Controlling Spending: Electoral Competition, Polarization and Endogenous Platforms∗ Jakob Svensson† August 2005
Abstract Many countries have the following structure: society delegates power over public policy to policymakers via elections. The policymakers in turn delegate the execution of those policies to non-elected civil servants. The concern voters have is to elect policymakers who will adopt the right policy mix and to have civil servants execute those policies in the most cost efficient manner. The voters’ problem is that they typically only have an indirect instrument to deal with the tradeoff (elections), and that voters may disagree over what leaders to choose and what is the right policy mix. This paper shows that greater electoral polarization is associated with less cost efficient policies and lower public output, but that allowing citizens to first elect candidates (platforms) may improve the tradeoff in the face of greater polarization of society.
∗
I am grateful for comments and suggestions by Alberto Alesina, Allan Drazen, Aart Kraay, Per Krusell, Peter Norman, Stefan Palmqvist, Torsten Persson, David Strömberg and several seminar participants. † Institute for International Economic Studies, Stockholm University, Development Research Group, The World Bank, and CEPR. Email:
[email protected].
1
Introduction
Most countries have the following structure: society delegates power over public policy to policymakers via elections, who in turn delegate the execution of those policies to non-elected civil servants. Countries differ in to what extent these elections are free, and to what extent the voters can also determine the candidates running for office (parties’ platforms). The concern voters have is to elect leaders who will adopt the “right policy mix” and to have civil servants execute those policies in the most cost efficient manner. But if elected officials and bureaucrats have different objectives than the median voter, society is confronted with the problem of determining how to induce the right behavior of parties and civil servants. The problem is that they typically only have one instrument to do this - - elections. Determining how this can be done and what tradeoffs are involved requires careful modeling of the behavior of voters, elected officials and civil servants. Heterogeneity of voter preferences (electoral polarization) complicates the issue by creating political conflict over what leaders to elect and what is the right policy mix. However regardless of the policy mix, voters still want policies executed at the lowest possible cost. Thus, we need to study how polarization affects the effectiveness of elections as an instrument for generating the right mix of policy and the cost-efficient implementation of that policy. We analyze a model in which two civil servants (or public agencies, local governments) each controls one project. If finalized the projects provide benefits for the citizens. There are two parties (or candidates) with different preferences over the two projects. The winning party determines the allocation of public funds and thus which projects are in the end finalized. The voters want the projects implemented at the lowest possible cost, but assign different weights to the two projects. An election induces a competion among public agencies. A more cost efficient project increases the probability that a party with a favorable view to the agency’s activity will win the election, and thus the likelihood that
1
the project receives sufficient financing. If the electorate is homogeneous, in the sense that voters’ assign similar weights to the two projects, the election effect will be powerful - - voters will vote for the party that will only fund costefficent projects and the public agencies have strong incentives to keep costs low. If the electorate is highly polarized on the other hand, few voters are willing to switch party and vote for a candidate with different preferences over public spending than her own, although that party might be able to implement policies in the most cost efficient manner. In this scenario the election outcome is driven almost exclusively by the policy mix issue, not about how well these policies are implemented. The civil servants will act accordingly and inflate project costs. As a result, in polarized economies projects become more costly on average, and fewer projects are implemented. Primary election, or more generally the process to choose whom to run for office, provides the voters with an additional instrument to control spending. By electing more moderate candidates (in a relative sense) in a polarized society, the election induced incentive scheme is sharpened, since civil servants might then still find it beneficial to improve performance in order to ensure funding from both parties. Thus, allowing the citizen to elect candidates (platforms) will improve the tradeoff in the face of greater polarization of society. The contribution of this paper is fourfold.
First, we provide a model
in which the electoral process imposes discipline on government action even though voting is prospective (forward-looking) and the parties’ preferences are known. In the agency models initiated by Barro (1973) and Ferejohn (1986), all politicians are identical and politicians can be disciplined by elections only provided that voting is retrospective.1 Rogoff (1990) and Banks and Sundaram (1993) relax the assumption of retrospective voting. They find that the incumbent’s past performance is still an important statistics for the voters and that the electoral process can raise effort or otherwise induce less opportunistic 1 Persson, Roland and Tabellini (1997) show that separation of powers between the executive and the legislative bodies can further strengthen the disciplining role of elections.
2
behavior provided that past performance reveals information about the policymaker’s type or “competence”, and that competence is persistent (see also Harrington, 1993; Besley and Case, 1995). However, if voting is prospective and the policymakers’ platforms are known, the incumbent’s performance prior to the election should play no role. In our model the electoral process still imposes discipline on government action by inducing competition among parts of government. Second, we show that primary elections can be viewed as a strategic delegation mechanism. Members of the two parties will choose candidates whose preferences are systematically different from their own, and the difference depends on the underlying degree of polarization in society. In contrast to the models in Rogoff (1985), and Persson and Tabellini (1994); where strategic delegation arises because of a time-inconsistency problem, or Chari et al. (1997); where it arises because of a free-rider problem, it arises here because of problems caused by electoral polarization. Alesina and Rosenthal (2000) also analyze the choice of party platforms.2 As in our model the choice of platforms in the first stage is driven by the voter equilibrium in the second stage (general election). They show that if policy is the result of an executive-legislative compromise, parties choose more “extreme” platforms in order to pull the equilibrium policy towards their own ideal points. We highlight another mechanism in which voters use the policy platforms as instrument to constrain the bureucrats. Third, the paper extends the theoretical literature on polarization and economic policy outcomes. Persson and Svensson, 1989; and Tabellini and Alesina, 1990; develop models in which the incumbent can affect the state of the world inherited by his successor through its policy choice in the current period.3 Our setup differs in one key aspect from these models. In our model several com2 See Alesina and Rosenthal (2000) for additional references to the literature on endogenous platforms and policy (platforms) convergence. 3 See also Cukierman, Edwards, and Tabellini, 1992; and Svensson, 1998. Aghion and Bolton, 1990; Milesi-Ferretti, 1995; Milesi-Ferretti and Spolaore, 1994; Jonsson, 1997; argue that the incumbent can use a state variable (e.g., public debt) to influence the electoral outcome, by affecting the preferences of the electorate.
3
peting agents (bureaucrats), not one, constrain the future action space of the government. The resulting competition between these agents can be exploited by the electorate so as to induce more cost efficient policies. Finally, the paper also presents some new evidence on the effects of polarization, in particular due to ethnic divisions. Easterly and Levine (1997) show that ethnic division adversely affects growth through its impact on public policy choices, such as polices affecting human capital formation, the functioning of the foreign exchange markets, and the development of financial systems. We find that in addition, ethnic polarization is associated with both higher government spending and a less cost-efficient (productive) spending profile. The latter result is reminiscent of the findings in Alesina et al. (1999), although our empirical method is different. They show that the share of spending on “productive” public goods in U.S. cities are inversely related to degree of ethnic division. We show that the impact of a given spending level is conditional on the degree of polarization. An implication of our findings is that it is important to separate the effects of public projects (output) on welfare, from the effects of public spending on public output. In other words, public spending (even on potentially “productive areas” such as health) and output are not necessarily correlated.4 This paper is organized as follows. Section 2 presents some empirical facts on the relationship between spending, outcomes, and polarization as a motivation for the theoretical part. Section 3 presents the model. Section 4 derives the equilibrium public policy with exogenous platforms. The parties’ platforms are endogenized in section 5, while section 6 concludes. 4
Pritchett (1996), Filmer and Pritchett (1997), Reinikka and Svensson (2004) make a similar argument.
4
2
Some empirical patterns on spending, outcomes, and polarization
2.1
Data and specification
In this section we provide some empircial facts on the relathionship between spending, outcomes and polarization. Unfortunately, we cannot directly measure the degree of polarization or cost efficiency of public spending. Omitted variables and endogeneity concerns further complicate the causal interpretation of the findings below. Thus, this section will be suggestive in nature and should be viewed as a motivation for the theoretical work. We look at two empirical patterns: (i) the relationship between polarization and government spending; and (ii) the relationship between polarization and provision of public goods. To study these relationships we need data on degree of polarization (π), government spending (G), and public output (y). Below we discuss how we attempt to measure these three variables, and the empirical specification and method used. A common measure of government spending is the ratio of government expenditure to GDP. However, government expenditure data are generally not comprehensive, with coverage of public enterprises being particularly inadequate. Instead we choose to rely on a narrower but more precise indicator, government consumption as share of GDP. Real government consumption data are available from the Summers-Heston data set (denoted by spending). The consumption data are based on 1985 international prices which facilitates crosscountry comparisons over time. We follow Easterly and Levine (1997) and choose a measure of ethnic diversity (polarization) as proxy for polarization. A vast political science literature links ethnic groups with redistributive politics in developing countries, particularly in Africa. polarization measures the probability that two randomly
5
selected individuals in a country will belong to different ethnolinguistic groups. polarization increases with the number of groups and the more equal is the size of the groups.5 The data on real government consumption and polarization are sufficient to study the first pattern, i.e. the relationship between polarization and government spending. Specifically, we estimate the following regression Gjt = δ z Zjt + δ p Pjt + εjt ,
(1)
where Gjt is government spending in country j at time t, Pjt is the degree of polarization, and Zjt is a vector of other variables that determine the level of spending in country j at time t. δ p is a coefficient scalar, δ z is a coefficient vector, and εjt is a zero mean error term. Preferably, to look at the second pattern, i.e., the relationship between polarization and provision of public goods, data on the share of spending put into productive use, or amount of public goods actually supplied, are needed. However, such data are not directly available. What is observed is total spending (G) but not the amount (public) capital/goods actually created. These two variables are not necessarly correlated. The empirical growth literature is abundant with explicit (and implicit) attempts to separate productive spending from redistributive expenditures. Typically, this has been done by ex ante determining what items are likely to be productive. As expected, the results of these types of empirical exercises are mixed.6 Leaving measurement problems aside, the approach to ex ante separate productive from non-productive spending is problematic exactly for the same reason as mentioned above: spending and output are not necessarily correlated. 5
Ethnic fractionalization is obviously not a necessary, and much less a sufficient, condition for the existence of competing social groups. Consequently, we do not claim that polarization is a valid measure for all countries. 6 Easterly and Rebelo (1993) find that public investment overall has a very low impact on growth, but that certain types of investment expenditures are correleted with growth (see also Landau, 1986). Devarjan, Swaroop & Zhou (1996) find that the standard candidates for productive expenditures had either a negative or insignificant relationship with growth.
6
To circumvent this identification problem we use an alternative method. Intuitively, we let the data reveal if a given country’s spending profile is productive (cost-efficient) or not. We use two outcome indicators (productivity measures), growth and child mortality, and two spending measures, total government consumption (spending) and the share of health spending in GDP (health spending). By relaxing the assumption of common parameters across countries we can then simply test if the partial regression coefficient on spending (health spending) is a function of the degree of polarization. Specifically, the question we are interested in is if government spending (health spending) will have significantly different effects on growth (health status) in a more polarized country than in a less polarized one. This alternative method is easily implementable. Specifically, we estimate the following equation γ jt = βx Xjt + β g Gjt + β p Pjt + εjt
(2)
separately for the subgroups of ”more” polarized and ”less” polarized countries, respectively. In (2), γ jt is the outcome indicator (growth in country j in period t, child mortality in country j at the end of time period t), Pjt is defined above, Gjt is total spending or health spending depending on which outcome indicator is used, and Xjt is a vector of controls. β g and β p are coefficient scalars, β x is a coefficient vector, and εjt is a zero mean error term. If the efficacy of public spending is related to the degree of polarization, the marginal impact of spending should differ across subgroups.7 We define the subgroup of more polarized countries as those countries with a score on the polarization index above the mean in the full sample, and consequently the group of less polarized countries as those with a score on the polarization index below the mean in the full sample. 7
An alternative approach is to estimate a pooled regression and interact spending and polarization. However, this approach imposes the restriction that the coefficients of the control variables are the same for the two groups of countries (and common error variance). This need not be the case in reality.
7
To abstract from business cycle fluctuations we study the relationship between polarization and economic outcomes over decades: 1970s, 1980s, and 1990s. Thus, each country has three observations, data permitting. The equations are estimated with OLS with decade fixed effects. Standard errors are corrected for heteroskedasticity. Summary statistics are provided in Table 1. Unless otherwise noticed, the data is obtained from the Global Development Network Growth Database (Easterly and Yu, 2000).8 All variables are defined in appendix B. To examine the sensitivity of the results, we experiment with different conditioning information sets (Zjt , Xjt ). We seek to reduce the chances that the cross-country regressions either omit an important variable or includes a specific set of regressors that yields a favored result. We report the results with three conditioning information sets for the growth regression. The simple conditioning information set includes a constant for each decade, the initial real per capita GDP, and the initial level of educational attainment. The initial income variable is used to capture the convergence effect and school attainment is used to control for the level of human capital. The policy conditioning information set includes the simple conditioning information set augmented with a battery of some of the most commonly used explanatory variables in crosscountry growth regressions: the black market exchange rate premium (see e.g., Barro and Sala-i-Martin, 1995); the central government surplus to GDP (see e.g., Easterly and Levine, 1997); a measure of openness (see e.g. Dollar, 1992; Sachs and Warner, 1995), and the degree political and civil rights (see e.g., Barro and Sala-i-Martin, 1995). The full conditioning information set includes the policy conditioning information set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, and East Asia. In the specification with child mortality as dependent variable we follow Filmer and Pritchett (1997) and include in the simple conditioning set a constant for each decade, real per capita GDP, and the level of educational at8
The data are available at http://www.worldbank.org/html/prdmg/grthweb/GDNdata.htm
8
tainment. Filmer and Pritchett (1997) show that a large part of the variation in mortality is associated with income. To minimize, although not eliminate, problems of reverse causation, both per capita GDP and educational attainment are measured at the start of the sample period. The full conditioning information set includes the simple conditioning set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, East Asia, and the proxy for democracy. For the spending equation (1) we also report the results with three conditioning information sets. The simple conditioning information set includes decade fixed effects. Following Alesina and Wacziarg (1997) and Rodrik (1998), the structural conditioning information set includes in addition to the constants for each decade, initial real per capita GDP, population, area, openness, and share of population above 65. The full conditioning information set includes the structural conditioning information set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, and East Asia.
2.2
Empirical facts
The results for the spending equation are depicted in Table 2. Polarization has a large significant effect on spending. The effect is reduced when more controls are added on to the regression, but the point estimate in column (3) still implies that going from the lowest to the highest score on the polarization index results in a 4-percentage point increase in consumption as share of GDP. The conditional findings of spending (health spending) on growth (child mortality) are reported in Table 3-4. As a benchmark, we pool the two samples. The results are depicted in column (1) in Table 3. Spending has a significantly negative effect on growth, a result which is consistent with most studies on growth and government size. Columns (2) and (3) show that this negative relationship is driven by the group of more polarized countries. In fact, while the adverse effect of spending on growth is large and highly significant in the group of polarized countries, the relationship is statistically insignificant in 9
the group of less polarized countries. Figures 4a and 4b illustrate the results reported in columns (2) and (3). Columns (4)-(7) show that the result is not driven by the omittance of other important variables, i.e. the variables in the policy and full conditioning information sets. Spending is negatively correlated with growth in polarized countries, while we find no robust relationship between spending and growth in the group of less polarized countries. While the results in Table 3 suggest that the efficacy of total government spending is related to the degree of polarization, a more specific test is provided in Table 4. Table 4 reports the results of estimating the relationship between health spending and health outcomes (infant and under-5 child mortality) conditional on degree of polarization. Column (1) again provides the benchmark, i.e., the results when pooling the two groups. The results suggest that on average, health spending reduces infant mortality. However, as shown in columns (2) and (3), this result masks the asymmetric effect of spending in more and less polarized countries. We find a strong negative relationship between health spending and infant mortality in less polarized countries, but no robust relationship in more polarized ones. This partitioning is robust to the inclusion of other potential controls, columns (4)-(5), and to alternative health indicators (under-5 child mortality), columns (6)-(9). Figures 5a and 5b illustrate the results reported in columns (2) and (3). To summarize: Polarized countries have higher government spending. Moreover, the effect of spending on outcome is conditional on degree of polarization. For highly polarized countries, the results suggest that there is no significant relationship between health spending and health outcomes. Thus, more spending (even on potentially “productive items” such as health) does not seem to imply more productive output.9 What can account for these empirical patterns? We 9 The regression results reported above pass a battery of diagnostic and sensitivity tests. As shown above, the results are robust to modifications in the conditioning information set. To further test this finding we added additional controls, including the inflation rate, and measures of political instability. These did not alter our findings.
10
turn to this question next.
3
Model
Below we lay out a model that can sheed some light on the empirical findings reported above. The model yields implications on both the relathionship between polarization and government spending and polarization and the provision of public goods that are consistent with the empirical facts. However, the model also highlights a political institution, which we broadly can think of as primary elections, that can mitigate the adverse consequences of polarization.
3.1
Basic setup
The public sector consists of two public agencies (or bureaucrats - we will use both terms interchangeably), denoted with subscript j ∈ {1, 2}. Each of them initiates and monitors a project in period 1 of a given size. The projects provide benefits to the citizens. Each public agency also chooses some cost parameter To check for the potential influence of outliers, we examined the residuals from the regressions. We removed all observations with residuals more than three-standard deviations away from zero and re-ran the regressions with the full conditioning set. This did not alter the results. Then, we removed all observations with spending levels more than three-standard deviations away from mean. This did not change the conclusions either. The strong relationships between polarization and spending, and the conditional impact of spending (health spending) on growth (health status) do not seem to be driven by outliers. To test that our assumption of allowing the coefficient vector on the control variables to differ between the two groups of countries did not drive the results, we pooled the sample and ran regressions with an interaction term between polarization and spending. The coefficient on the interaction term captures the conditional effect of spending on growth. The results (available upon request) are qualitatively similar to those reported above. The marginal impact of spending on growth is significantly correlated with degree of polarization for high values of polarization. In the growth regressions discussed above we have treated spending as predetermined. However, recent literature on the determinants of government size as well as the findings reported in Table 2 cast doubts on this presumption. Alesina and Wacziarg (1997) and Rodrik (1998) suggest that country size, openness and government size are interconnected. Unfortunately, most variables in the spending equation also belong in the growth regression, thus making it difficult to instrument for spending. A second best strategy to avoid contemporaneous correlation is to use lagged spending levels as explanatory variables. The results of this test confirm the basic findings (results available upon request). While the effect of spending on growth is smaller in absolute terms when using lagged spending levels, the significant difference between more and less polarized countries remain.
11
for the project, cj ∈ [c, ∞), where c
γ a , j = 1, 2, the elected candidate will not provide any accommodating finance and neither of the projects will be implemented. Thus, (x1 , x2 , t) = (0, 0, 0). Second, if cj ≤ γ b , the elected party will ensure that both projects receive sufficient financing. Hence, (x1 , x2 , t) = (1, 1, 2cj ). If cj lies in between these threshold values; that is, cj ∈ (γ b , γ a ], party a (b) will provide additional financing only so as to ensure that project 1 (2) is implemented. It is constructive to think of the bureaucrat’s problem as choosing between two strategies: a “safe strategy” (denoted by S) in which cj is chosen so as to ensure that the jth agency’s project is implemented with certainty, and an “uncertain strategy” (denoted by U) in which the bureaucrat chooses a higher cost structure realizing that the project will not be implemented with some positive probability. For later references, the S-equilibrium (U-equilibrium) is the subgame perfect equilibrium in which both bureaucrats choose S (U). Note that since the bureaucrats utility is increasing in c, the S-equilibrium has both public agencies choosing cj = γ b . Consider next the U-equilibrium. The jth bureaucrat’s maximization program is simply max p(c1 , c2 )u(xj cj )
(3)
s.t. c ≤ cj ≤ γ a
(4)
where p(c1 , c2 ) denotes the probability that the party biased in favor of the jth agency’s project wins the election. Assume c ≤ γ b , implying that both strategies are feasible. We can now determine who wins the election. To this end let pa (c1 , c2 ) denote the probability that party a wins the election as a function of c1 and c2 . Consider the case where both bureaucrats find it optimal to choose U. Given c1 and c2 , agent i will vote for party a if y − c1 + αi ≥ y − c2 + (1 − αi ) 15
(5)
which can be simplified to αi ≥
1 2
[c1 − c2 + 1] ≡ ψ. Thus, the probability that
party a wins the election is ·µZ 1−ϕ+µ Z a f (αia , µa )dαia + pa (c1 , c2 ) = Pr ψ
1−ϕ+µb
ψ
¡ ¢ g αib , µb dαib
= Pr [m ≥ c1 − c2 ] = 1 − H(c1 − c2 ) where m ≡ (µa + µb ), and
H(q) =
2 π + q 2 + 2qπ if q ≤ 0 2π 2 2 2 π − q + 2qπ if q > 0 2π 2
¶
¸ ≥1 = (6)
(7)
and where pb (·) = 1−pa (·). By definition, m ∈ [−π, π]. Note that dpa (c)/dc1 < 0 and dpa (c)/dc2 > 0. Everything else equal, the less efficient cost structure chosen by bureaucrat 1 the lower the probability that party a wins the election. The polices chosen by the two public agencies in the U-equilibrium, c˜1 and c˜2 , are defined by the Nash conditions that both public agencies play their best response to each others actions. The best-response functions cj = C(c−j ), j = 1, 2, can be derived by solving the maximization program (3)-(4). In the c). unique symmetric U-equilibrium, cj = c˜, where c˜ is the fixed point c˜ = C(˜ To characterize the equilibrium it is convenient to define the threshold platform γ¯ k , implicitly defined by c) = 0 u(1 − γ¯ a ) − 12 u(˜
(8)
where γ¯ b = 1 − γ¯ a by symmetry. In (8) γ¯ k is the platform for which the bureaucrat’s expected utility of choosing S is equal to the expected utility of choosing U. We are now ready to determine the subgame perfect equilibrium level of cj ,
16
denoted by c∗ ,
12
c˜ if γ a > γ¯ a c∗ = γ b if γ a ≤ γ¯ a
(9)
Thus, if γ a is sufficiently low both bureaucrats will choose to get their projects implemented with certainty. If this is not the case, both bureaucrats choose U, realizing that in equilibrium there is 50 percent chance that their project will not be implemented. ¢ ¡ Assumption 1 (Interior solution feasible): γ a = 1 − γ b ≥ c˜ . We can now prove the following.
Proposition 1: Assume that assumption 1 holds. Then the higher the degree of polarization, the less cost-efficient policies will be chosen in equilibrium, and the wider the range of parameter values for which only one project will be implemented. Proof. See appendix 1. The competing public agencies face soft-budget constraints in that cost overruns will be bailed out (or accommodated) by the party biased in their favor. However, if voters are mobile and willing to switch party, an election induces competition among the bureaucrats, decreasing cost overruns, and hardening soft-budget constraints. Such a competition effect is stronger the less polarized the electorate is; that is, the less attached voters are to the two parties. This result warrant two remarks. First, increased polarization is associated with both higher government spending and lower supply of public goods. Thus, it is important to separate the effects from public capital (projects) on welfare, from the effects of public spending on public capital. Public spending and output are not necessarily correlated. 12
We assume that in the (indeterminate) case, where the expected payoffs of the two strategy profiles are equal, the Pbureaucrats coordinate on the equilibrium which maximizes aggregate expected utility E j u(cj ). If aggregate expected utility is equal, we assume that the bureaucrats coordinate on the S equilibrium.
17
Second, an interesting implication of the model is that “ideologically motivated” projects; that is, projects that yield very different payoffs for the two parties (high γ a ), must be more cost-efficient to survive both administrations.
4.1
Examples
The relationship between the degree of electoral polarization, the parties’ platforms and the cost of projects can be seen clearly by considering a few special cases. First, consider the case where utility (bureaucrats’) is linear; that is, uj = xj cj . With linear utility, c∗ = π/2 if γ b < π/4, and c∗ = γ b otherwise. This case is illustrated in Figure 2. /(1 − σ), where σ is the coefficient of relative risk Next, let uj = xj c1−σ j aversion, with 0 < σ < 1. Then, c∗ = π(1 − σ)/2 if γ b < (1/2)1/(1−σ) π(1 − σ)/2 and c∗ = γ b otherwise. Note that as the bureaucrats become more risk averse (increase in σ), the cost of the projects fall as does the cutoff for choosing c˜. Thus, if society wants to ensure low cost projects, it should appoint risk-averse civil servants. To better understand the role elections play in the model it is interesting to consider the case where instead of elections voters could vote up-and-down on the two projects. Voter i would support project 1 (2) if c1 ≤ αi (c2 ≤ 1−αi ). In this scenario there is no safe strategy available, so the problem for bureaucrat j is simply to choose cj so as to maximize pˆ(cj )xj cj (given linear utility), where pˆ(cj ) is the probability that the voters would approve the project. It is straightforward to solve the problem, yielding an unique cost-level c∗ = p π/2 + 12 (1 + 2π). The second term constitutes the additional cost-reducing effect of elections. Intuitively while the power to shut down projects provide
a certain degree of control over project costs, elections introduce an additional gain: a competition among public agencies in which lower costs by agency 1 not only increase the chances of party a winning the election, but also reduce the chances that agency 2:s project will be implemented, thus “forcing” it to also cut costs. 18
The competition effect can be further illustrated by looking at the case in which there is no polarization; that is, we consider the case where ϕ →
1 2
and
π → 0. In this scenario the bureaucrats know that all voters view the projects as perfect substitutes. When both bureaucrats play U, it follows that 1 if c1 < c2 a 1 p (c) = if c1 = c2 . 2 0 if c > c 1 2
The unique pure strategy subgame perfect equilibrium has the two agencies £ ¤ choosing c∗ = max γ b , c and both projects are implemented. Note that this
example also shows the important role played by the parties’ platforms. Pro-
vided that γ b (γ a ) is low (high) enough, both projects will be implemented at the lowest possible cost. As γ a , γ b → but as cj =
5
1 2
1 2
both projects will still be implemented
> c the equilibrium will no longer be cost-efficient.
Equilibrium public policy with endogenous parties
Thus far we have assumed that the parties’ platforms (or candidates) are exogenously given. In real elections, of course, political candidates are citizens who decide to run for office. In this section we extend the model by making these platforms endogenous. We study a specific political institution, primary elections, in which citizens in the two groups vote for candidates in the upcoming election. In an earlier version of this paper, we also show that other mechanism of choosing candidates, such as “citizen candidates” where each citizen initially decides whether or not to become a candidate for public office, produce identical results. We show below that by allowing the citizens to determine the parties’ platforms the incentive problems discussed above will become less pronounced. Thus, the institution governing the choice of candidates may serve an important role by mitigating the adverse consequences of polarization. Following the previous section we let each citizen be a member of one of the 19
two parties, citizens of type a (group a) are members of party a and citizens of type b are members of party b. In order to endogenize the platforms we introduce two additional assumptions. The first is a natural assumption on the information structure and the distribution of the voters’ preferences prior to the primary elections. Assumption 2: The distribution of voters in the two parties are given by F (αa , µea ) and G(αb , µeb ), where E[µl ] ≡ µel , l = 1, 2. Thus, the distribution of the party members is known. This assumption makes it easy to identify the median voter in each party. Note though that if we allow for uncertainty prior to the primary elections, the qualitative results of the analysis remain unchanged, at the cost of considerable increase in notational complexity. The second assumption is a normalization. Assumption 3: 1 − γ a = c. where γ a is implicitly defined by u(1 − γ a ) − 12 u(γ a ) = 0
(10)
This assumption is made to simplify the exposition as it limits the possible equilibrium outcomes.
5.1
Primary elections
In this section we add an initial voting stage over the parties’ platforms. In most democratic societies the parties platforms are determined through some social choice mechanism among the parties’ members. To be more precise, we let the voters in the two parties elect their candidate in closed primaries. The sequence of events is as follows: (i) Primary elections are held simultaneously in the two parties; (ii) The bureaucrats choose c1 and c2 , taking the identity of the two candidates and F (αia , µa ) and G(αib , µb ) as given; (iii)
20
A general election is held; (iv) The winning candidate (party) implements its most preferred policy. An equilibrium is defined by two conditions. It must be a Nash equilibrium among the two bureaucrats. It must also be a political equilibrium. That is, each candidate is preferred to any other candidate in the group by a majority of the members; given the outcome in the other primary; the reaction functions of the bureaucrats; and given how the elected candidates behave once in office, and the elected party receives at least 50 percent of the votes in the general election. We solve the problem through backward induction. We still consider only symmetric equilibria; that is, γ b + γ a = 1. The characterization of the equilibrium is parallel to that given in section 3, except that we add a stage at the beginning of the game. Thus, in the last stage of the game the elected party chooses the project implementation-financing triple (x1 , x2 , t). Both public agencies observe γ a and γ b before they choose cj . As in section 3, the public agencies choose between two strategies {S, U }, and the equilibrium configuration is given by (9). Next consider the first stage of the game, the primary elections, in which the two parties elect candidates. The previous discussion can be summed up by noticing that the choice of candidates determines the choice of cj and thereby the triple (x1 , x2 , t). To study the voters’ incentives in the primaries, let the expected utility of voter i in party l as a function of the candidates’ identities be h i h i ¡ i ¤ ¢ h £ E vli (x1 , x2 , t) = y − E t(γ a , γ b ) + αil E x1 (γ a , γ b ) + 1 − αil E x2 (γ a , γ b ) (11)
where E [(x1 , x2 , t)] =
(1, 1, 2(1 − γ a )) if γ a ≤ γ¯ a ¡ 1 , 1 , c˜¢ 2 2
(12)
if γ a > γ¯ a
There are two things to note about expression (11). First, it is linear in αil , 21
which is the only parameter that distinguish voters in party l, implying that there exists a unique Condorcet winner, namely the candidate with the median value of α in the group, say αm l . Second, the expected values of x1 , x2 and t are discontinuous in γ k . The optimal policy stance from the point of view of the median voter in party a (αm a ) is ³ ³ ´ ´ m a b m x γ Ev , γ (α ) , x (·) , t (·) γ*a = arg max 1 2 a b a γ
(13)
Note that the median voter in party a takes median voter in party b (αm b ) as given. This is because primary elections are held simultaneously within the two groups. The optimal choice of candidate from the point of view of the median voter in party b is defined analogously as ³ ³ ´ ´ b , x2 (·) , t (·) γ*b = arg max Evbm x1 γ a (αm a ), γ γb
(14)
The equilibrium platforms are defined by the Nash conditions that both median voters play their best response to each others actions. Proposition 2: Assume that assumption 2-3 hold. In the unique symmetric subgame perfect equilibrium voters in the two groups elect candidates, γ ∗a = γ a and γ ∗b = γ b and both projects are implemented. Proof. See appendix A.2. This equilibrium has two important properties. First, both projects will always be implemented. Second, the candidates’ preferences (or parties’ platforms) will typically differ from the median voters’ positions. In fact, relative to the party mean, the voters elect more moderate candidates in countries with high π, and less moderate candidates in homogeneous societies. The intuition for this result is clear if we interpret the political constitution as an incomplete contract, where the sharpness of the incentive scheme is determined by the difference in the candidates’ positions. Specifically, γ a and γ b determine the two bureaucrats’ expected payoff of choosing U and S, 22
¡ ¢ respectively. To illustrate, assume initially that π = π 0 ≡ 2γ a − 1 . Then ¡ a b¢ m (αm a , αb ) = γ , γ . If the median voters would run for office themselves, both
projects would be implemented at the lowest possible cost. An increase in π
implies that the competition across public agencies falls and the difference between the median voters’ positions rises.13 If the median voters would continue to run for office themselves, the public agencies’ expected utility of choosing S (U ) falls (increases). Hence, the bureaucrats would prefer to choose U. However, by electing more moderate candidates the voters can ensure that both public agencies will continue to choose S, implying that both projects are implemented and rents to special interests are minimized. Thus, primary elections counterbalance the adverse consequences of polarization. Conversely, a fall in π (starting at π 0 ) results in increased competition across public agencies and a smaller difference between the two parties. If the median voters would run for office themselves both projects would still be implemented but to a higher cost. Again, this effect could be counterbalanced at the primary election stage. Each party can be tougher and choose a more “extreme” candidate. Thus, in homogeneous societies primary elections strengthen the incentive scheme facing the bureaucrats, leading to reduced government spending and lower rents to special interests. The difference in equilibrium project costs as a function of π with and without primary elections is illustrated in Figure 3. To summarize, primary elections, or more generally party formation, perform an important role by shaping the degree of competition among government bodies. By adjusting the policy platforms according to the underlying degree of social division the voters improve their trade-off between policy mix and cost-efficiency in the face of greater polarization of society.14 Another implication worth emphasizing is that if there are no primary elecb m tions and γ a = αm a (γ = αb ), Proposition 1 holds even if we relax assumption 13
m Note that αm a = (1 + π)/2 and αb = (1 − π)/2. Interpreting primary elections more broadly as an additional instrument for participation and extended political rights, this result is broadly consistent with the findings in Rodrik (1997) and Svensson (2000). 14
23
1, since an increase in π implies that γ a increases, γ¯ a falls, and γ a remains the same.
6
Discussion
The cross-country data suggest that polarization is associated with both lower provision of public goods and higher government spending. We have provided a model which can account for these empirical findings, although it should be noted that there are potentially other mechanisms beside those explored in the model that could generate similar predictions. The key assumption we make in the model is that there exists a multiple of agents with ability to exploit the central government’s budget constraint. These agents could be ministries, local governments, public enterprises, public agencies or more generally bureaucrats. There is plenty of anecdotal evidence supporting this assertion. In the model we have captured the bureaucrats’ ability to exploit the central government’s budget constraint by assuming that the bureaucrats choose project costs prior to the general election. A democracy is characterized by periodic elections, while bureaucrats typically make continuous choices throughout the period. We have collapsed these choices into one, and it is intuitive to think of this choice as effort exerted by the bureaucrats in between elections. Another important assumption we make is that primary elections are held prior to the bureaucrats’ decisions. An objection to this assumption is that it gives the bureaucrats a rather narrow window of opportunity to exploit the government’s budget constraint. While technically correct, this critique takes the model’s structure too much at face value. Primary elections are meant to capture a process in which the citizens can influence the choice of candidates to stand for office, or more generally determine the competing parties’ policy platforms. In reality this is also a continuous process. Over time, the equilibrium platforms will reflect the fact that the candidate-pair running for office
24
influences the way the bureaucracy acts. Our timing assumption is intended to capture this “long-run” effect. A third assumption we make is that policy differences can be collapsed into a one-dimensional space (α). This simplifying assumption implies the voters in the two parties will always elect candidates that will ensure that both projects are implemented at the lowest cost. However, a slight generalization of the model, for instance letting voters favor “their” party (or candidate) even if the candidates if elected will implement the same policies (projects), would introduce a trade-off between these “non-economic” benefits (likely to increase in π) and cost-efficiency of public projects, implying that rents to special interest, while lower with primary elections, still would increase with π. Finally, by assuming a specific set of institutions we have been able to derive a number of results within a fairly simple model structure. At the same time, the model implicitly precludes various other ways of reducing inefficiency. Our choice of not dealing with these issues in the present framework does not imply that we think they are not important, but simply suggests that further research on the complex relationship between the degree of polarization and public sector performance appears fruitful.
7
Concluding remarks
A common concern of voters across countries is to elect leaders who will adopt the “right policy mix” and to have civil servants execute those policies in the most cost efficient manner. The problem is that they only have an indirect instrument to deal with this trade-off - - elections, and that voters may disagree over what leaders to choose and what is the right policy mix. This paper has shown that greater polarization of voter preferences is associated with less cost efficient policies and lower public output. Furthermore we have shown that primary election, or more generally the process to choose whom to run for office, provides the voters with an additional instrument to control both the
25
policy mix and implementation. By adjusting the policy platforms according to the underlying degree of social division, the electorate improves its trade-off in the face of greater polarization of society. In the model polarization is associated with both higher government spending and with lower provision of public goods. This pattern is also present in the data.
References [1] Aghion, P. and P. Bolton, 1990, Government Debt and the Risk of Default: A Political Economic Model of the Strategic Role of Debt, in R. Dornbusch & M. Draghi (eds.), Public Debt Management: Theory and Practise, Cambridge University Press. [2] Alesina, A., 1994, Political Models of Macroeconomic Policy and Fiscal Reforms, in S. Haggard & S. Webb (eds.), Voting for Reform: Democracy, Political Liberalization, and Economic Adjustment, New York, NY: Oxford University Press. [3] Alesina, A., R. Baqir, and W. Easterly, 1999, Public Goods and Ethnic Divisions, Quarterly Journal of Economics 114 (4): 1243-84. [4] Alesina, A., and H. Rosenthal, 2000, Polarized platforms and moderate policies with checks and balances, Journal of Public Economics, 75: 1-20. [5] Banks, J. and R. Sundaram, 1993, Long-Lived Principal, Short Lived Agents, University of Rochester, Working Paper. [6] Barro, R., 1973, The Control of Politicians: An Economic Model, Public Choice, 14: 19-42. [7] Barro, R.J., and X. Sala-i-Martin, 1995, Economic Growth, McGraw-Hill, Inc. [8] Besley, T. and A. Case, 1995, Incumbent Behavior: Vote-Seeking, Tax-Setting, and Yardstick Competition, American Economic Review, 85: 25—45. [9] Besley, T. and S. Coate, 1997, An Economic Model of Representative Democracy, Quarterly Journal of Economics, 112: 85-114. [10] Chari, V.V., L.E. Jones, and R. Marimon, 1997, The Economics of Split-Ticket Voting in Representative Democracies, American Economic Review, 87 (5): 957976. [11] Coate, S. and S. Morris, 1995, On the Form of Transfers to Special Interests, Journal of Political Economy, 103 (6): 1210-1235. [12] Dollar, D., 1992, Outward-oriented developing economies really do grow more rapidly: evidence from 95 LDCs, 1976- 1985, Economic Development and Cultural Change, 40: 523-544.
26
[13] Easterly, W. and R. Levine, 1997, Africa’s Growth Tragedy: Policies and Ethnic Divisions, Quarterly Journal of Economics, 112: 1203-1250. [14] Ferejohn, J., 1986, Incumbent Performance and Electoral Control, Public Choice, 50, 2-26. [15] Filmer, D., and L. Pritchett, 1997, Child mortality and public spending on health: How much does money matter?, Policy Research Working Paper no. 1864, The World Bank. [16] Grandmont, J., 1978, Intermediate Preferences and the Majority Rule, Econometrica, 46: 317-330. [17] Harrington, J.E., 1993, Economic Policy, Economic Performance, and Elections, American Economic Review, 83, 27-42. [18] Milesi-Ferretti, G.M., 1995, The Disadvantage of Tying Their Hands: On the Political Economy of Policy Commitments, The Economic Journal, 105, (November), 1381-1402. [19] Milesi-Ferretti, G.M. and E. Spolaore, 1994, How Cynical Can an Incumbent Be? Strategic Policy in a Model of Government Spending, Journal of Public Economics. [20] Murphy, K., 1983, Macroproject Development in the Third World: An Analysis of Transnational Partnerships, Westview Press. [21] Myerson, R. B., 1999, Theoretical comparisons of electoral systems, European Economic Review, 43 (4-6): 671-697. [22] Osborne, M. and A. Slivinski, 1996, A Model of Political Competition with Citizen Candidates, The Quarterly Journal of Economics, vol. 109, 65-96. [23] Persson, T., G. Roland and G. Tabellini, 1997, Separation of Powers and Political Accountability, The Quarterly Journal of Economics, vol. 112: 1163-1202. [24] Persson, T., and L.E.O. Svensson, 1989, Why A Stubborn Conservative Would Run A Deficit: Policy with Time-Inconsistent Preferences, Quarterly Journal of Economics, 104: 325-46. [25] Persson, T., and G. Tabellini, 1994, Representative Democracy and Capital Taxation, Journal of Public Economics, 55 (1): 53-70. [26] Pritchett, L., 1996, Mind Your P’s and Q’s: The cost of public investment is Not the value of public capital, Policy Research Working Paper no. 1660, The World Bank. [27] Reinikka, R. and J. Svensson, 2004, Local Capture: Evidence from a Central Government Transfer Program in Uganda, Quarterly Journal of Economics, vol. 119 (2): 679-705. [28] Rodrik, D., 1997, Where Did All the Growth Go? External Shocks, Social Conflict, and Growth Collapses, memo, Harvard University.
27
[29] Rodrik, D., 1998, Why do more open economies have bigger governments?, Journal of Political Economy, 106: 997-1032. [30] Rogoff, K., 1990, Equilibrium Political Budget Cycles, American Economic Review, 80: 21-36. [31] Sachs, J.D., and A. Warner, 1995, Economic Reform and the Process of Global Integration, Brookings Papers on Economic Activity, 1-118. [32] Svensson, J., 1998, Investment, Property Rights and Political Instability - Theory and Evidence, European Economic Review 42: 1317-1341. [33] Svensson, J., 2003, Who Must Pay Bribes and How Much? Evidence from a Cross Section of Firms", The Quarterly Journal of Economics, vol. 118 (1): 207-30. [34] Tabellini, G. and A. Alesina, 1990, Voting on the Budget Deficit, American Economic Review 80, 37-49.
A
Appendix
A.1
Proof of proposition 1.
The first order condition of bureaucrat 1:s maximization program is −h(˜ c1 − c˜2 )u(˜ c1 ) + pa (˜ c1 , c˜2 )uc (˜ c1 ) = 0
(15)
The second-order condition is −hc (·)u(·) − 2h(·)uc (·) + pa (·)ucc (·) < 0 Using the first-order condition (15) to substitute for u(·), yields £ ¤ −hc (·)h−1 (·)pa (·) − 2h(·) uc (·) + pa (·)ucc (·) < 0
(16)
(17)
∗ condition is that £Thus, a sufficient ¤ for cj = c , j = 1, 2,∗ to provide a local maximum −1 a −hc (·)h (·)p (·) − 2h(·) ≤ 0 for all cj near c . Since hc (·) > 0 for cj ≤ c∗ , (17) is strictly negative ∀cj ≤ c∗ . Using (7) to simplify we get, £ ¤ −hc (·)h−1 (·)pa (·) − 2h(·) = −3(π − m) ≤ 0
for cj = c∗ + ε, where ε is a ”small” positive number. Since π ≥ m, c∗ provides a local maximum. The first part of proposition 4.1 follows by applying the implicit function theorem to the first-order condition (15) − [hπ u + Hπ uc ] dπ + [−hc u − 2huc + pa ucc ] dc1 = 0
(18)
where the terms in the second bracket is the second-order condition |D|. Evaluating (18) at cj = c∗ , j = 1, 2, gives dc1 −π−2 = >0 dπ |D|
28
(19)
The subgame perfect equilibrium is found by comparing the bureaucrats expected utility of choosing U and S; i.e. u(γ b ) and 12 u(˜ c). We know that only one project will d d u(γ b ) = 0 while dπ u(˜ c) > 0. be implemented in the U-equilibrium. Note that dπ a Given γˇ , let π ˇ be the degree of polarization for which (8) holds. Then, an increase in π above π ˇ leads to lower public project provision. Since 2ˇ γ b < c˜, due to the concavity of u(·), an increase in π above π ˇ also leads Pto higher government spending, even though fewer projects are implemented. Thus, j (cj − c) rise.
A.2
Proof of proposition 5
£ ¤ ¡ ¢ ¡ ¢ Note first that E vli (x1 , x2 , t) is maximized at γ a , γ b = γ a , γ b . Suppose that the median voter in party a [b] would elect a candidate γ a > γ a [γ b < γ b ]. Then it follows from (9) that bureaucrat 1 would choose c1 = c˜, and bureaucrat 2 would choose c2 = 1 − γ b and only one project £ would ¤ be implemented in equilibrium. From (12) we see that this would reduce E vli (·) . Suppose instead that median voter in group a [b] would elect a candidate γ a < γ a [γ b > γ b ]. Then both projects would still £ i ¤ be implemented ¡ ain bequilibrium ¢ ¡ a b ¢but to a higher cost. Thus, E vl (x1 , x2 , t) attains its maximum at γ , γ = γ , γ .
B B.1
Appendix Data description
africa = Dummy variable for Sub-Saharan African countries. area = square kilometer [Source: GDN, 2000]. surplus = Fiscal Surplus/GDP. Decade average of ratio of central government surplus to GDP [Source: GDN, 2000]. bmp = Log of 1+ black market premium, decade average [Source: Global Development Network Growth Database (GDN) 2000]. democracy = Ranking of political liberties on a scale from 0 to 6, where 6 is most free [Source: Gastil (1982 and subsequent issues)]. easia = Dummy variable for East Asia countries. financial = Financial Depth. Ratio of liquid liabilities of the financial system to GDP, decade average [Source: GDN, 2000]. growth = Growth rate of real per capita GDP, World Bank [Source: GDN, 2000]. health spending = health spending as % of GDP based on IMF: Government Financial Statistics: Central Govt. [Source: GDN, 2000]. human = Log of 1 + average years of school attainment, quinquennial values (1970-75, and 1980-85) [Source: Barro & Lee (1993)]. infant mortality = mortality rate, infant (per 1,000 live births) [Source: GDN, 2000]. latin = Dummy variable for Latin America and the Caribbean. lgdp = Log of initial Income. Log of real per capita GDP measured at the start of each decade [Source: PWT5.6]. log of population = Log of population [Source: GDN, 2000]. openess = Export plus imports as percent of GDP (%) [Source: GDN, 2000]. polarization = logarithm of 1 + the index of ethnic polarization (ethnolinguistic fractionalization), 1960 [Source: GDN, 2000]. spending = Real government consumption as share of GDP, percent, 1985 intl. prices, [Source: PWT5.6]. under 5 mortality = mortality rate, under-5 (per 1,000 live births) [Source: GDN, 2000].
29
Figure 1: Distribution of voters’ preferences
g (α b , µ b )
f (αa , µa )
1−α
α 1 2
biased towards project 2
biased towards project 1
Figure 2: Determination of equilibrium
π u(1 − γ a ) = γ b 1 2
c~ 1 u(c~) 2
1 2
Note: Linear utility. Equilibrium in bold.
γ
a
γa
Figure 3: Comparison of project cost with and without primary elections cost per project Without primary elections
With primary elections
c
π0 Note: Linear utility.
c~ = γ a ; αlm = γ l , l = a, b
π
Figure 4a. Partial correlation between growth and spending in polarized countries coef = -.11707273, se = .04423965, t = -2.65
growth | X
5.81945
-10.1853
-12.3035
spending | X
30.7403
Note: The sample of polarized countries includes countries with a score on the polarization index above the mean in the full sample. X is the simple conditioning information set
Figure 4b. Partial correlation between growth and spending in less polarized countries coef = -.03298143, se = .0520545, t = -.63
growth | X
4.89078
-7.94421
-9.35564
spending | X
22.1007
Note: The sample of polarized countries includes countries with a score on the polarization index above the mean in the full sample. X is the simple conditioning information set
Figure 5a. Partial correlation between infant mortality and health spending in polarized countries coef = -.03039838, se = .06247158, t = -.49
infant mortality | X
.934093
-.911222 -2.3807
health spending | X
2.92513
Note: The sample of polarized countries includes countries with a score on the polarization index above the mean in the full sample. X is the simple conditioning information set.
Figure 5b. Partial correlation between infant mortality and health spending in less polarized countries coef = -.07062394, se = .02607999, t = -2.71
infant mortality | X
.944719
-.595879 -3.16965
health spending | X
4.45463
Note: The sample of less polarized countries includes countries with a score on the polarization index below the mean in the full sample. X is the simple conditioning information set.
Table 1. Summary statistics: Pooled data (1970, 80, 90) Mean
Median
Min
Max
Std. dev.
No.
Growth
1.61
1.65
-10.36
9.38
2.75
318
Spending
19.21
17.38
3.90
50.83
7.93
318
Polarization
3.34
3.76
0
4.54
1.10
318
Health spending
2.11
1.62
7.76
0.04
1.68
255
Infant mortality
3.32
3.35
1.31
5.22
1.07
255
Under 5mortality
3.51
3.42
5.77
1.61
1.13
226
Table 2. Effect of polarization on spendinga Equation Dependent variable Polarization
Simple conditioning setb
(1)
(2)
(3)
Spending
Spending
Spending
1.950 (.353) [.000]
0.861 (.343) [.013]
0.680 (.335) [.045]
X
Structural conditioning setc
X
Full conditioning setd No. obs R2
X 321 0.08
293 0.33
293 0.39
Notes: (a) Dependent variable is real government consumption as share of GDP. Estimation by OLS, with decade fixed effects. Standard errors, reported in parentheses, are corrected for heteroskedasticity (White, 1980). P-values in brackets. (b) Simple conditioning set includes decade fixed effects. (c) Structural conditioning set includes the simple conditioning set plus initial real per capita GDP, population, area, and openness. (d) Full conditioning set includes the structural conditioning set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, and East Asia.
Table 3. Conditional effects of spending on growtha Equation
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Growth
Growth
Growth
Growth
Growth
Growth
Growth
All
More Polarized
Less Polarized
More Polarized
Less Polarized
More Polarized
Less Polarized
Spending
-0.089*** (.023)
-0.117*** (.032)
-0.032 (.034)
-0.072** (.032)
-0.024 (.034)
-0.067** (.026)
-0.045 (.031)
Polarization
-0.484*** (.141)
-1.541** (.671)
-0.845*** (.282)
-1.217* (.707)
-0.658** (.261)
-1.876** (.725)
-0.292 (.252)
X
X
X X
X X
X
110 0.32
92 0.57
Dependent variable Sample
Simple conditioning set b Policy conditioning set c Full conditioning set d No. obs R2
270 0.30
160 0.30
110 0.34
110 0.43
92 0.48
Notes: (a) Estimation by OLS, with decade fixed effects. Standard errors, reported in parentheses, are corrected for heteroskedasticity. *** [**] (*) denotes significance at the 1 [5] (10) percent level. (b) Simple conditioning set includes decade fixed effects, the logarithm of initial real per capita GDP and initial level of educational attainment. (c) Policy conditioning set includes the simple conditioning set plus the black market exchange rate premium, central government surplus to GDP, openness, and degree political and civil rights. (d) Full conditioning set includes the policy conditioning set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, and East Asia, and openness. (e) lagged spending as explanatory variable.
Table 4. Conditional effects of health spending on child mortalitya Equation
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Infant mortality
Infant mortality
Infant mortality
Infant mortality
Infant mortality
Under 5mortality
Under 5mortality
Under 5mortality
Under 5mortality
All
More Polarized
Less Polarized
More Polarized
Less Polarized
More Polarized
Less Polarized
More Polarized
Less Polarized
Health spending
-0.055*** (.017)
-0.030 (.045)
-0.071*** (.016)
-0.019 (.028)
-0.055*** (.017)
-0.051 (.048)
-0.063*** (.018)
-0.039 (.029)
-0.059*** (.018)
Polarization
0.093*** (.026)
0.281 (.198)
0.194*** (.040)
0.600*** (.139)
0.159*** (.060)
0.355* (.213)
0.227*** (.060)
0.642*** (.141)
0.160** (.069)
X
X
X
X
X X
X
84 0.93
78 0.92
Dependent variable Sample
Simple conditioning set b Full conditioning set d No. obs R2
185 0.87
98 0.86
87 0.91
X
X
98 0.92
87 0.92
84 0.86
78 0.90
Notes: (a) Estimation by OLS, with decade fixed effects. Standard errors, reported in parentheses, are corrected for heteroskedasticity. *** [**] (*) denotes significance at the 1 [5] (10) percent level. (b) Simple conditioning set includes decade fixed effects, the logarithm of initial real per capita GDP and initial level of educational attainment. (c) Policy conditioning set includes the simple conditioning set plus dummy variables for Sub-Saharan Africa, Latin America and Caribbean, and East Asia, and an index of civil and political rights.
