Online Appendices on“Internalization of Externalities and Local Government Consolidation: Empirical Evidence from Japan”
Takeshi Miyazaki
Department of Economics, Kyushu University, Fukuoka, Japan
6-19-1 Hakozaki Higashi-ku Fukuoka 812-8581, Japan
Tel.: +81-(0)92-642-4272, Fax: +81-(0)92-642-4272,
E-mail:
Online Appendix 1. Alternative forms or estimations of demand functions
Table Online Appendix 1 presents the results of estimation of public goods demandusingan alternative sample or alternative demand models. As shown in the table, parameter estimates for the sample with populations of less than 150,000 and for the camaraderie and exploitation hypothesis models demonstrate that the estimates of the congestion parameter are lower than those of the baseline and yet remain statistically significant. The production function parameters are statistically positive, although with relatively small point estimates for the camaraderie function. Although still significant, in the camaraderie model the size of the spatial autocorrelation largely differs from that of the baseline, probably because the level value of population in Eq. (10), , strongly affects the spatial lag estimate. However, these coefficients are highly significant, and, more importantly for the calculation of spillover, the publicness and production function parameters remain significant with almost the same coefficient sizes.
Table Online Appendix 1. Estimation of Public Goods Demand under Alternative Sample or Functional Forms
Variable / Population < 150,000 / Camaraderie effect model / Exploitation hypothesis model(1) / (2) / (3)
L(MEDSHARE) / / 0.157*** / 0.019 / 0.271***
(0.046) / (0.049) / (0.050)
L(WAGE) / / 0.174*** / 0.096* / 0.293***
(0.054) / (0.050) / (0.053)
L(MEDINC) / / 0.622*** / 0.594*** / 0.701***
(0.049) / (0.054) / (0.051)
L(POP) / / -0.076 / -0.270*** / -0.735***
(0.049) / (0.053) / (0.072)
L(POPDEN) / -0.122*** / -0.121*** / -0.126***
(0.006) / (0.006) / (0.006)
POP65 / 0.010*** / 0.012*** / 0.012***
(0.001) / (0.001) / (0.001)
FOREIGNER / 0.034*** / 0.037*** / -0.013**
(0.007) / (0.006) / (0.007)
POP / 0.858***
(0.066)
L(POPDAY) / 0.804***
(0.048)
Parameters
Congestion / / 0.799*** / 0.716*** / 0.841***
P-value / [0.000] / [0.000] / [0.000]
Production function / / 0.150*** / 0.095* / 0.231***
P-value / [0.000] / [0.058] / [0.000]
Spatial lag parameter / / 0.951*** / 0.207***
(0.014) / (0.004)
Spatial error parameter / 0.226*** / 0.933***
(0.004) / (0.027)
Table Online Appendix 1 (Continued)
Variable / Population < 150,000 / Camaraderie model / Exploitation hypothesis model(1) / (2) / (3)
Specification test of spatial lag and error
Spatial lag:
Robust Lagrange multiplier test / 182.68 / 176.29
P-value / [0.000] / [0.000]
Spatial error:
Moran's I / 40.032 / 41.429
P-value / [0.000] / [0.000]
Robust Lagrange multiplier test / 961.584 / 1047.413
P-value / [0.000] / [0.000]
Likelihood ratio test H0: coeff = 0 / 1022.592
Observations / 3019 / 3143 / 3143
Log pseudo likelihood / 189.848 / 459.994
Adjusted R2 / 0.795
Notes: Robust standard errors are in parentheses. * is significant at 10%; ** is significant at 5%; *** is significant at 1%. P-values are in brackets. Variable definitions and units are listed in Table 1. Based on the results of specification tests for spatial lag and error, we adopt the spatial autoregressive model with spatial autocorrelation in disturbances (SARAR). "Congestion" and "Cobb-Douglas function" parameters are calculated from the estimates, according to the definition stated under Eq. (9) in the text.
Table Online Appendix 2 Estimation of Voting Behavior Using Alternative Public Goods Demand Functions
Variable / Demand function for municipalities with population <150,000 / Camaraderie congestion function case / Exploitation hypothesis case(1) / (2) / (3)
ATS / 16.829** / 0.003** / 19.926**
(7.858) / (0.001) / (9.243)
ΔCOST / 0.011** / 0.012** / 0.011**
(0.005) / (0.005) / (0.005)
ΔPOP / 0.00001 / 0.00001** / 0.00001
(0.00001) / (0.00001) / (0.00001)
ΔTAXINC / -0.010 / -0.009 / -0.010
(0.008) / (0.008) / (0.008)
SHAREPOP / -0.426* / -0.496** / -0.426*
(0.225) / (0.221) / (0.224)
(SHAREPOP)2 / 0.006** / 0.006** / 0.006**
(0.003) / (0.003) / (0.003)
ΔMEDINC / -0.023* / -0.022* / -0.022*
(0.012) / (0.012) / (0.012)
ΔEDUC / 0.877 / 0.793 / 0.879
(0.903) / (0.908) / (0.903)
ΔPOPDEN / 2.030 / 1.721 / 2.035
(2.256) / (2.304) / (2.254)
ΔDEBT / 0.004 / 0.004 / 0.004
(0.005) / (0.005) / (0.005)
ΔSPEC_GRANT / 0.028 / 0.025 / 0.028
(0.039) / (0.039) / (0.039)
ΔUNCON_GRANT / -0.039*** / -0.045*** / -0.039***
(0.015) / (0.014) / (0.015)
Observations / 308 / 308 / 308
Adjusted R2 / 0.159 / 0.156 / 0.159
Notes: * is significant at 10%, ** is significant at 5%, and *** is significant at 1%. Standard errors are adjusted by a two-step procedure proposed by Murphy and Topel (1995) and are described in parentheses. All regressions contain year and prefecture dummies. Column (1) uses the estimates of the demand function using municipalities with population under 150000 to calculate ATS, column (2) uses those of the demand function assuming camaraderie congestion, column (3) uses the estimates of the demand function based on the exploitation hypothesis. Variable definitions and units are described in Table 1.
Table Online Appendix 2 presents the regression results using alternative formulations of the demand function in the first-step estimation. Column (1) employs the same specification as the baseline model, except that the sample in the first-step estimation is limited to municipalities with a population of less than 150,000. Column (2) presents the estimates obtained using the camaraderie effect model, as in Eq. (10), whereas column (3) adopts the exploitation hypothesis as the formulation of public service spillovers, as in Eq. (11). As shown inthe table,the is significant at 5% significance level, and in columns (1) and (3), the sizes of the coefficients are almost equal to those of the baseline shown in Table 6. The of the camaraderie effect model is extremely low, primarily because in contrast to the cases using the baseline demand functions, where the is composed of the logs of per capita expenditure, wage, and population (shown in Eq. (12)), theunder the camaraderie function is computed using the level value of population as well, which,on average,considerably amplifies the magnitudes ofthe. In this case, we are unable to interpret the coefficient of the as a sort of elasticity of approval rates with regard to degree of spillovers. Nevertheless, these results seem to support the findings that spillovers among potential consolidation municipalities have positive impacts on consolidation preference. As for other covariates, , , , and are again significant and have the same signs as before.
Online Appendix 2.Robustness check
Table Online Appendix 3 provides the results of the robustness check.[1] Column (1) exploits the efficiency gains computed from cost parameters estimated after including sociodemographic variables as controls (see column (2) of Table 4), because municipal expenditures are likely to depend on sociodemographic and economic factors. Column (2) includes political proximity variables as controls to deal with the effects of political closeness among the mayors of merging municipalities, because political proximity between mayors could be one of the determinants for municipal consolidation. To control for political effects, we create five political proximity variables owing to the presence of five major political parties in Japan: Social Democratic Party, Komeito, Japanese Communist Party, Liberal Democratic Party of Japan, and Democratic Party of Japan. Measurements and units of the political proximity variables are provided in Panel B of Table 1. In column (3), year dummies are omitted from the explanatory variables to examine another functional form with respect to year of consolidation.
Next, we consider the effects of the consolidation pattern and objective of the referendum. There are two types of municipal consolidation in Japan: i) to create a new municipality by merging several surrounding ones equally, and ii) to allow a large, central municipality to absorb other municipalities and play a dominantrole in the merged municipality. To examine this feature, column (4) includes as a control, which takes the value one when a new municipality is created through consolidation. For our data, the objective of the referendum is twofold: i) to decide whether to consolidate counterparts, and ii) to decide whether to institute a merger consultation committee, which associated municipalities have to create prior to consolidation to discuss issues about consolidation. In column (5), this aspect is examined by including , which takes one if the referendum question is whether to consolidate, and zero if the question is whether to establish the committee.
Table Online Appendix 3 Robustness Check for Estimates of Voting Behavior
Variable / Scale economy effects with sociodemographic variables / Including political proximity / Excluding year dummy / Including dummy for creating municipality / Including dummy for consolidation(1) / (2) / (3) / (4) / (5)
ATS / 18.260** / 18.101** / 16.339* / 17.171* / 18.404**
(8.977) / (9.068) / (8.646) / (9.080) / (9.019)
ΔCOST / 0.010** / 0.010** / 0.011** / 0.012** / 0.011**
(0.005) / (0.005) / (0.005) / (0.005) / (0.005)
SHAREPOP / -0.452** / -0.443* / -0.433* / -0.432* / -0.420*
(0.219) / (0.231) / (0.224) / (0.224) / (0.226)
(SHAREPOP)2 / 0.006** / 0.006** / 0.006** / 0.006** / 0.006**
(0.003) / (0.003) / (0.003) / (0.003) / (0.003)
ΔMEDINC / -0.023* / -0.022* / -0.022* / -0.022* / -0.023*
(0.012) / (0.012) / (0.012) / (0.012) / (0.012)
ΔUNCON_GRANT / -0.040*** / -0.040** / -0.040*** / -0.038** / -0.039**
(0.015) / (0.015) / (0.015) / (0.015) / (0.015)
ΔPOLIT_SDP / -8.747
(8.062)
ΔPOLIT_KM / -3.391
(6.429)
ΔPOLIT_JCP / 3.604
(6.147)
ΔPOLIT_LDP / -1.653
(5.716)
ΔPOLIT_DPJ / 8.737
(7.492)
NEWCREAT / 2.582
(2.210)
CONSOLIDATION / 1.898
(2.466)
Observations / 308 / 308 / 308 / 308 / 308
Adjusted R2 / 0.158 / 0.148 / 0.157 / 0.159 / 0.157
Notes: Robust standard errors adjusted by a two-step procedure proposed by Murphy and Topel (1995) are in parentheses. * is significant at 10%, ** is significant at 5%, and *** is significant at 1%. Regression (1) exploits scale economy effects computed from the fitted values of the cost function with sociodemographic variables. Column (2) includes political proximity variables as controls. All regressions except column (3) contain year and prefecture dummies. Estimates of ΔEDUC, ΔPOPDEN, ΔDEBT, and ΔSPEC_GRANT are omitted. NEWCREAT is a dummy that takes one when a new municipality is created through consolidation, and CONSOLIDATION is a dummy that takes one if the referendum question is whether to merge and zero if it is whether to establish a merger consultation committee.
The table displays that the coefficients of , , , and do not change much in all the abovementioned specifications. They are statistically significant and have the expected signs. Besides, and remain significant, their point estimates being almost equal to those obtained above. The political proxy variables in column (2) are all statistically insignificant, thus indicating that closeness in political convictiondoes not influence municipal consolidation. One explanation is thatin Japan,in fact,a number of mayors do not receive official approval or endorsement from a party while running for election, thereby pointing to the low variation in political proximity. As shown in columns (4) and (5), and are also not significant.
1
[1] Estimates of , , , , , and are omitted from the table because they are not significant in the current regressions as inTables 6 and7.