Supporting Information for Representation and Redistribution in Comparative Perspective Tiberiu Dragu and Jonathan Rodden December 17, 2010
1 Data Below we list information regarding the source of our data for each country in the sample. Argentina: Grants data and gross provincial product are from the Ministry of Economy, Subsecretary of provinceal Programming, adjusted for inflation using the CPI developed by Sanguinetti and Tommasi (1997). Provincial population data are from the National Institute of Statistics and Census. Australia: Grants data are from Australian Bureau of Statistics, Government Finance Statistics State government series, adjusted for inflation using the CPI of the largest city in the state (produced by ABS). Gross state product and population data are from ABS state accounts. All data were obtained directly from the ABS. Brazil: Grants data were obtained directly from the Ministry of Finance: Inflation adjustment was conducted using the INPC deflator prepared by IBGE. Population and gross state product are from IBGE provinceal accounts. Canada: All data are from Statistics Canada, CANSIM series; fiscal data are deflated using provincial-level CPI. Germany: Grants data are from the Statistisches Bundesamt, accessed from http://www.statistikbund.de (no longer in service, replaced by www.destatis.de). Land-level GDP, population, and Land-specific deflators were provided directly by the Baden-Wuerttemberg Ministry of Finance. Mexico: All data are from INEGI: Grants are from Finance Statistics of the States and Municipalities, and population and GDP are from System of National Accounts: Internal Product by Federal Entity. Spain: Grants data are from Ministry of Economy and Finance, General Secretary of the Treasury, Budgets of the Autonomous Communities. GDP and population data are from National Instititute of Statistics, provinceal Accounts. Switzerland: All data are from Statistisches Lexikon der Schweiz. Accessed in September 2009 from http://www.bfs.admin.ch USA: Fiscal and population data were obtained directly from the Census Department. Fiscal data were adjusted for inflation with the national CPI produced by the U.S. Department of Commerce, Bureau of Economic Analysis (BEA). CPI and gross state product were obtained from the BEA web page: www.bea.gov. Our sample presents substantial variation on the relationship between representation and income: some federations clearly over-represent the poor, some over-represent the wealthy, and others display no pattern. Figure 1 demonstrates this by plotting each province s average seats per capita (expressed as a share of the national average) against average real provincial GDP per capita (also expressed as a share of the national average). The upper and lower chambers are displayed separately. The sample contains other important variation as well. First, by looking at the distribution of provinces along the vertical axis of Figure 1, one can see that these federations demonstrates considerable variation in the extent of malapportionment, as well as whether it exists in one legislative chamber or two. In addition, following the discussion above, Figure 2 plots legislative representa- 1
tion against logged population density, and demonstrates that while many of our federations are characterized by rural over-representation, others display no relationship, or even over-represent densely populated industrial provinces or city-states. 2 Robustness Checks In the first set of robustness checks, we focus exclusively on cross sectional variation. Our baseline model is a simple OLS regression which links the governmental grants a province obtains (in a given year) to malapportionment conditional on country fixed-effects and provinceal variables such as income, population, and province size, grants i N(α 0 + X i β + ɛ i, σɛ 2 ), (1) On this baseline model, we do two robustness check. Specifically, we estimate the interaction between malapportionment and provinceal population and between malapportionment and provinceal income. We further model the country-level of analysis (instead of estimating country fixed-effects) while controlling for country-level invariant factors to check whether the effect of malapportionment is robust to this test as well. For this analysis, the individual-level regression model is, grants i N(α 0 + α c[i] + X i β + ɛ i, σɛ 2 ) (2) And the country-level model is the following: α c N(T c τ, σc 2 ) (3) where c indexes the countries, T c is a matrix of country-level covariates, τ is the vector of coefficients for the country-level regression, and σ c is the standard deviation of the unexplained country-level errors. We include three country-level variables: presidential system, the number of province, and democracy age. The presidential system variable that takes the value 1 if the country is a presidential system and the value 0 otherwise. One could argue that each country s legislative institutions structure inter-provincial bargains in different ways. For example, legislative coalition-building among provincial representatives is most readily on display in countries like the United States, Brazil, Argentina, Mexico, and Switzerland, where the chief executive lacks the threat of a noconfidence procedure with which to force legislative cohesion among co-partisan legislators (Dier- 2
meier and Feddersen 1998). By contrast, parliamentary democracies are characterized by stronger and more cohesive parties in the legislature, where party leaders might determine how grants are distributed rather than a decentralized bargaining procedure in which each member tries to get as much as possible for the geographic area she represents. Second, we include a variable, democracy age, that takes the value of the year in which a country has become a democracy. The rationale for including this variable is that one could argue that countries that have had a shorter tradition as a democracy are more likely to be malapportioned. For example, researchers have noted that in Latin American countries before the transition to democracy, elites have attempted to give more representation to areas from which those elites garnered electoral support (Gibson 2004; Bruhn, Gallego, and Onorato 2008). Third, we include a variable, number of provinces, that takes the value of the number of provinces in a given federation. The rationale is that the number of bargaining parties may affect the bargaining process and consequently the distribution of grants. Table 1 presents the results of these robustness checks: Table 1 shows that the effect of malapportionment is robust to interaction effects between malapportionment and income and malapportionment and population (Model 1 and 2). It also suggests that it is robust to controlling for country-invariant variables (Model 3). We also provide several robustness checks on our cross sectional time series analysis. For this analysis, our baseline model estimate a multilevel regression with country and year fixed-effects. The individual level model is the following: grants i N(α 0 + α s[i] + X i β + ɛ i, σɛ 2 ) (4) The province-level model is the following: α s N(Y s γ, σs) 2 (5) where s indexes the provinces, Y s is a matrix of province-level covariates, γ is the vector of coefficients for the province-level regression, and σ s is the standard deviation of the unexplained province-level errors. On this baseline model, we do two robustness checks. We estimate the interaction between malapportionment and provinceal population and the interaction between malapportionment and provinceal income. Table 2 presents the results of these robustness checks: Table 2 suggests that the effect of malapportionment is robust to including interaction effects between malapportionment and provinceal income (Model 1) and between malapportionment and provinceal population (Model 2). Finally, we provide robustness checks on our baseline the multilevel analysis by modeling the 3
country-level of analysis while controlling for country-invariant variables that may affect the distribution of governmental funds. For these analysis, the individual level model is grants i N(α 0 + α s[i] + α c[i] + X i β + ɛ i, σɛ 2 ), (6) The province-level model is the following: α s N(Y s γ, σs) 2 (7) where s indexes the provinces, Y s is a matrix of province-level covariates, γ is the vector of coefficients for the province-level regression, and σ s is the standard deviation of the unexplained province-level errors. And the country-level of analysis is α c N(V c κ, σc 2 ) (8) where c indexes countries, V is a matrix of country-level covariates, κ is the vector of coefficients for the country-level regression, and σ c is the standard deviation of the unexplained country-level errors. Table 3 presents the results of this estimation: Table 3 suggests that the effect of malapportionment is robust to including country-invariant factors (Model 1) as well as to including interaction effects between malapportionment and provinceal income (Model 2) and between malapportionment and provinceal population (Model 3). Also, the effect of malapportionment is robust to interaction with whether the country is a presidential system or a parliamentary system (Model 4). 4