transcription

1 Reapportionment and Redistribution: Consequences of Electoral Reform in Japan Yusaku Horiuchi Jun Saito National University of Singapore Yale University Does reapportionment in a legislature affect policy outcomes? We examine this question from a comparative perspective by focusing on reapportionment associated with the electoral reform in Japan. First, we show that the reform of 1994 resulted in an unprecedented degree of equalization in legislative representation. Second, using municipal-level data, we present evidence that municipalities in overrepresented districts received significantly more subsidies per capita, as compared to those in underrepresented districts, in both prereform and postreform years. Third, by examining the relationship between the change in the number of seats per capita and the change in the amount of subsidies per capita at the municipal level, we show that the equalization in voting strength resulted in an equalization of total transfers per person. Does reapportionment in a legislature affect policy outcomes? 1 Since the United States Supreme Court established the principle of one-person, one-vote in the 1960s, a number of American legal scholars and political scientists have examined this question. 2 In this article, we examine the same question from a comparative perspective by focusing on reapportionment associated with the electoral reform in Japan. We argue that Japan s electoral reform of 1994 provides an important case for examining the policy consequences of reapportionment, because it approximates a natural experiment for the following reasons. First, since the reform reduced the overall level of malapportionment in a very short span of time, we can examine the effects of reapportionment on policy outcomes, while holding other social, economic, and demographic factors almost at a constant. In contrast, although the court-ordered re- districting in the United States literally embodied the oneperson, one-vote principle, the process of reapportionment was gradual and diffuse. Therefore, policy changes, if any, could have been influenced by a variety of factors other than reapportionment. Second, and more importantly, although the electoral reform reshaped Japan s Lower House districts throughout the nation, it kept intact other important institutional characteristics: rules electing politicians in other government organs (i.e., Upper House members, governors, mayors, and local legislators), constitutional provisions for budget compilation and intergovernmental relations, statutes concerning subsidy allocation, and so forth. In other democracies, an electoral reform in a particular legislative body is often accompanied by reforms in other institutions, for example, electoral and administrative reforms at subnational levels. In such cases, it is difficult to differentiate the Yusaku Horiuchi is Assistant Professor, Department of Political Science, National University of Singapore, Block AS1, Level 4, 11 Arts Link, Singapore (polyh@nus.edu.sg). Jun Saito is a Doctoral Candidate, Department of Political Science, Yale University, P.O. Box , New Haven, CT (jun.saito@yale.edu). He is also a member of the House of Representatives, Japan. The earlier drafts of this article were presented at Jisshō Seiji Bunseki Kenkyū-kai (Tokyo, June 2001), the 2001 Annual Meeting of the American Political Science Association (San Francisco, September 2001), and Political Science Department Seminar, National University of Singapore (Singapore, September 2001). We would like to thank the participants at these seminars and conference for their helpful comments and suggestions. Horiuchi also acknowledges the financial support of the Faculty of Arts and Social Sciences at National University of Singapore. All analyses and arguments in this article are of the authors, not of any institution to which the authors are affiliated. 1 In this article, reapportionment means both change in geographical boundaries of electoral districts (i.e., redistricting ) and change in the number of seats per geographical unit (i.e., narrowly-defined reapportionment ). 2 Following a series of cases beginning with Baker v. Carr in 1962, the U.S. Supreme Court ruled that it was unconstitutional to have unequal district populations in state legislatures. For a recent comprehensive study of reapportionment processes and consequences in the United States, see Ansolabehere, Gerber, and Snyder (2002) and Cox and Katz (2002). American Journal of Political Science, Vol. 47, No. 4, October 2003, Pp C 2003 by the Midwest Political Science Association ISSN

2 670 YUSAKU HORIUCHI AND JUN SAITO effects of a particular legislative equalization on policy outcomes. Using the methodologically ideal case of Japan, we give an answer to the empirical question raised earlier. Furthermore, through such an empirical inquiry, we intend to make contributions to two broader theoretical debates. The first is a persistent and nagging question for political scientists (Ansolabehere, Gerber, and Snyder 2002, 767): Does representation matter? Plaintiffs in suits, civil organizations, and the media often claim that representation bias causes policy bias, particularly in the distribution of public expenditure. The logic behind this claim is simple: if representatives have equal power in the legislative process, 3 people in rural areas, which tend to be overrepresented and have relatively more representatives within a legislature, receive an unfairly greater share of money. 4 Thus, it is claimed that equalization in legislative representation results in equalization in redistribution. Although this claim may sound highly intuitive, the corresponding empirical evidence has not been as obvious as one might think; a number of scholars have failed to find statistically significant policy effects of reapportionment decisions. In fact, after a few decades of research, scholars of American politics had reached a near consensus that the court-ordered redistricting in U.S. state legislatures had no impact on policy outcomes. 5 3 Bargaining among equally powerful individual legislators is often regarded as an ideal of representative democracy (e.g., Dahl 1998). This is also a widely shared view among political scientists who analyze and interpret legislative politics (Ansolabehere, Gerber, and Snyder 2002, 776). 4 More specifically, if all representatives are endowed with an equal ability and authority to elicit government resources, the malapportionment of seats inevitably creates inequalities in the per capita amount of government resources brought into each district. For example, if five representatives from five single-member districts divide $500, each representative brings $100 into his/her district, but the per capita amount is inevitably larger in districts with smaller populations. Atlas et al. (1995) presented an alternative logic for the same causal relationship. They argue that the smaller the number of constituents represented by politicians, the higher the marginal productivity of lobbying for public resources, thus, the greater the level of efforts made by politicians in seeking benefits. We consider this mechanism as supplementary in producing correlations between representation bias and policy bias. If politicians in overrepresented districts, as compared to those in underrepresented ones, put in more effort to bring resources back to their constituencies, policy bias in favor of the overrepresented areas is intensified. Yet, even when representatives put in the same level of effort and slice a pie equally, malapportionment inevitably produces, as explained, an unequal distribution of per capita benefits received by constituents. 5 Ansolabehere, Gerber, and Snyder (2002) extensively reviewed the literature on policy effects of reapportionment and argued that this Why did the extant literature fail to substantiate this intuitive expectation in the real world? One may argue that these results simply imply that representation does not matter in shaping distributive policies. Economists and scholars of public administration argue that socioeconomic and demographic characteristics of administrative units, as well as macrolevel economic and budgetary conditions, matter the most. Political scientists argue that politics matter, but not formal representation per se. They show that budgetary politics is more a function of political parties, electoral competition, fiscal centralization, divided government, and so on (e.g., Alt and Lowry 1994; Garand 1988; Hicks and Swank 1992; Levitt and Snyder 1995; McCubbins and Thies 1997; and Meyer and Naka 1998). Ansolabehere, Gerber, and Snyder (2002), however, present an alternative view. They argue that previous studies assessing policy consequences of the reapportionment revolution in the United States suffer from methodological problems which, in turn, lead to unclear conclusions. These commonly examine whether levels of spending (overall or on specific programs) changed after reapportionment. Ansolabehere, Gerber, and Snyder, however, argue that the proper test should examine whether the distribution of spending per capita changed as a result of changes in the distribution of seats per capita. Based on this approach and on very comprehensive county-level data, they found that reapportionment did, indeed, change redistribution quite significantly, even after the effects of other relevant political, economic, and demographic variables were controlled. We employ the same approach adopted by Ansolabehere, Gerber, and Snyder. Using municipal-level Japanese data, we examine determinants of the distribution of subsidies from the central government to municipalities (i.e., cities, wards, towns, and villages) and how the distribution changed after the electoral reform in Japan. These investigations show that the reapportionment associated with the electoral reform did, indeed, change the allocation of public resources in Japan. Using data from Japan, which is totally different from that of the United States, we examine the same question, take the same methodological approach, and reach the same conclusion as Ansolabehere, Gerber, and Snyder. Thus, our analysis from this comparative perspective advances our understanding of representative democracy: formal representation matters even in a country with very different political incentives, social characteristics, and cultural backgrounds from those of the United States. consensual view could be found in Carp and Stidham (1993, 370), Erikson (1973, 280), McCubbins and Schwartz (1988, 388), and Rosenberg (1991, ). Also see Cox and Katz (2002, 4).

3 REAPPORTIONMENT AND REDISTRIBUTION 671 The second debate is more specific to Japan: Did the 1994 electoral reform matter? The literature on the consequences of Japan s electoral reform is extensive, but almost all of it examines the impacts on political parties, factions, candidates, or voters (e.g., Cox, Rosenbluth, and Thies 1999; Gallagher 1998; McKean and Scheiner 2000). A limited number of scholars argue or speculate that the reform would not solve Japan s deep-rooted problems of money politics (Kinken Seiji) and pork-barrel politics (Rieki-yudo Seiji), thus, the reform, in itself, might not bring about any desirable policy change (e.g., Christensen 1996; Gallagher 1998; Huang 1996; and Natori 2002). This is the first systematic and empirical analysis of the policy consequences of Japan s electoral reform and, contrary to the dominant and pessimistic view, we show that the reform did, indeed, matter in shaping policy outcomes. We proceed in the following manner. The next section reviews the historical process of reapportionment decisions in Japan and shows that the electoral reform of 1994 resulted in an unprecedented degree of equalization in legislative representation. The third section discusses data and empirical methods. The fourth section introduces dependent and independent variables and discusses expected effects of the independent variables. The fifth section reports our empirical findings from multivariate regressions. The last section concludes our discussion. Reapportionment Decisions in Japan Ever since the first election under the new Constitution was held in 1947, inequalities in legislative representation steadily deteriorated until the mid-1970s. This was caused by rapid population movement from rural to urban areas, which was associated with rapid economic growth as well as the government s inability to rectify the problem. Although the Public Offices Election Law (Kōshoku Senkyo Hō) stipulates that a reapportionment of seats be made based on each national census, which takes place every five years (e.g., in 1990, 1995, and 2000), the government made only minimal efforts to reapportion the legislative seats. 6 We can trace the steady deterioration of inequalities in Lower House representation through a variety of measures. Table 1 lays out two indices of inequality in repre- 6 See Christensen (2002), Horiuchi and Saito (2001), and Meyer and Naka (1999) for political involvement in, and problems of, the reapportionment process in Japan. TABLE 1 Malapportionment of Seats in the Japanese Lower House Election Year Maximin Ratio LH Index (0.33) (0.006) (0.47) (0.022) (0.03) (0.000) (0.10) (0.007) (0.09) (0.010) (0.25) (0.011) (0.53) (0.026) 1967 (Minor 3.50 ( 0.05) (0.002) Reapportionment in 1964) (0.83) (0.011) (0.66) (0.010) 1976 (Minor 3.50 ( 1.49) ( 0.018) Reapportionment in 1975) (0.38) (0.003) (0.08) (0.001) (0.46) (0.006) 1986 (Minor 2.92 ( 1.48) ( 0.009) Reapportionment in 1986) (0.26) (0.012) 1993 (Minor 2.82 ( 0.36) ( 0.010) Reapportionment in 1992) 1996 (Electoral 2.32 ( 0.50) ( 0.053) Reform in 1994) (0.15) (0.002) Note: The LH index stands for the Loosemore-Hanby Index. The numbers in parentheses are differences from the last election. sentation. 7 The first index is the ratio of the maximum over the minimum number of seats per capita (the maximin ratio ). This measure is sensitive to the presence of outlying observations and does not accurately reflect the overall tendency of inequalities. Nevertheless, as in the United States, it is the most frequently cited measure in judicial decisions in Japan. The figure started off at 1.76 in 1947 and reached a record high of 4.99 in the 1972 election. The two earliest reapportionment decisions in The data sources for Table 1 are Jichishō (Various issues) and Kawato (n.d.). Note that we used the number of electors (tōjitsu yūkensha sū) instead of the district population (senkyoku jinkō) because the district populations from earlier years are unavailable.

4 672 YUSAKU HORIUCHI AND JUN SAITO and 1975 did not hurt any incumbent, because no existing districts suffered a cutback in the number of seats. These decisions only divided a few urban districts into two and increased the number of seats in several districts. As the total number of seats reached the capacity limit of the Diet Hall, it became necessary to reduce the number of seats in several districts. The 1986 reapportionment added one seat each to eight districts and subtracted one each from seven districts, without changing district boundaries. 8 The 1992 reapportionment adopted a similar process: nine districts gained one seat each and ten districts lost one each. Not surprisingly, these four reapportionment decisions, which merely trimmed the few extreme cases, failed to alleviate the overall level of inequality. This is evidenced by another measure of inequality in representation the Loosemore-Hanby (LH) index of electoral disproportionality (Taagepera and Shugart 1989). 9 The index takes the value of zero when seat allocation is perfectly proportional and approaches the value of one where the allocation of seats is more concentrated in fewer districts. The LH index is less frequently used than the maximin ratio, but it reflects the overall tendency for malapportionment more accurately because the index takes into account the level of malapportionment in all districts. Consider the two cases in which the maximin ratio showed a significant decline: one in 1976 and another in Table 1 shows that although the maximin ratio dropped from 4.99 to 3.50 in 1976 and from 4.41 to 2.92 in 1986, the LH index shows only a slight decline, from to in 1976 and from to in It was not until the 1994 electoral reform that reapportionment was conducted in a categorical manner. This reform introduced a combination of the 300 seats elected by the single-member district (SMD) plurality rule and another 200 seats elected by the proportional representation (PR) system from 11 regional blocs. 10 The reform was mainly intended to remove excessively personalized electioneering styles under the single nontransferable vote system (SNTV) with multimember districts (MMD) 8 There were only three cases that experienced minor redrawing of district boundaries. They are Wakayama First and Second, Ehime First and Third, and Ōita First and Second. 9 The LH index is defined as 0.5 s i p i,wheres i is the seat share and p i the population share of the ith district against the national total. See Samuels and Snyder (2001) for various measures of malapportionment of seats in legislatures. 10 The regional PR blocs are apportioned fairly equally; the maximin ratio was 1.09 and the LH index was 0.02 in Therefore, we neglect this portion and only examine the effects of the apportionment of SMD seats on policy outcomes. and to install party-centered competition. 11 One consequence, however, was that the reform brought about a sizable reduction in malapportionment. 12 Although the 1994 reapportionment did not fully achieve the goal of the one-person, one-vote principle, it significantly alleviated the overall level of malapportionment; the LH index dropped from to This dramatic change is better evidenced in Figure 1, which provides three scatter plots of pre- and postreapportionment voting weight. Voting weight is defined as the number of seats per million district-populations (subtracted from its national average). Each data point corresponds to the voting weight in approximately 3,300 municipalities in Japan. 13 The 45-degree line serves as the benchmark of no change, while the regression line summarizes the data. If a reapportionment decision strictly embodies the one-person, one-vote principle, every observation will be aligned on a horizontal regression line. Therefore, the more horizontal the regression line, the more significant is the equalization of legislative representation. Panels A and B show that reapportionment decisions in 1986 and 1992 did not alleviate inequalities in a sizable manner. 14 Most observations are concentrated along the 45-degree line. However, it is evident from Panel C that reapportionment associated with the electoral reform in 1994 led to a major reduction in inequality in representation. 15 Although the positive slope of the regression line shows that municipalities in previously overrepresented districts still maintained heavier voting weight under the new electoral system, the slope of the regression line is more horizontal than those in the two previous cases of reapportionment (i.e., 0.20 in Panel C, as opposed to 0.78 in Panel A and 0.80 in Panel B). 11 For detailed descriptions of the recent electoral reform in Japan and discussions of its causes and consequences, see Christensen (1994, 1996, 1998), Hrebenar (2002, ch. 1-2), McKean and Scheiner (2000), Otake (1998), Reed and Thies (2001a, 2001b), and Shiratori (1995). 12 Reapportionment and redistricting in 1994 were conducted in the following manner. After allocating one seat each to 47 prefectures, the remaining 253 SMD seats were allotted on the basis of each prefecture s population. The district borders were then drawn for each prefecture. 13 Municipalities in a given electoral district have the same degree of voting weight in a given year. We will explain in a later section how administrative municipalities and electoral districts are related. 14 Reapportionment in 1986 and 1992 entailed transfers of a few municipalities across district borders. Outlying municipalities near the center of each panel reflect these cases. 15 Note that the voting weight for 1995 in Panel C is calculated on the basis of the prereform multimember districts.

5 REAPPORTIONMENT AND REDISTRIBUTION 673 FIGURE 1 Comparison of Recent Reapportionment Decisions in Japan Voting weight in Panel A Voting weight in 1985 Voting weight in Panel B Voting weight in 1992 Voting weight in Panel C Voting weight in 1995 Note: Voting weight is defined as the number of seats per million district-populations subtracted from its national average. The solid lines indicate regression lines, while the broken lines indicate the 45-degree lines (i.e., no-change baselines). In sum, reapportionment in 1994 was considerably different from that of previous years. The reform not only drew new district borders but also alleviated the overall level of malapportionment. In the following sections, using the minor equalization in 1992 for contrast, we focus on the major equalization in 1994 and examine whether the electoral reform did, indeed, affect policy outcomes. Data and Methods Despite the fact that the malapportionment of seats in the Lower House has long been considered a serious problem in Japan, there exists only a limited number of empirical studies that examine how malapportionment affects intergovernmental transfers. What is more notable is that these studies found inconsistent results. Some of them found indeterminate effects (Yoshino and Yoshida 1988) or no significant effects (Kikuchi 1989) of malapportionment on fiscal transfers, but others found significant results (Meyer and Naka 1999; Onizuka 1997). As with the American case, we consider that this lack of consistency derives from inappropriate data and methods. The main problem is that literally all of the existing studies in Japan used data aggregated at the prefecture level. Prefecture-level data is very easy to access and is convenient for examining the relationships between various types of political and economic variables. Nevertheless, with only 47 observations, the estimated effects suffer from a problem of inefficiency. More importantly, Lower House seats are malapportioned not only across prefectures but also within prefectures. For example, the largest within-prefecture maximin ratio was 3.75 in 1972 (Chiba), 2.47 in 1993 (Tokyo), and 1.74 in 2000 (Aichi). Therefore, studies using prefecture-level data average away analytically relevant pieces of information that could, eventually, lead to biased inferences. Our analysis improves upon existing studies by using municipal-level data. In Japan, there are roughly over 3,300 municipalities (shi, ku, chō, and son) within 47 prefectures (to, dō, fu, and ken). Electoral districts, typically, include multiple municipalities, and there are only a few municipalities that belong to multiple electoral districts. In other words, electoral districts under both the old SNTV-MMD and the current SMD rules respect municipality borders. 16 Methodologically, this is very important as municipalities provide a stable frame of reference within which we can compare intertemporal differences of subsidy allocation under different districting schemes. 17 Another problem with existing studies is that they examine the relationship between the apportionment of seats and subsidy allocation across prefectures in some particular years. To put it in another way, they did not examine how the change in apportionment (i.e., reapportionment) affects the change in redistribution. To date, no 16 Under the old SNTV rule, there were only four cases where Lower House district borders split a single municipality into more than two districts: Kōriyama-shi, Chiba-shi, Chuō-ku of Ōsaka, and Okayama-shi. Districting under the new electoral rule continues to respect municipality borders. 17 It should be also noted that the geographic borders of municipalities were fairly stable for the period under investigation.

6 674 YUSAKU HORIUCHI AND JUN SAITO scholar has ever examined the consequences of not only the minor reapportionments but also the major reapportionment attributable to the 1994 electoral reform, which was, as we have argued, the single most drastic equalization of legislative representation in postwar Japanese history. Our article is the first, among the available literature, that investigates the effects of reapportionment in Japan. We consider that examining how the change in representation affects the change in redistribution, using differenced cross-section data, is a more direct and appropriate method for testing the effects of institutional change on policy outcomes. 18 It is also methodologically preferable because by taking the first-order differences, we can remove the effects of time-invariant municipalityspecific factors, which may lead to omitted variable biases in conventional cross-sectional regressions. As a matter of fact, such location-specific factors are substantively important in Japan. For instance, the allocation of subsidies is affected by the presence of power plants, national defense bases, oil stockpiling facilities, and so on. Let us now introduce our data and methods more specifically. To begin with, we conduct cross-sectional OLS regressions using data from four fiscal years before (i.e., FY1991 to FY1994) and after (i.e., FY1995 to FY1998) the electoral reform. Having acknowledged some of the methodological problems with this approach, we employ this cross-sectional estimation for a first-cut analysis. The purpose is to find out if the estimated effects of the apportionment of seats on the allocation of money changed after the electoral reform. Then, more importantly, we examine the covariates of the changes in subsidies per capita. When choosing pairs of fiscal years, we considered the following two criteria. First, the most recent Lower House election must have been held prior to the beginning of the budget cycle in the summer. 19 Second, no election must be held during that fiscal year. When these conditions are satisfied, the budgets, including supplementary ones, are approved by politicians elected in the most recent Lower House election and should be free from any outright influence by politicians who are to be elected in the next election. The 18 As we introduced earlier, Ansolabehere, Gerber, and Snyder (2002) first adopted this approach to examine policy consequences of the reapportionment revolution in U.S. state legislatures. 19 Japan s fiscal year begins on April 1 and ends on March 31. All government ministries start drafting budget proposals in the preceding summer. The cabinet approves the ministerial proposal at the end of December. After a series of deliberations in the Diet, the members of both Houses vote on the budget proposal, usually by the end of March. After the budget is approved, one or two supplementary budgets are approved during the fiscal year. For details of the budgetary process in Japan, see Ishi (1996). fiscal years that satisfy these two conditions include 1992, 1995, and The most recent Lower House elections held nearest to these fiscal years were those in February 1990, June 1993, and October Minor reapportionment was carried out in 1992 and major reapportionment (i.e., the electoral reform) in Therefore, the first pair of fiscal years, FY1995-FY1992, includes the minor reapportionment decision, while the second pair, FY1998- FY1995, includes the major reapportionment decision. 20 By comparing the regression estimates of these two cases, we intend to see if the effect of reapportionment differs according to the magnitude of reapportionment decisions. Dependent and Independent Variables Let us clarify our operationalization and definition of the dependent and independent variables. For our dependent variables, we use the per capita amount of total transfers from the central government to municipal governments (in logs). The total transfers are defined as the sum of general (chihō kōfuzei kōfukin) and specific (kokko shishutsukin) subsidies. 21 They include both formulaic and nonformulaic portions for a wide range of programs (e.g., construction, health and welfare, education, and disaster restoration). As with Ansolabehere, Gerber, and Snyder (2002), we prefer to use the total transfers rather than program-specific or type-specific (i.e., either formulaic or nonformulaic) transfers for the following reasons. First, the program breakdown of the subsidies for small towns and villages is simply not readily available. More importantly, when we use program-specific or typespecific subsidies as our dependent variable, we may not satisfactorily measure the overall political effects of reapportionment. This is because a particular project is often financed by various pockets. For example, a public home for the aged (rōjin hōmu) may be financed by both 20 One may argue that it is inappropriate to use the 1995 data as prereform data, because the first redistricting plan under the new SMD rule was published in November 1994, when the 1995 budget was being compiled. We consider, however, that the allocation of subsidies in the 1995 budget was not influenced by the new districting scheme for the following reasons. First, when the 1995 budget was compiled and approved, the LDP had not decided which incumbent to field in the brand new single-member districts. Second, existing personal support groups were still present in accordance with the existing SNTV-MMD district borders. 21 To be more precise, we use the amount of subsidy reported in the account settlement (Chihō Zaisei Chōsa Kenkyū Kai, Various issues). They reflect both the main and supplementary budgets of each fiscal year. The municipality population is based on Kokudo Chiri Kyōkai (Various issues).

7 REAPPORTIONMENT AND REDISTRIBUTION 675 formulaic and nonformulaic portions under construction and welfare programs. Thus, estimating political influences on decomposed public transfers should suffer from measurement errors. The apportionment of seats, our key independent variable, is measured in terms of the number of seats divided by the district population (in logs). 22 As explained in the introductory section, we expect that this variable is positively correlated with the dependent variable. One may argue that subsidy allocation during prereform years may have partly influenced the process of equalizing legislative representation through the electoral reform in If such an effect exists, our regression estimates would suffer from an endogeneity bias. We consider that this argument is inapplicable, because we could find no evidence suggesting that the 1994 electoral reform in Japan was intended to alleviate the problem of politically biased resource transfers across municipalities. As we have explained, its primary goal was to remove excessively personalized electioneering styles under the SNTV-MMD system. Thus, the equalization of transfers, if any, should be regarded as an unintended consequence. Besides the number of representatives per capita, there are a variety of factors that may affect the geographical allocation of subsidies. As discussed earlier, formal representation may not be a significant factor when other relevant variables are appropriately controlled. Thus, to cope with omitted variable biases and to capture other formulaic and nonformulaic determinants of subsidy allocation, we included a set of control variables: LDP s seat share, municipality fiscal strength index (zaisei-ryoku shisū), population, income, dependent population, industrial structure, and municipality urbanness. The LDP s district seat share (i.e., the number of LDP Lower House members divided by the number of seats; Mizusaki n.d.) is intended to control the effect of partisanship (Meyer and Naka 1998). If the LDP piles up subsidies for its constituents even after providing the formulaic portion, the coefficient should be positive. On the other hand, the LDP may want to use subsidies to buy off marginal voters who had not previously voted for the party. Under this scenario, the coefficient should be negative. 23 The municipality fiscal strength index (Chihō Zaisei Chōsa Kenkyū Kai, Various issues) controls the effect of formulaic portions of transfers. Roughly speaking, it indicates the fraction of a municipality s fiscal demand that can be financed by its local taxes. 24 Since the central government developed this index for the purpose of determining the amount of general (i.e., formulaic) subsidies to each municipality, there is no doubt that the coefficient is negative and highly significant. The municipality population (Kokudo Chiri-in, Various issues a; in logs) and the municipality population density (Kokudo Chiri Kyōkai, Various issues; Kokudo Chiri-in, Various issues; in logs) are also intended to control the effects of formulaic portions, because the central government also considers the size and density of populations when allocating general subsidies. Previous literature suggests that both municipality population size and population density negatively affect the allocation of subsidies per capita. There are two more theoretical justifications for including the population size. The first is the economy of scale in administration. Existing studies have shown that municipality population size is negatively correlated with the cost of administration (Saito and Nakai 1995). Another justification is the asymmetry of lobbying resources. When Lower House representatives lobby for additional subsidies from the central government, mayors and members of the prefectural assembly typically pipeline construction projects. Since both the number of mayors and prefectural assembly members per capita is inversely related to the municipality population, we expect that smaller municipalities obtain more subsidies per capita. The taxable income per capita (Nippon Mākettingu Kyōiku Sentā, Variousissues) isexpectedto have anegative effect because some portions of the central-to-municipal subsidies are used explicitly to assist the poor and the unemployed. The share of dependent population also enters the formulaic calculation of subsidies (Sōmuchō Tōkeikyoku, Various issues). The larger the share of population under 14, the higher the demand for education spending. The same is true for the share of population over 65 with regard to welfare spending. 22 The number of seats and composition of municipalities in each electoral district are based on Mizusaki (n.d.). The district population is calculated as a total of the municipality population based on Kokudo Chiri Kyōkai (Various issues). If a municipality is divided into multiple electoral districts, the district population from this municipality is weighted by the number of electors in each district. 23 Note that our intention of using this variable is to avoid a possible omitted variable bias, rather than to test specific hypothesis about incumbency or partisanship. 24 More specifically, it is the average of the following ratio in the past three fiscal years: the amount of standard fiscal revenues (kijun zaisei shūnyū-gaku) to the amount of standard fiscal demands (kijun zaisei juyō-gaku). The indices for 23 special wards in Tokyo are unavailable because they do not receive general subsidies from the central government (thus, the central government does not need to report the indices for these wards). In our analysis, we estimated the values of this variable for these wards based on the average ratio noted above.

8 676 YUSAKU HORIUCHI AND JUN SAITO We also set controls for the industrial structure in each municipality. Specifically, we use the ratio of the number of persons employed in the agricultural sector against the total number of employed persons in each municipality (Sōmuchō Tōkeikyoku, Various issues). A similar variable is also included for the service sector. These two industrial sectors are expected to give positive effects for the following reasons. First, since agriculture and service are vulnerable to international competition, these sectors raise their demands for protective measures. Second, the base category for comparison is the manufacturing industry, in which the strength of unions prevented a penetration of the LDP s partisan support. The ratio of the population in Densely Inhabited Districts (DID) against the total population (Sōmuchō Tōkeikyoku, Various issues) is a measure of municipality urbanness. A large portion of construction expenditure is used for purchasing project sites, and the cost is typically higher in urban areas than in rural areas. Therefore, urbanness is expected to have a positive effect on the amount of subsidies. We also include a dummy variable for cities designated by ordinance (seirei shitei toshi). 25 This variable should be included as a control because these cities, as compared to other municipalities, have different legal, administrative, and budgetary relationships with the central government. With everything else constant, they are expected to receive more subsidies than other municipalities. One important characteristic of the period under investigation is the Kobe earthquake of January 17, Since then, the central government has provided a large amount of special subsidies to disaster-stricken municipalities. 26 We control this effect with a dummy variable. These municipalities, coded as one (i.e., disaster-stricken municipalities), comprise 10 cities and 10 towns in Hyōgo Prefecture, for which the Disaster Relief Act (Saigai Kyūjyo Hō) was invoked. Since census-related data is collected and published every five years, the data for noncensus years was linearly interpolated. Given the practical limitation that the 2000 census results are not yet available, the data for FY1998 was linearly extrapolated based on data from 1990 and There are twelve cities designated by ordinance: Sapporo, Sendai, Yokohama, Kawasaki, Chiba, Nagoya, Kyōto, Ōsaka, Kōbe, Hiroshima, Kita-Kyūshū, and Fukuoka. 26 The government started providing earthquake-related subsidies from FY Observations with nonsensical extrapolation were removed from the analysis. Since the number of cases removed is very small, their inclusion or exclusion does not affect the conclusion of our analysis. Results Our analysis of the relationship between legislative representation and fiscal transfers boils down to the following two questions. Do municipalities in overrepresented districts, in a given year, receive disproportionately more subsidies from the central government? Did the equalization of representation result in an equalization of transfers? We present our answers to these questions in turn. Table 2 shows the results of cross-sectional regressions using data from prereform years (i.e., FY1991 to FY1994). In all four regressions, the number of seats per capita had a significantly positive effect on the receipt of total transfers per capita. The elasticity estimates range between 0.08 and To use an example from the FY1991 sample, suppose that there are two municipalities that have almost identical demographic and political characteristics, except for the apportionment of seats. Our finding implies that if the first municipality is represented twice as much as the second municipality, the receipt of subsidies per capita increases by 11%, even after the formulaic portion of transfers is taken into consideration. Table 3 shows the results from postreform years (i.e., FY1995 to FY1998). The elasticity estimates of the number of seats per capita are not considerably different in FY1995 and FY1996 than from those in prereform years. After the first election under the new electoral system was held in October 1996, the estimates increased slightly (i.e., from 0.11 in FY1996 to 0.16 in FY1997 and FY1998). This increase, however, does not necessarily mean that the equalization in representation had some effect on resource transfers, because the steeper slope estimate could be observed even without any change in the total transfers per capita. 29 What is more important in Table 3 is that the effects of the number of seats per capita on the total transfers per capita are highly significant even after the reform. Tables 2 and 3 clearly suggest that municipalities in overrepresented districts received disproportionately more subsidies from the central government during both pre- and postreform periods. It should be emphasized that this finding is highly robust. In our preliminary analysis, we conducted a number of other regression analyses with different sets of control variables, but found that the number of seats per capita always has a positive and highly significant effect in all fiscal years. We also checked the robustness of our finding by changing the unit of 28 Since relevant variables are already log transformed, we can interpret the coefficients as partial elasticity estimates for these variables. 29 Note that the smaller the variance of an independent variable, the steeper the slope of a regression line, if all other variables including the dependent variable are held constant.

9 REAPPORTIONMENT AND REDISTRIBUTION 677 TABLE 2 Determinants of Total Transfers Per Capita in Prereform Years Independent Variables FY1991 FY1992 FY1993 FY1994 Seats Per Capita (log) (0.02) (0.02) (0.02) (0.03) LDP s Seat Share (0.03) (0.03) (0.03) (0.03) Municipality Fiscal Strength Index (0.08) (0.07) (0.10) (0.12) Municipality Population (log) (0.01) (0.01) (0.01) (0.01) Municipality Population Density (log) (0.01) (0.01) (0.01) (0.01) Taxable Income Per Capita (log) (0.05) (0.04) (0.05) (0.06) Age under 14 Municipality Population (0.35) (0.35) (0.37) (0.36) Ageover65 Municipality Population (0.21) (0.22) (0.22) (0.21) Agricultural Sector Workers Ratio (0.07) (0.07) (0.07) (0.07) Service Sector Workers Ratio (0.08) (0.08) (0.08) (0.08) DID Population Ratio (0.04) (0.04) (0.04) (0.04) Cities Designated by Ordinance (0.10) (0.09) (0.10) (0.10) Earthquake in Kōbe 0.33 (0.08) Constant (0.34) (0.31) (0.30) (0.37) Number of Observations 3,255 3,255 3,255 3,255 R Square Root of MSE Note: The numbers in parentheses are White-Huber robust standard errors. analysis to the district-level, but that also did not alter our conclusion. 30 Let us now briefly look at the estimates of our control variables in the cross-sectional regressions shown in Tables 2 and 3. Most of the control variables exhibit statistically significant effects in the expected directions. In addition, the sizes of these coefficients are fairly stable across sample years. The exceptions are taxable income per capita and the share of population below 14 years of age. These variables are insignificant in some fiscal years. 30 We employed similar sets of regressors in this district-level analysis except for population size in log, which was dropped due to perfect multicollinearity. Another important exception is the LDP s seat share, which shows some notable patterns. First of all, contrary to Doi and Seriya s (1997) analysis using prefectural data, 31 our regression estimates suggest that in most (but not all) years, the LDP s seat share had a negative effect on the subsidy allocation. This finding is not necessarily puzzling, as some existing studies based on prefectural data report similar results (e.g., Onizuka 1997; Nagamine 2001). As discussed earlier, we might interpret the negative estimate as a consequence of the LDP s efforts to buy off marginal voters. 31 Doi and Seriya did not consider the effect of malapportionment in their analysis. Thus, their estimates might suffer from an omitted variable bias. We should also note that their measure of the LDP s legislative strength was made up of the LDP s vote share.

10 678 YUSAKU HORIUCHI AND JUN SAITO TABLE 3 Determinants of Total Transfers Per Capita in Postreform Years Independent Variables FY1995 FY1996 FY1997 FY1998 Seats Per Capita (log) (0.03) (0.03) (0.03) (0.03) LDP s Seat Share (0.03) (0.03) (0.01) (0.01) Municipality Fiscal Strength Index (0.10) (0.10) (0.11) (0.11) Municipality Population (log) (0.01) (0.01) (0.01) (0.11) Municipality Population Density (log) (0.01) (0.01) (0.01) (0.01) Taxable Income Per Capita (log) (0.05) (0.05) (0.06) (0.05) Age under 14 Municipality Population (0.35) (0.33) (0.32) (0.29) Ageover65 Municipality Population (0.19) (0.17) (0.18) (0.17) Agricultural Sector Workers Ratio (0.07) (0.07) (0.08) (0.07) Service Sector Workers Ratio (0.08) (0.08) (0.08) (0.08) DID Population Ratio (0.04) (0.04) (0.03) (0.03) Cities Designated by Ordinance (0.10) (0.10) (0.11) (0.09) Earthquake in Kōbe (0.14) (0.13) (0.11) (0.07) Constant (0.38) (0.38) (0.43) (0.42) Number of Observations 3,255 3,254 3,250 3,249 R Square Root of MSE Note: The numbers in parentheses are White-Huber robust standard errors. What is puzzling is that after FY1994, the magnitude of the coefficient dropped and the effect became statistically insignificant. There are several possible interpretations for these changes. They may reflect the dichotomous nature of LDP representation in single-member districts, vis-à-vis semiproportionality under the SNTV- MMD rule. 32 They may relate to the fact that regressions after FY1994 include the earthquake dummy. 33 Alterna- 33 We conducted additional cross-sectional regressions for FY1994 to FY1998 without the earthquake dummy. By excluding this varitively, the LDP may have, indeed, lost their power to allocate public resources to their constituencies after Japan entered into an era of coalition governments after Since arguing this point would take us beyond the scope of this article, we leave the further examination of this important puzzle to our future research. Now, our first-cut analysis with the cross-sectional regressions shows that the number of seats per capita is 32 Cox (1991) shows that the SNTV-MMD rule and the d Hondt PR rule are equivalent. able, the LDP s seat share became significantly negative in FY1995 and FY1996, but it is still insignificant in FY1994 (negative), FY1997 (positive), and FY1998 (positive). Note that even without the earthquake dummy, the effects of our main independent variable (i.e., the number of seats per capita) are highly significant and positive.

11 REAPPORTIONMENT AND REDISTRIBUTION 679 TABLE 4 Effect of Reapportionment on Change in Total Transfers Per Capita Independent Variables FY1995-FY1992 FY1998-FY1995 Seats Per Capita (log) (0.02) (0.01) Seats Per Capita (log) in (0.02) Seats Per Capita (log) in (0.02) (0.01) Seats Per Capita (log) in (0.02) LDP s Seat Share (0.01) (0.01) (0.01) (0.01) Municipality Fiscal Strength Index (0.24) (0.24) (0.33) (0.34) Municipality Population (log) (0.18) (0.18) (0.20) (0.21) Taxable Income Per Capita (log) (0.04) (0.04) (0.05) (0.05) Age under 14 Municipality Population (0.51) (0.52) (0.51) (0.51) Ageover65 Municipality Population (0.58) (0.57) (0.60) (0.58) Agricultural Sector Workers Ratio (0.23) (0.23) (0.20) (0.19) Service Sector Workers Ratio (0.33) (0.34) (0.26) (0.27) DID Population Ratio (0.10) (0.10) (0.05) (0.05) Earthquake in Kōbe (0.09) (0.09) (0.07) (0.07) Constant (0.02) (0.16) (0.02) (0.27) Number of Observations 3,255 3,255 3,249 3,249 R-square Square Root of MSE Note: The numbers in parentheses are White-Huber robust standard errors. positively correlated with the per capita fiscal transfers. It also shows that the elasticity estimates increased after the first election under the new electoral system. This change, as we have argued, does not necessarily indicate that there existed some substantive changes in redistribution. In order to fully examine the relationship between the change in legislative representation and the subsequent change in subsidy allocation, we conducted regressions with firstdifferenced data. 34 We also conducted additional first- 34 Since differenced municipality area size is almost always zero, we dropped the municipality population density from the list of differenced regressions by replacing the change in the seats per capita with the preequalization seats per capita and the postequalization seats per capita. The purpose of these regressions is to examine whether the apportionment of seats had parametrically different effects, between preand postequalization years, on the growth of per capita transfers. The results of these regressions are shown in Table 4. explanatory variables. We did not drop the earthquake dummy, although it remains constant across years, because we considered that the magnitude of its effect on the dependent variable is different depending on fiscal year.