de05_Article3 11/16/05 3:59 PM Page 1025

Journal of Economic Literature Vol. XLIII (December 2005), pp. 1025-1048

Constitutions, Politics, and Economics: A Review Essay on Persson and Tabellini’s The Economic Effects of Constitutions

∗ DARON ACEMOGLU

In this essay, I review the new book by Torsten Persson and Guido Tabellini, The Economic Effects of Constitutions, which investigates the policy and economic con- sequences of different forms of government and electoral rules. I also take advantage of this opportunity to discuss the advantages and disadvantages of a number of popu- lar empirical strategies in the newly emerging field of comparative political economy.

1. Introduction is an excellent introduction to basic models of political decision-making and their impli- olitical economy has now emerged as an cations for economic outcomes. The objec- active and flourishing subdiscipline of P tive of their new book, The Economic Effects economics, with the ambition of offering of Constitutions, also published by MIT insights into the causes of the large cross- Press, is to continue where the first book left country differences in major economic out- off, and confront theory with data. comes and policies.1 Torsten Persson and The book deserves enthusiastic reception Guido Tabellini deserve much of the credit for a number of reasons, including the for making this happen. Their book, Political authors’ role in advancing the political Economics: Explaining Economic Policy, economy research program, its ambitious which was published by MIT Press in 2000, objectives, its careful scholarship, and the long list of findings that come out of the ∗ Acemoglu: Massachusetts Institute of Technology. I research methodology. Equally, however, thank Joshua Angrist, Victor Chernozukov, Roger Gordon, such a bold attempt at confronting theory Chad Jones, Torsten Persson, Guido Tabellini, Simon Johnson, and Pierre Yared for useful comments and Pierre with data requires a critical look. This moti- Yared for excellent research assistance. vates my review of their new book, focusing 1 While there is renewed interest in political econo- on the identification issues that arise both my, in large part in the context of understanding the sources of large cross-country differences in policies, in this context and more generally in recent institutions and economic performance, this work is pre- political economy research. ceded by the important results of the earlier “social This review essay is organized as follows. choice” literature. David Austen-Smith and Jeffrey S. Banks (2000) provide an excellent introduction to the Section 2 introduces the context and the major results in this literature. objectives of Persson and Tabellini’s new

1025 de05_Article3 11/16/05 3:59 PM Page 1026

1026 Journal of Economic Literature, Vol. XLIII (December 2005)

Policy Preferences

Economic Constitutional Outcomes Rules

Markets Political Outcomes

Policy Decisions

Figure 1: The Policymaking Process (Figure 1.1, page 3)

book. Section 3 provides a detailed discus- the rental market works. The major advance sion of econometric issues and identification that came with political economy is encapsu- problems that Persson and Tabellini have lated in a new question: why do policies dif- rightly emphasized in their work. Section 4 fer over time, across countries and more discusses the major empirical results of the broadly across polities? Why, for example, do book. Sections 5–8 discuss a number of some governments impose rent control problems and challenges facing empirical while others do not intervene in rental mar- work in political economy, and reevaluate kets. Before, we had no better answer than Persson and Tabellini’s contribution in this “(some) politicians just don’t get it.” Political light. Sections 9 and 10 discuss two sets of economy is about developing better answers. topics that I view as important areas for This task requires a framework for think- future research in political economy. ing about why policies differ across coun- tries. This is what much of political economy 2. Background and Objectives does. Starting with Kenneth J. Arrow’s cele- brated (im)possibility theorem (Arrow 1963, Persson and Tabellini (henceforth PT) set Duncan Black 1948), Anthony Downs’s the scene with figure 1.1 in their book, median voter result (Downs 1957, Harold which I replicate here. The box on the left is Hotelling 1929), and George J. Stigler’s work where much of economics falls. We have a on regulation (Stigler 1970, 1972), the cen- good, though far from perfect, understand- tral working hypothesis has been that ing of how different organizations of markets agents—as voters, lobbyists, revolutionaries, lead to different economic outcomes. But politicians—have induced preferences over why are markets organized differently in dif- policies. This means that they understand ferent societies? Economists and other that different policies will map into different social scientists have long realized the outcomes, and consequently their prefer- importance of “policies.” If the government ences over policies are shaped by their pref- imposes rent control, this will impact how erences over the outcomes that will be de05_Article3 11/16/05 3:59 PM Page 1027

Acemoglu: Constitutions, Politics, and Economics: A Review 1027

induced by the policies.2 Political economy comparative political economy more specif- is then about understanding how these ically. What makes this work even more induced preferences over policies are aggre- impressive is the fact that the authors are gated. The recent work in political economy investigating predictions of their earlier is also part of this research program, but models. more explicitly investigates why collective In reality, PT have an even more ambi- decisions and policies differ across societies. tious goal. Many scholars attempt to docu- Much work in political science, on the ment a set of correlations in the data that are other hand, as succinctly captured by the box related to predictions of existing models. on the right in Persson and Tabellini’s figure Instead, PT are interested in more than cor- is about how different political procedures relations. They undertake the ambitious task lead to different political outcomes. Let us of identifying the causal effects of the form refer to political procedures as “political of political institutions on economic and institutions” and make the link between political outcomes. They write on page 7: political outcomes and policies. This gives us “Our ultimate goal is to draw conclusions a natural framework for thinking about why about the causal effect of constitutions on different societies choose different policies— specific policy outcomes. We would like to they have different political institutions. answer questions like the following: if the PT’s first book, Political Economics, devel- United Kingdom were to switch its electoral oped various models of collective decision- rule from majoritarian to proportional, how making and derived predictions on how would this affect the size of its welfare state differences in political institutions translate or its budget deficits? If Argentina were to into differences of policies and economic abandon its presidential regime in favor of a outcomes. This new book is the next step to parliamentary form of government, would push this framework further by confronting this facilitate the adoption of sound policy its predictions with data. towards economic development?” Although this type of comparative analysis How would we get to the causal effects of has been vibrant in political science, there a set of political institutions on outcomes? are no systematic theoretical analyses deriv- PT have a well-specified strategy. First, ing precise predictions from micro-founded they focus on a subset of political institu- models to confront with data. Here PT are tions; the constitutionally determined form much more ambitious. They take the pre- of electoral rules and the form of govern- dictions of a set of theoretical models, in ment among democracies. This leaves out a particular, the probabilistic voting model of large part of political institutions, for exam- Assar Lindbeck and Jörgen Weibull (1987), ple, those that relate to whether a country extensively studied in their first book, and is a democracy or not, and thus a lot of an extended version of John Ferejohn’s interesting questions related to what (1986) model of politician accountability, makes constitutions credible in the first 3 studied in Persson, Gerard Roland, and place. Nevertheless, narrowing the scope Tabellini (1997, 2000), seriously. Although one can have qualms about the specific 3 See, for example, Avinash Dixit, Gene M. Grossman, assumptions that are important to derive and Faruk Gul (2000). This question perhaps is not first- order for PT because they focus on democracies that are predictions in these models, PT’s is the right “fully consolidated” and do not switch to dictatorship, and approach in social science in general, and in investigate how electoral rules affect policies within the democratic framework. Nevertheless, whether a society is a fully consolidated democracy is endogenous and poten- 2 This approach is referred to as “rational choice” in tially also depends on the form of government and elec- political science. To economists, it is the natural approach toral rules in democracy (see Acemoglu and Robinson without apologies. 2005 for a discussion). de05_Article3 11/16/05 3:59 PM Page 1028

1028 Journal of Economic Literature, Vol. XLIII (December 2005)

of the investigation is the right approach to on a model that is linear in the covariates be able to estimate causal effects. Second, and features constant (marginal) effects. × they use a variety of state-of-the-art tech- Let Xi be a K 1 vector of (observable) niques developed in the micro-economet- exogenous covariates and suppose that the rics literature, in particular, matching structural relationship of interest is = + β estimators, propensity score methods, E(Yi|Xi) Si X i , or: instrumental variables estimation, and =+ β + parametric selection corrections. (1) YSiiiiX u, Since the identification of causal effects where measures the effect of Si on the out- is the major objective of this book, and also β × because the econometric issues that arise come of interest, is a K 1 vector of coef- here are becoming increasingly important ficients associated with X, and ui captures the effect of all unobserved determinants as in the political economy literature, in the 5 next section I discuss PT’s econometric well as random influences on Yi. methodology in detail. PT use four different empirical strategies to estimate (1) or variants: ordinary least- squares (OLS), matching and propensity 3. Econometric Methodology score estimators to relax the linearity assumption in (1), instrumental variables 3.1 The Basic Framework (IV), and parametric selection corrections due to Heckman. PT’s major interest is to identify the These strategies deal with the identifica- causal effect of two political institutions, tion problem in (1) in two different ways, presidentialism (versus parliamentary sys- which can be classified as “selection on tems) and majoritarian electoral rules (ver- observables” and “exclusion restrictions” sus proportional representation), on the (see Angrist and Krueger 1999). amount and the composition of government 3.2 Selection on Observables spending, and on politicians’ rents and other political outcomes. Ultimately, part of The identifying assumption for strategy 1 the interest is also to estimate the effect of is that conditional on Xi, Si and ui are orthog- these different constitutional forms on onal. This last condition implies that the aggregate productivity and income per OLS estimator is consistent. To see this capita levels. briefly, let variables with the tildes denote Following PT’s exposition in chapter 5, I the original variable after the effect of the 6 simplify the problem by focusing only on vector Xi has been partialed out. Then, (1) one binary constitutional feature, denoted can be rewritten as: by S ∈{0,1}, and a generic outcome vari- i =+ able, Yi, with i denoting country in a cross- (2) YSuiii, sectional dataset.4 PT consider a relatively general econometric framework, incorpo- which implies that rating potentially heterogeneous effects. For the purposes of this review, let us focus 5 It is worth noting that equation (1) can be estimated without the constant marginal effects and linearity assumptions to obtain “average treatment effects” with weights depending on the estimation strategy, see Angrist 4 See, for example, Joshua D. Angrist and Guido W. and Krueger (1999). Imbens (1995), Angrist, Imbens, and Donald B. Rubin 6 In words, these are the residuals from the original

(1996), Angrist and Alan B. Krueger (1999), James J variable’s linear projection on the vector Xi, given by Heckman (1997), and Jeffrey M. Wooldridge (2002), for yt = (I − X(XX)1X)y, where X denotes the K × N matrix

more detailed treatments. obtained from the Xi’s. de05_Article3 11/16/05 3:59 PM Page 1029

Acemoglu: Constitutions, Politics, and Economics: A Review 1029

distribution of noninstitutional political ( ) OLS Cov SY, power between various groups, differences plimˆ = ii Var(S ) in culture, and differences in other institu- i tional features affecting potential outcomes. ( ) Cov Suii, Since the conceptual issues arising with (3) =+ , ( ) both linear regression and matching estima- Var Si tors are similar, in what follows I focus on where the variance and covariance terms the identification issues with OLS, which refer to population values. Consequently, the identifying assumption for strategy 1 is are simpler to discuss. t t = To highlight the potential bias in OLS Cov(Si,ui) 0 (or that conditional on Xi, ui be statistically independent from Si). estimation, suppose, again using the tilde The fact that Si and ui need to be orthog- notation: onal conditional on X highlights the impor- i (4) uL˜ =+λω˜ , tance of selection on observables. It is quite iii possible that S is correlated with some i where Li is some other potential determi- other determinants of Y , but once we con- ~ i nant of the outcomes of interest, and L i dition on the covariates, this correlation dis- denotes its residuals from the projection on appears. Therefore, for this class of ω t ω = the Xi’s, while i satisfies Cov(Si, i) 0. This strategies controlling for the right set of implies that observables is of primary importance. Cov (SL, ) What distinguishes the OLS and match- plimˆ OLS =+λ ii. ing strategies is the way in which they con- ( ) Var Si trol for the effect of these observables. OLS regression, as in (1), imposes linearity, Therefore, the OLS estimator will be incon- whereas the matching estimators allow for sistent if there exists a potential determinant ≠ non-linear effects of observables. With Li of the outcome of interest (i.e., 0), matching estimators, the parameters of which is also correlated with the form of gov- interest are estimated conditional on a set of ernment or electoral rules (i.e., if t t ≠ observables, and then the average effect is Cov(Si,Li) 0. This corresponds to the usual obtained by averaging these conditional omitted variable bias. This notation can also parameter estimates. The advantage of this capture the measurement error (attenua- tion) bias whereby we observe a noisy signal estimator is that it allows the parameter of t interest to depend flexibly on the character- of the true institutions, Si , while what we t = t − τt τt istics used for matching. Propensity score care about is Ti Si i, where i is a classical t τt = methods enable consistent estimation when measurement error term, with Cov(Ti , i) 0. t =−τt λ = there are many characteristics according to In this case, Li i and , so that OLS = t t τt which observations can be matched (see plim ˆ Var(Ti)/(Var(Ti) Var( i)). Angrist and Krueger 1999, Wooldridge 3.3 Exclusion Restrictions and Instrumental 2002, chapter 18). Variables With both strategies, however, there are t t = numerous reasons why Cov(Si,ui) 0 may The most important line of attack that not be valid. In this context this might be, researchers have against potential biases of for example, because the constitutional fea- the OLS estimator is to use an instrumental tures of a country are partly determined by variables (IV) strategy, which is PT’s third omitted factors that also influence policy strategy. The IV strategy relies on a M × 1 choices and economic outcomes. These fac- vector of instruments, Zi, which has predic- tors may include differences in the nature of tive power for the endogenous regressors, but economic opportunities across societies, the is orthogonal to ui in (1) (where naturally, de05_Article3 11/16/05 3:59 PM Page 1030

1030 Journal of Economic Literature, Vol. XLIII (December 2005)

M ≥ K + 1). The exclusion restriction refers (5) SZv˜˜=+ , to this last requirement, and is equivalent to iii = the condition E(Ziui) 0. This means that where vi denotes the residuals from the linear t t t = the vector of instruments, Zi, has no effect regression of Si on Zi, thus Cov(Zi, vi) 0. The on the outcomes of interest other than its probability limit of the instrumental variable impact through the endogenous regressors. estimator of is given by A common view in the applied economics Cov (ZY, ) literature is that the IV strategy is “atheoreti- IV = ii (6) plim ˆ ( ) cal.” This view originates from the fact that Cov ZSii, typically the IV strategy does not necessi- tate a fully specified structural model. Cov (Z ,u ) Nevertheless, as I will argue in greater detail =+ i i γ ( ) . below, every IV strategy requires an under- Var Zi lying theory. In essence, the IV strategy This equation encapsulates the two require- starts with an essentially nontestable hypoth- ments for a valid instrument; (1) relevance, esis, that the vector of instruments, Z , is i which requires that there should be a first- orthogonal to omitted influences captured stage relationship, i.e., λ ≠ 0;7 (2) excludabil- by the error term ui. This presumption must t ity, which means that Cov(Z ,ut ) = 0 (which be based on some theoretical consideration, i i is equivalent to Cov(Z ,u |X ) = 0, or to and the IV procedure would have no justifi- i i i E(Z u ) = 0). cation without a compelling (a priori) theory. i i To illustrate the usefulness of the IV strat- This underlying theory can be extremely egy, let us return to the model of the unob- simple, for example, the hypothesis that, served effect in equation (4), and suppose everything else equal, a change in compulso- t t that Cov(S ,L ) ≠ 0, so that OLS is inconsis- ry schooling laws or their enforcement i i tent. In this case, using the IV strategy leads should affect schooling (e.g., Angrist and to an estimate of , ˆ IV, with the following Krueger 1991, Acemoglu and Angrist 2000), probability limit: but have no effect on earnings through other channels; or more involved, for example, the λCov (ZL, ) ˆ IV =+ ii hypothesis that all information contained in plim γ ( ) . the past income realizations should already Var Zi be incorporated in the current consumption t t = decisions and should have no predictive Therefore, if Cov(Zi,Li) 0, the IV estimator ≠ power for consumption growth (e.g., Robert will be consistent (given 0). The condi- t t = E. Hall 1978, Lars Peter Hansen and tion that Cov(Zi,Li) 0 is sometimes loosely Kenneth J. Singleton 1982). I return below referred to as the requirement that “the to the theoretical justifications of the instru- instrument should be orthogonal to omitted ments used in PT’s work and in other work in effects.” Moreover, the above discussion of political economy. measurement error illustrates that a valid For now, let me simplify the discussion by assuming that all of the variables in Xi 7 Note also that if 0, then even though plimˆ IV, are exogenous, and there is a single exclud- in small samples ˆ IV could systematically deviate from , able instrument, Z , for the only endoge- causing the “weak instruments” problem (see, for example, i Douglas Staiger and James H. Stock 1997, Stock, Jonathan nous regressor, Si. So the instrument vector H. Wright, and Motohiro Yogo 2002). The sister problem = is simply Zi (Xi,Zi). Using the tilde nota- is the “too many instruments” problem, which arises when tion, the estimating equation can be writ- the vector Zi includes a large number of excluded instru- ments (relative to the number of observations), leading to ten as (2), with a corresponding first-stage a spurious first-stage relationship (e.g., Christian Hansen, relationship of Jerry Hausman, and Whitney Newey 2004). de05_Article3 11/16/05 3:59 PM Page 1031

Acemoglu: Constitutions, Politics, and Economics: A Review 1031

instrument also removes the attenuation bias first-stage relationship (while the entire Xi is arising from classical measurement error. included in the second stage). In practice, the

In all of this, the set of covariates included in procedure is the following: first, Si is predict- 1 Xi is important because one can imagine that Zi ed on the basis of Xi and Zi (using linear ˆ could be correlated with the error term in the regression). Denote the predicted values by Si. second stage, but this correlation disappears Then the parameters of interest are estimated ˆ once we condition on a set of covariates (i.e., using an OLS regression with Si instead of Si. 1 ≠ PT we may have that while Cov(Zi,ui|Xi) 0, It can be shown that for this estimator, ˆ , to 2 = Cov(Zi,ui|Xi) 0 for some (not necessarily be consistent the Xi’s that are omitted from the mutually exclusive) different sets of covariates first stage should have no predictive power for 1 2 Xi and Xi). Interestingly, this issue does not the endogenous regressors (conditional on the arise when there is random assignment due to other covariates).8 Otherwise, the residuals real experiments or because of natural experi- from the first stage, which mechanically go ments. This makes the identification problem into the second stage, are no longer orthogo- more difficult and subtle in political economy nal to the covariates, and this biases the esti- models than in many other contexts, and mate of the vector β, and the bias from β makes the distinction between the selection on carries over to the estimate of the parameter observables and exclusion restriction strategies of interest, .9 Therefore, the estimator used I tried to highlight above less clear-cut. PT include income per capita, democracy score, age of democracy, measures of trade, 8 In particular, let the true (statistical) first-stage rela- 1 π 2 π tionship be Si Zi (X i) 1 (X i) 2 i, with Sˆi Zi measures of age composition of the popula- 1 π 2 π (X i) 1 (X i) 2 denoting the predicted component of Si. tion, a dummy for federal structure and a The second-stage equation of the traditional 2SLS proce- ˆ 1 β 2 β dummy for OECD in X in their main spec- dure is Yi Si (X i) 1 (X i) 2 ui. The assumption i β ifications. The instruments are various com- E(Ziui) 0 ensures consistent estimation of ( , ) with the standard IV procedure. In contrast, the predicted binations of the timing of adoption of the b b 1 πb value from the PT procedure is S i Zi (X i) 1 ˆ b 1 πb πb 2 π Si ( )Zi (X i) ( 1 1) (X i) 2, where the second country’s constitution, fraction of the popu- b equality substitutes for the definition of Si. So the second lations that speak English or another 1 β 2 β b stage in this case becomes: Yi Sˆi (X i) 1 (X i) 2 ui, European language, and latitude. b b 1 πb π 2 π where ui ui (( )Zi (X i) ( 1 1) (X i) 2). Defining b b I will discuss the validity of these instru- Wi (Xi,Si), PT’s estimator is − ments below. Before doing this, however, an N 1 N ⎛ − ⎞ ⎛ − ⎞ ηˆ PT = ⎜ NNY1 ∑∑WW⎟ ⎜ 1 W ⎟ important estimation issue needs to be ⎝ ii⎠ ⎝ ii⎠ = = raised. The above derivation of the consis- i 1 i 1 − tency of the instrumental variable estimator N 1 N ⎛ − ⎞ ⎛ − ⎞ = η + ⎜ NN1 ∑∑WW⎟ ⎜ 1 Wu ⎟ . made it clear that the same set of covariates, ⎝ ii⎠ ⎝ ii⎠ = = X , needs to be included both in the first and i 1 i 1 i b b For this estimator to be consistent, we need E(Wiui) 0. the second stage. This was made explicit by π b However, when 2 0, ui will be no longer be orthogonal to the “tilde” notation above, where the regres- 2 X i, and this will imply that the entire vector of coefficients η sions include the variables after the vector Xi is not estimated consistently. In other words, we will typ- b b PT has been partialed out. Equations (5) and (6) ically have E(Wiui) 0 and plim ˆ , even if E(Zi ui) 0. 9 For example, Wooldridge (2002, p. 91) notes: “In show immediately why the IV estimator, practice, it is best to use a software package with a 2SLS given the identification assumption rather than explicitly carry out the two-step procedure. t Cov(Z ,ut ) = 0, is consistent in this case. Carrying out the two-step procedure explicitly makes one i i susceptible to harmful mistakes. For example, the follow- PT, instead, choose a nonstandard econo- ing seemingly sensible, two-step procedure is generally metric method, and exclude some of the X ‘s inconsistent: (1) regress xK on 1, z1,…,zM [the set of exclud- i t from the first stage. In particular, suppose that ed instruments for xK] and obtain the fitted values, say xK ; (2) run the regression [of the dependent variable on the the vector of covariates is partitioned as t other covariates and on xK]...” This example is equivalent to = 1 2 1 Xi (Xi,Xi), and only Xi is included in the the procedure that PT implement. de05_Article3 11/16/05 3:59 PM Page 1032

1032 Journal of Economic Literature, Vol. XLIII (December 2005)

by PT may be inconsistent even when the measured as government spending as a standard identification assumption, fraction of GDP). Majoritarian systems t t = = 10 Cov(Zi,ui) 0 or E(Ziui) 0, is satisfied. also appear to have smaller welfare state One might wonder whether this is an impor- spendings and budget deficits. tant source of bias in practice, and I return 2)While presidential or majoritarian sys- to this problem below. tems in general do not have a robust Finally, PT’s Strategy 4, parametric selec- effect on political rents, corruption and tion corrections, is related to the IV strategy. aggregate productivity, certain details of This is an attempt to deal with the failure of the electoral system, in particular the t t = the assumption Cov(Si,ui) 0 by positing a size of the electoral districts and functional form for selection on unobserv- whether voters cast their ballots for ables, based on the work by Heckman (1976, individual politicians or for party lists, = 1979). Specifically, PT posit that Si 1 do have significant effects. + ≥ = if G(Zi) vi 0, and Si 0 otherwise. Fol- 3)Countries with parliamentary systems lowing the common practice, they assume have more persistent fiscal outcomes

that the vi’s are normally distributed, which than countries with a presidential sys- enables a parametric correction for selection tem. Namely, in parliamentary systems, (see, e.g., Wooldridge 2002). Parametric increases in government spending dur- selection approaches exploit the functional ing downturns are not reversed during

form assumptions on vi, but otherwise rely booms. There is a similar, but weaker on the same set of exclusion restrictions. pattern, for countries with proportional Consequently, though they are a useful representation (relative to those with check on standard IV estimates, they will majoritarian elections). Finally, consis- also lead to inconsistent estimates if the tent with the predictions of the political underlying exclusion restrictions are not business cycle models, proportional valid. This motivates my focus on the linear representation also appears to generate IV approach for most of the discussion, a greater expansion in welfare spending though I will also contrast the results of the in the proximity of elections. parametric selection approach to the OLS These results are exposed with great clari- and IV estimates. ty, and PT show that they are robust. For example, the results are similar with linear regressions, or with the matching and 4. Main Results and Assessment propensity score methods. Importantly, results are also similar with the instrumental Using all four strategies, PT obtain a num- variables and parametric selection correction ber of results that paint a broadly consistent strategies, though sometimes less precise. picture. The most important results are: What makes these results even more 1)Presidential and majoritarian systems important than this brief description would have smaller governments than parlia- suggest is that they are closely related to mentary and proportional representa- the predictions of the models that have tion systems (where government size is motivated PT’s study. For example, the models that are most popular in the litera- 10 PT justify this econometric method by arguing that, ture suggest that electoral rules, in particu- because the major source of bias is not reverse causality but omitted variables, excluding some second-stage covariates lar, how broad or narrow districts are, from the first stage might be justifiable. This may be an should affect both the overall size of gov- π = argument for why we should expect 2 0 in terms of the ernment, the composition of spending and notation in footnote 8. They further suggest that this might be useful because it would lead to a stronger first-stage the extent of political rents. These are con- relationship, avoiding weak instrument problems. firmed by the empirical results in the book. de05_Article3 11/16/05 3:59 PM Page 1033

Acemoglu: Constitutions, Politics, and Economics: A Review 1033

In addition, models based on the impor- 5. Identification in Political Economy tance of separation of power, such as 5.1 General Issues Persson, Roland, and Tabellini (1997, 2000), suggest that presidential systems PT have put identification of causal effects where there is typically greater separation as the primary objective of their work. In this of powers should have smaller govern- they are not alone. Much recent research of ments, which is also the pattern that PT political economy is about understanding find in the data. the effects of various policies or institutional These are remarkable results. Only a very features on economic development. The brave or uninformed scholar could attempt issues of identification are therefore at the to write on comparative political economy forefront. I will argue that even though PT without seriously studying this book, and have undertaken a careful and thorough only a very stubborn researcher would have investigation of the questions at hand, they his or her posteriors remain unchanged may have not estimated the causal effect of after studying it. PT have already achieved the constitutional features on the outcomes something very few scholars can: a body of of interest. Moreover, the identification work for not only the current generation of problems that arise in this work are shared researchers, but also for the next genera- by many other studies in political economy. tion. In fact, the impact of the book might Let us start by recalling that the basic even be greater than this discussion sug- approach is to understand differences in gests. If the results indeed correspond to policies and economic outcomes as a func- the causal effects of the form of government tion of the underlying political institutions of and electoral rules on policies and econom- societies. This methodological approach ic outcomes as PT claim, we have learned builds on the notion that agents have more with this book than from the entire induced preferences over policies because comparative politics literature of the past they recognize the implications of these poli- fifty years. cies for their well-being. Although I whole- Despite these remarkable results, there heartedly agree with this methodology, I also are reasons to question whether this think that there is an immediate next step research has successfully uncovered causal that one can (and should) take. By exactly effects. The OLS and matching estimates the same token, agents should have induced ultimately rely on the exogeneity of politi- preferences over political institutions. This is cal institutions. Nevertheless, political because different political institutions will institutions are equilibrium outcomes, generate different policies, and thus lead to determined by various social factors that different economic outcomes, and rational are not fully controlled for in the empirical economic (and political) agents should models. This makes me believe that, understand not only the implications of dif- although the OLS results uncover interest- ferent policies but also the implications of ing patterns, they do not typically identify different political institutions. The logic of causal effects. For causal effects, we have political economy therefore forces us to also to turn to estimates using the exclusion think of political institutions as endogenous. restrictions. Here PT’s work is again the This leads to an important issue: while a state-of-the-art, but the exclusion restric- large body of empirical literature argues that tions are not necessarily compelling, mak- political institutions are “predetermined” or ing it unlikely that the IV estimates are given by history and treats them as exoge- uncovering causal effects either. The next nous, the reasoning of the political economy three sections discuss these problems in approach implies that these institutions are greater detail. also endogenous and likely determined by the de05_Article3 11/16/05 3:59 PM Page 1034

1034 Journal of Economic Literature, Vol. XLIII (December 2005)

same factors that make policies unappealing elite prefer presidential systems to parlia- variables to treat as exogenous. mentary systems. In this case, when the elite This point has featured in one way or are more powerful, the equilibrium form of another in the comparative politics litera- government is more likely to be presidential. ture. It is emphasized in its most general If, in addition, the strength of the elite is form by the Charles A. Beard’s classic 1913 persistent, there will be an empirical corre- study, An Economic Interpretation of the lation between presidentialism and smaller U.S. Constitution, where he argues that the governments, not only at the time the con- primary objective of the government and the stitution is written, but also in the subse- constitution is to ensure favorable economic quent decades. But this correlation would conditions for those holding political power, not necessarily reflect the causal effect of and the form of government should be seen presidentialism on equilibrium policies.11 In as a (secondary) feature serving the same terms of the econometric discussion above, t t = objective—by “secondary” Beard does not the requirement that Cov(Si,ui) 0 in equa- mean less important, but chosen in the same tion (2), is violated because there is an omit- t way and for the same objectives as the over- ted factor, like Li in equation (4), in this case all structure of political power. He writes: the political strength of the elite, which will (1913, p. 13) affect both the form of government and the Inasmuch as the primary object of a govern- outcomes of interest. The same concerns ment, beyond the mere repression of physical generalize to the other potential outcome violence, is the making of the rules that deter- variables. mine the property relations of society, the In this light, the OLS (non-IV) estimates dominant classes whose rights are thus to be determined must perforce obtain from the of the causal effects of the form of govern- government such rules as are consonant with ment and electoral systems should not be the larger interests necessary to the continu- interpreted as causal. Instead, they likely ance of their economic processes, or they correspond to interesting robust correlations must themselves control the organs of gov- in the data. For causal estimates, we have to ernment. In a stable despotism the former takes place; under any other system of gov- rely on IV approaches. ernment, where political power is shared by 5.2 The Instrumental Variables Approach in any portion of the population, the methods and nature of this control become the prob- Practice lem of prime importance—in fact, the funda- PT attempt to estimate the causal effects mental problem in constitutional law. The social structure by which one type of legisla- of the form of government and electoral sys- tion is secured and another prevented—is a tems by using a simple IV approach. Their secondary or derivative feature arising from potential instruments are three constitution- the nature of the economic groups seeking al dating variables, indicating whether a positive action and negative restraint.

If Beard is correct in his assessment, then 11 A similar argument, though with the opposite impli- many features of constitutions will be influ- cation, is suggested by another major figure in comparative enced by factors that also have a direct political science, Stein Rokkan. Rokkan argued in his 1970 book that the emergence of proportional representation in impact on policy and economic outcomes. Continental Europe was a result of the previous elites’ To illustrate this possibility, imagine a world attempts to manage the transition to democracy in a man- consisting of a group of politically powerful ner that was consistent with their interests. He suggests that a key objective of the proportional representation sys- elite and citizens. The elite oppose redistrib- tem was to make the emergence of a socialist majority ution, while the citizens favor it. When the more difficult. If Rokkan is correct, then because propor- elite are more powerful, they will be able to tional representation emerged in societies where the elites were more powerful, we may expect a negative correlation limit the amount of redistribution and the between proportional representation and redistribution, as size of government. Imagine also that the opposed to the positive correlation documented by PT. de05_Article3 11/16/05 3:59 PM Page 1035

Acemoglu: Constitutions, Politics, and Economics: A Review 1035

country adopted its current form of govern- countries fell into different waves depend- ment and current electoral rule after 1981, ing on when they adopted their constitution between 1951–80, between 1921–50, with or declared their independence. The two before 1921 as the omitted category; two lan- language variables and latitude are included guage variables, indicating the fraction of the as proxies for “European influence” follow- population in the country speaking one of ing the arguments in Hall and Charles I. the major European languages, and the frac- Jones (1999). Hall and Jones use these vari- tion speaking English as a native language; ables as instruments for the overall quality and latitude (distance from the equator). of institutions (social infrastructure), and Although the IV approach adopted by PT PT argue that these could also have affect- is important on methodological grounds, I ed the form of government and electoral will argue that it is unlikely to be estimating rule. The reasoning for these Hall–Jones the causal effect of the form of government instruments is discussed in the next section. and electoral rules on policy and economic These potential instruments are highly outcomes for two reasons: first, because the correlated with the form of democratic insti- three constitutional timing variables have tutions. Typically, countries farther away little effect on the endogenous regressors, from the equator and those with a greater identification ultimately relies on the frac- fraction of the population speaking English tion of the population with European lan- as a native language (but not those with other guage as native tongue and latitude, which European languages) are less likely to have are not convincing instruments for institu- presidential systems. For majoritarian sys- tional features in general; second, even if tems, the effects of English and other they were valid for some general institu- European languages are reversed. There is tional features, by implication they could some effect from the constitutional dating not be valid instruments for specific institu- variables, but these are typically weaker, and tional features such as the form of govern- less consistent (depending on what covariates ment or electoral rules.12 In this are included in the first and second stages). subsection, I show the importance of the The general pattern suggests that with language and latitude instruments for iden- only the constitutional variables, there is no tification, and also briefly look at the impli- first stage for the instrumental variable strat- cations of not including second stage egy.13 This is illustrated in table 1 where I use covariates in the first stage. PT’s data to replicate and further investigate Let us begin with the reasoning for these their first-stage relationships. Columns 1a and instruments. The three timing variables 1b are identical to the first two columns of could be relevant for the form of govern- PT’s table 5.1, and separately estimate OLS ment and for the electoral system because first-stage relationships with dummy variables there may have been waves (or “fads”) in for presidential and majoritarian systems on the type of constitutions, and different the right-hand side (respectively PRES and MAJ). Both columns include seven variables; three dummies for the dates of the adoption 12 Another potential problem is that, as discussed in of the constitution, latitude, fraction of the section 3, a successful IV strategy also requires E(X iui) 0, population speaking English and fraction of i.e., the covariates that are treated as exogenous to be the population speaking another European orthogonal to the error term. These covariates here include income per capita and trade, which are jointly determined with the size and composition of government spending, and their inclusion in the equation may lead to biased estimates. Nevertheless, reestimating PT’s IV mod- 13 Lack of a first-stage relationship even at the conven- els excluding income per capita and trade leads to results tional significance levels is an extreme form of the weak similar to their baseline IV estimates. instruments problem mentioned in footnote 7. de05_Article3 11/16/05 3:59 PM Page 1036

1036 Journal of Economic Literature, Vol. XLIII (December 2005)

language as native languages, and age of are essential in the first-stage relationships democracy (the latter is included as a covari- reported by PT. Columns 3a,b and 4a,b show ate in the second stage equations as well, that they are, and that once these variables while the others are excluded instruments). are left out, there is a very weak or no first- They also report F-tests for joint signifi- stage relationship left. Without the second- cance of all the excluded instruments and stage covariates, the constitution dummies the constitutional variables at the bottom. are significant at 3 percent and 5 percent These two columns show a reasonably respectively in columns 3a and 3b, and when strong first-stage relationships, which are the second-stage covariates are included, then used by PT in their 2SLS estimation they are highly insignificant. (the R2 is 0.48 in the first column and 0.40 This implies that an instrumental-vari- in the second column). However, we can ables strategy relying only on the constitu- see that the major determinants of the tional timing dummies would not achieve endogenous regressors are the Hall–Jones identification. instruments, not the constitution timing Table 2 shows the corresponding 2SLS dummies. In both columns 1a and 1b, the estimates of the form of government and the six instruments are jointly significant at less electoral system on the size of government. than 1 percent, but the three constitutional Column 1 replicates PT’s results (the corre- dummies are only significant at 2 percent in sponding first stages are in columns 1a,b of column 1a (for the presidential dummy) table 1). Column 2 shows that the estimates and very far from statistically significant in are slightly less precise when the second- column 1b (for the majoritarian dummy). stage covariates are included in the first stage, Columns 2a and 2b show that the joint sig- though the dummy for a presidential system nificance of the excluded instruments, and continues to be statistically significant. This especially of the constitutional variables, is shows that leaving the second-stage covari- significantly reduced when the second-stage ates out of the first stage is unlikely to have covariates are also included in the first stage. caused a major bias in this case. Recall that these second-stage covariates are Columns 3 and 4 exclude the Hall–Jones log income per capita, the proportions of the instruments and show that there is no longer population between the ages of fifteen and a significant effect of the form of govern- sixty-four, and over the age of sixty-five, the ment or the electoral system on the size of sum of exports plus imports divided by GDP, government. For example, the coefficient on the average of political and civil liberties the presidential dummy in column 3 is –2.01 indices of Freedom House, and dummies for (standard error = 5.11) as compared to –8.65 a federal structure and OECD (see PT, espe- (standard error = 3.61) in column 1. cially table 6.2 and the data appendix for Evidently, the Hall–Jones instruments are details). As discussed above, PT include essential for PT’s instrumentation strategy. these covariates in the second stage, but not Table 3 shows the 2SLS results for the in the first-stage relationship. Once they are other major outcome variable that PT look at also included in the first stage, the constitu- in their chapter 6, the composition of spend- tional dummies are no longer jointly signifi- ing, proxied by the share of social spending cant for either presidential or majoritarian and welfare in GDP. The structure of this dummies. table is identical to that of table 2. It also I will argue in greater detail in sections 6 shows that the IV estimates of the effect of and 7 that latitude and European languages constitutional features on policy outcomes may not be valid instruments for the form of are sensitive to excluding the Hall–Jones government and electoral rules. In this light, instruments. Furthermore, in this case, it is important to understand whether they including the second-stage covariates in de05_Article3 11/16/05 3:59 PM Page 1037

Acemoglu: Constitutions, Politics, and Economics: A Review 1037 al regimes and 4b include but do erwise. See Persson and Tabellini efficients on CON2150, he coefficients on CON2150, STIMATES E TAGE S Cross-Section OLS IRST : F Dependent Variable is Constitution Dependent Variable Persson and Tabellini (2003) Sample Persson and Tabellini TABLE 1 TABLE ELECTION S ONSTITUTION C (1a) (1b) (2a) (2b) (3a) (3b) (4a) (4b) PRES MAJ PRES MAJ PRES MAJ PRES MAJ MAJ PRES MAJ PRES MAJ PRES PRES (0.138) (0.162) (0.149) (0.161) (0.145) (0.153) (0.132) (0.169) (0.182) (0.236) (0.220) (0.248 (0.167) (0.152) (0.165) (0.222) (0.203) (0.252) (0.245) (0.263) (0.205) (0.183) (0.206) (0.272) (0.335) (0.388) (0.520) (0.619) (0.119) (0.122) (0.127) (0.174) (0.113) (0.134) (0.153) (0.186) (0.308) (0.295) (0.320) (0.245) (0.401) (0.329) (0.326) (0.413) [0.02] [0.67] [0.30] [0.27] [0.02] [0.06] [0.36][0.00] [0.00] [0.10] Observations 78 78 77 77 78 78 77 77 77 78 CON2150 CON5180 78 CON81 LAT01 –0.39 77 –0.161 –0.070 –0.132 0.095 –0.323 –0.095 0.302 0.277 0.169 –0.120 0.074 0.022 –0.319 0.197 0.131 ENGFRAC EURFRAC 77 0.266 0.055 –0.029 0.161 0.238 0.589 0.004 0.358 AGE –0.644 1.035 –0.692 0.916 Second stage covariates 78 F-test on all constitution variables –1.366 –0.884 –0.786 –0.410 0.425 –0.349 0.431 –0.409 3.66 F-test on all instruments NO Observations 78 0.540 0.197 0.703 0.083 0.52 0.280 0.150 0.602 0.358 24.39 NO 1.23 26.69 YES 1.35 7.97 YES 3.35 12.07 NO 2.61 NO 1.09 YES 2.20 YES R-squared Robust standard errors in parentheses. F-test on all constitution variables refers to the F-statistic for joint test that t CON5180, and CON81 are each zero. F-test on all instrumental variables refers to the F-statistic for joint test that co ENGFRAC, and EURFRAC are each zero. P-value for the F-statistic is in brackets. Columns 2a, 2b, 4a, CON5180, CON81 , LAT01, TRADE, PROP1564, PROP65, GASTIL, FEDERAL, and OECD. PRES is a dummy variable which equal to 1 in presidenti not report LYP, 0.48 0.40 (2003) for data and definitions. 0.55 0.48 0.18 0.07 0.41 0.23 zero otheriwse. MAJ is a dummy variable which equal to 1 if all the lower house elected under plurality rule and oth de05_Article3 11/16/05 3:59 PM Page 1038

1038 Journal of Economic Literature, Vol. XLIII (December 2005)

TABLE 2 SIZE OF GOVERNMENT AND CONSTITUTIONS: SECOND STAGE INSTRUMENTAL VARIABLE ESTIMATES

Persson and Tabellini (2003) Sample Cross-Section 2SLS (1) (2) (3) (4) Dependent Variable is Central Government Expenditure as % GDP PRES –8.653 –8.653 –2.005 –4.098 (3.614) (4.713) (5.108) (9.183) MAJ –3.901 –4.233 –9.384 –9.194 (3.448) (3.466) (6.900) (7.896) Constitution instruments in first stage YES YES YES YES Hall-Jones instruments in first stage YES YES NO NO Second stage covariates in first stage NO YES NO YES Observations 75 75 75 75 Standard errors in parentheses. All columns include but do not report AGE, LYP, TRADE, PROP1564, PROP65, GASTIL, FEDERAL, and OECD in the second stage. All columns include CON2150, CON5180, CON81, and AGE in the first stage. Columns 1 and 2 include LAT01, ENGFRAC, EURFRAC in the first stage. Columns 2 and 4 include all second stage covariates in the first stage. See Table 1 for first stage regressions. PRES is a dummy variable which is equal to 1 in presidential regimes and zero otheriwse. MAJ is a dummy variable which is equal to 1 if all the lower house is elected under plurality rule and zero otherwise. See Persson and Tabellini (2003) for data and definitions.

the first stage also matters to some extent, touches upon the importance of including and the effect of a majoritarian electoral second-stage covariates in the first stage of system is no longer significant at 5 percent the IV estimation). The implication is that in column 2. the IV strategy in PT’s book heavily relies This brief discussion illustrates the on the validity of the Hall–Jones instru- importance of the Hall–Jones instruments ments. Could these instruments be valid in for PT’s identification strategy,14 (and also the current context? I believe there are two fundamental rea- sons for these instruments not to be valid. 14 PT are upfront about this issue, and argue that The first is that the Hall–Jones instruments “Admittedly, these [Hall–Jones] variables could be corre- are unlikely to be valid for the overall qual- lated with other unobserved historical determinants of fis- cal policy or corruption . . . [As] we are confident about the ity of institutions. The second is that even exogeneity of the time dummies for constitutional adop- if these instruments were valid for the tion, we can test the validity of the additional instruments overall quality of institutions, they would by exploiting the overidentifying restrictions” (p. 131). In addition, the parametric selection models that they report by implication be invalid for a specific fea- also yield very similar estimates to the linear IV estimates. ture of the institutional structure such as Nevertheless, these checks are unlikely to be sufficient to presidentialism or a majoritarian electoral validate the IV estimates. The overidentifying restrictions have very little power, since, as we saw above, the consti- system. I discuss these two issues in the tutional timing variables have little explanatory power in next two sections. This discussion is not the first stage. The parametric selection models, on the only relevant to PT’s work, but also to a other hand, also use the same exclusion restrictions in addition to the parametric restrictions, so the similarity in number of recent papers adopting similar the results is not too surprising. strategies. de05_Article3 11/16/05 3:59 PM Page 1039

Acemoglu: Constitutions, Politics, and Economics: A Review 1039

TABLE 3 COMPOSITION OF GOVERNMENT SPENDING AND CONSTITUTIONS: SECOND STAGE INSTRUMENTAL VARIABLE ESTIMATES

Persson and Tabellini (2003) Sample Cross-Section 2SLS (1) (2) (3) (4) Dependent Variable is Central Government Expenditure on Social Services and Welfare as % GDP PRES 0.296 0.527 0.615 1.585 (1.791) (2.609) (2.461) (4.734) MAJ –3.633 –3.645 –3.920 –3.857 (1.663) (1.928) (2.493) (3.041) Constitution instruments in first stage YES YES YES YES Hall-Jones instruments in first stage YES YES NO NO Second stage covariates in first stage NO YES NO YES Observations 64 64 64 64 Standard errors in parentheses. All columns include but do not report AGE, LYP, TRADE, PROP1564, PROP65, GASTIL, FEDERAL, and OECD in the second stage. All columns include CON2150, CON5180, CON81, and AGE in the first stage. Columns 1 and 2 include LAT01, ENGFRAC, EURFRAC in the first stage. Columns 2 and 4 include all second stage covariates in the first stage. See Table 1 for first stage regressions. PRES is a dummy variable which is equal to 1 in presidential regimes and zero otheriwse. MAJ is a dummy variable which is equal to 1 if all the lower house is elected under plurality rule and zero otherwise. See Persson and Tabellini (2003) for data and definitions.

6. Are the Hall–Jones Variables Valid This makes it useful to reconsider the the- Instruments for Institutions? oretical justification of these instruments. What is the theory underlying the Hall and Jones wrote a very important Hall–Jones instruments? In broad strokes, and influential article emphasizing the Hall and Jones’s theory is as follows. “Good importance of social infrastructure, or insti- institutions” originated in Western Europe tutions on aggregate productivity and eco- and spread from there to other countries. nomic growth. Hall and Jones suggested Therefore, a potential determinant of the that the origins of good institutions around quality of institutions is the extent to which a the world lie in Europe, so “proximity” to country has been influenced by Europe’s Europe, as measured by the fraction of culture, values and institutions. Hall and those speaking European languages and lat- Jones isolate two channels of European itude, can be used as instruments for the influence. The first is through a shared lan- quality of institutions. guage, and the second is through geography. Hall and Jones’s paper has already Consequently, they argue, countries with a become a classic. It is not only cited exten- greater fraction of the population speaking sively, but their arguments are often invoked European languages and those farther from to use latitude and the fraction of the popu- the equator, which were less densely popu- lation speaking European languages as lated and geographically more similar to native tongues as instrument for various Europe hence more conducive to European institutional features in the recent empirical migration, have benefited from a literature. de05_Article3 11/16/05 3:59 PM Page 1040

1040 Journal of Economic Literature, Vol. XLIII (December 2005)

benign/beneficial European influence. Islands clearly illustrates the adverse effects Based on this theory, they use the fractions of Europeans, which set up repressive of the population speaking European lan- regimes based on slavery and forced labor. guages and distance from the equator (lati- Instead, the evidence is more consistent tude) as instruments for the overall quality with the theory that European influence of institutions. could be either beneficial or harmful to Some studies, including PT, cite my work institutional development. Whether or not with Simon Johnson and James A. Robinson it was beneficial depended on the as supporting this view. Nevertheless, the European powers’ colonization strategy. theory, and therefore the instrumentation Is it possible that the use of these instru- strategy, in Acemoglu, Johnson, and ments leads to consistent estimates, even Robinson (AJR) (2001) is different. First, in though the story for identification is not AJR we focus on former European colonies entirely substantiated? There is an argument where European powers had a large effect to be made here. For example, AJR (2001) on institutional development, often by show that in the sample of former European imposing the institutions and running these colonies, once mortality of European settlers areas as colonies. Second, we do not argue is used as an instrument for institutions, lat- that European influence was positive (or itude has no additional effect on income per negative). Instead, the argument is that capita today. It may then be argued that lat- Europeans had very different effects on dif- itude could be used as an instrument and ferent colonies depending on what was the would capture the same sort of variation as most attractive colonization strategy. In settler mortality. This argument is not com- places where they faced high mortality rates, pelling, however, mostly because latitude is they could not settle, and they were more being used as an instrument for the whole likely to opt for an extractive strategy, with world sample whereas the “experiment” in associated extractive/bad institutions. In question, European colonization, applies places with low mortality rates, they were only to the former European colonies. There more likely to settle, and good institutions is no valid theoretical argument for extend- were more likely to emerge, because these ing this experiment to the entire world (and were institutions that Europeans themselves for the sample of former European colonies, would live under. This contrast is illustrated there is no reason to use latitude instead of by a comparison of the United States or settler mortality). For example, for other Australia to the European colonies in countries not affected by European colo- Central America, the Caribbean, South Asia, nization, geographic considerations may or Africa. In a companion paper, AJR (2002), have had other, different effects, which may we documented the similar effect of indige- or may not be orthogonal to omitted deter- nous population density. Europeans were minants of the outcome variables (see, for more likely to settle and less likely to pursue example, AJR 2004 on the changing effect of extractive strategies in colonies that were access to the Atlantic for European nations less densely populated. during the early modern period). Let us now revisit the theoretical founda- Consequently, there is no justification for tions of the Hall–Jones instruments in the using latitude, or for that matter anything to light of this discussion. Neither the results do with settler mortality rates, in a sample in these papers nor historical evidence sup- that includes countries that were themselves port the theory that European influence European or were never European colonies. was generally beneficial (see also Stanley L. Overall, the theoretical foundations of the Engerman and Kenneth L. Sokoloff 1997). Hall–Jones instruments are not entirely Indeed, the history of the Caribbean compelling, especially when used for the de05_Article3 11/16/05 3:59 PM Page 1041

Acemoglu: Constitutions, Politics, and Economics: A Review 1041

entire world, and consequently, there are and wealth cannot easily survive with a set of good reasons to suspect that they may not be political institutions that distribute political excludable from the regressions of interest. power equally. Those with political power would be greatly tempted to use their power to redistribute income and change the eco- 7. Cluster of Institutions Versus Specific nomic institutions in line with their interests. Institutions Economic institutions that lead to a very The above discussion centered around the unequal distribution of income and wealth question of whether the Hall–Jones variables are only consistent with a similarly unequal are valid instruments for the overall quality distribution of political power, i.e., with dic- of institutions. I now turn to a discussion of tatorships and other repressive regimes. In the problems involved in using these vari- this case, sources of variation that affect a ables as instruments for specific institutions. broad cluster of institutions (e.g., economic This has become a common practice in the and political institutions together) would not newly flourishing empirical political econo- be useful in identifying the role of specific my literature. PT also follow this practice institutional features. and use these variables as instruments for As an example of the difficulty of this type the form of government and electoral rules of strategy to estimate the effect of specific (indicators of presidential or majoritarian institutions, consider the quasi-natural systems). I will argue that there are serious experiment due to international politics, the problems in this procedure because of inher- division of Korea into North and South. The ent complementarities between different two parts of Korea before the division were types of institutions. ethnically, culturally, economically, and To develop this argument, let us put aside socially very similar. But because of the the concerns raised in the previous section geopolitical balance between the United and suppose that the Hall–Jones instruments States and Soviet Union, the South ended up (or perhaps settler mortality rates in the for- largely capitalist, while the North became mer European colonies sample) are valid for communist. In the following forty years, we the overall quality of institutions (or the witnessed a large divergence between these broad cluster of institutions). This means we two countries. This is a good source of varia- now suppose that they are excludable from a tion to understand the effect of the broad second-stage regression of economic out- cluster of institutions at the level of “quasi- comes (such as aggregate output or produc- capitalist” versus “communist” systems. tivity) with the broad measure of institutions Suppose now that we try to use this source of as the endogenous regressor. variation to understand the effect of some Here the distinction between a broad specific institutional feature, say financial cluster of institutions and specific institu- development, on economic growth. It tions is crucial. In AJR (2001), we defined a should be clear that this strategy will lead to broad cluster of institutions as a combination a highly biased estimate. It is true that South of economic, political, social and legal insti- Korea is financially more developed than tutions that are mutually reinforcing. For North Korea. It is also true that the reason example, it is impossible to think of a system for this is the division in 1946 (had it not like the plantation economies in the been for the division, the North and the Caribbean Islands until the nineteenth cen- South would probably have similar levels of tury together with democratic political insti- financial development). But this does not tutions. This is because a set of economic make the division a good experiment to institutions, like the plantation system, that understand the effect of financial develop- lead to a very unequal distribution of income ment, because this division also caused many de05_Article3 11/16/05 3:59 PM Page 1042

1042 Journal of Economic Literature, Vol. XLIII (December 2005)

other institutional changes. It is a good labo- Next, suppose that we have an instrument

ratory for the study of broad institutions, but Zi, in particular a country-specific variable, not for a study of the specific institutions. which is potentially related to each of the Another example directly related to PT’s specific institutions, i.e., empirical work may also be useful. In the kk k (9) SZ=+ζυ, first-stage relationships shown in table 1, the i ii fraction of the population speaking English is and is a valid instrument for the cluster of = a strong predictor of a majoritarian system, institutions, Gi, i.e., Cov(Zi,ui) 0. So if we and this fraction may be a predetermined have a good measure of Gi, equation (7) can variable, shaped, for example, by colonial his- be estimated consistently by IV. tory. Nevertheless, this is not sufficient for it However, the objective here is to estimate to be an excludable instrument in estimating the effect of specific institutions. To simplify µ the causal effect of majoritarian systems, the discussion, also assume that Cov(Zi, i) = k = k = even if as Hall and Jones claim, English influ- 0, and Cov( i,vi) Cov(Zi,vi) 0 for all k k j = ≠ k ence is conducive to the development of and Cov(vi,vi) 0 for all j k. Note that S i’s k good institutions overall. For example, as we are correlated even if vi ’s are independent, know from Rafael La Porta, Florencio since they are all affected by Zi. This corre- Lopez-de-Silanes, Andrei Shleifer, and lation is at the root of the identification Robert W. Vishny (1998), countries more problems facing IV estimates of the impact influenced by the English heritage also have of specific institutions. more developed financial markets. So the In the last section, I questioned whether correlation between majoritarian electoral latitude could play the role of Zi to estimate systems (instrumented by fraction of the the structural relationship in (7). Now let us population speaking English) and the size of put aside the issues raised there, and sup- = government may reflect the effect of finan- pose that indeed Cov(Zi,ui) 0. Can we then cial development, which is omitted from the use Zi as an instrument for a specific institu- regressions. tion, say S1? The answer is no. If we were to To make these issues a little more precise, do this, all of the Sk’s would also load onto S1. consider the following structural model: More explicitly, we are now estimating =+1 1 β 1 +1 =++ β YSiiiiX u (7)YGiiiiX u , with IV using Z as the excluded instru- Y β i where i denotes the outcome of interest, ment. Given (7) above, the true value of α1 is again a vector of coefficients associated is 1 = 1. Moreover, from equations (8) with X, which is now assumed to stand for a and (9), set of noninstitutional covariates. Gi is a K measure of a broad “cluster of institutions.” 1 k k (10) uu=+∑νµ S + . Moreover, suppose that there are K > 1 specif- ii i i k=2 ic institutions, each denoted by Sk, and We can now see that since, by construction, K K νk k t α1,IV k k S is correlated with Z , plim (8) GS=+∑νµ k 2 i i i i i 1,IV. In particular k=1 ν k where denotes the effect of specific insti- 1 k Cov (Zu, ) tution S on the cluster of institutions. plimˆ 11,IV =+ ii ≠ ζ 1 ( ) Suppose that Cov(ui, i) 0 and/or Var Zi k ≠ Cov(ui,S i) 0 for some k, so that equation K kk ∑ = νζ Var(Z ) (cluster) cannot be estimated consistently =+1 k 2 i ζ 1 ( ) by OLS. Var Zi de05_Article3 11/16/05 3:59 PM Page 1043

Acemoglu: Constitutions, Politics, and Economics: A Review 1043

K kk ∑ = νζ and presidential systems on the amount of =+1 k 2 ,, ζ 1 redistribution, political rents and aggregate productivity. So one might argue that judged on the basis of their strong objective of esti- where the second equality exploits (10) and = = mating causal effects, the book is only a par- the fact that Cov(Zi,ui) Cov(Zi, i) 0. tial success. This would be the wrong Note that in the most extreme case where conclusion, however. 1 → = 0, even though Cov(Zi,ui) 0, we can When the body of work in the book is get arbitrarily biased estimates of the effect taken as a whole, it is a tremendous success. 1 of S on the outcome of interest. The correlations that PT document between This discussion also makes it clear that the the form of government and electoral sys- problem of instrumenting for a specific tems and various economic outcomes are 1 institution, such as Si, is in many ways simi- very important. Few comparative political lar to the omitted variable bias, since other economy papers can be written from now on specific institutions that make up the cluster that do not take these stylized facts serious- of institutions, Gi, are omitted from the ly. PT have not necessarily estimated the regression. Even if we include proxies for causal effects of the form of government and some of them, unless we can correctly esti- electoral systems. But even noncausal but mate the causal effects of all of those, IV robust relationships are important and valu- regressions will fail to estimate the causal able inputs into our thinking and into our effect of the specific institution of interest, models. 1 Si, consistently. Therefore, I believe that overall PT have For the identification strategy of using Zi largely achieved their ambitious aim of revo- 1 as an instrument for S to be valid, we need lutionizing comparative political economy, k = = in addition that 0 for all k 2, . . . K. and this book is the most significant contri- Consequently, even if we were convinced bution to this field since Lipset’s work almost that the Hall–Jones instruments are valid for fifty years ago. PT have not only pushed the broad cluster of institutions, there is lit- comparative political economy forward, but tle justification for using them as an instru- they have provided a set of findings that will ment for specific institutions such as challenge all economists and social scien- presidential or majoritarian systems. This tists, and likely pave the way for a large body discussion implies that in IV approaches in of new work in this area. With such an political economy, there is often a first-order important contribution, it is then the right question of unbundling, meaning going from time to wonder what will (and perhaps an understanding of the role of a broad clus- should) come next. ter of institutions to pinpointing which spe- cific institutions are more important for the economic outcomes of interest. I discuss this 9. Unbundling Institutions problem further below. I argued above that a source of variation in the broad cluster of institutions is not 8. What Have We Learned? sufficient to separately estimate the effects of specific institutional features. In other I have so far argued that some of the com- words, we can find clever instruments, mon empirical strategies in the political from history, sometimes from geography, economy literature will have difficulty in or international politics, that affect the uncovering the causal effect of specific insti- whole social organization of a society, but tutions. Relatedly, PT’s work may not have this is only the first step. It does not enable estimated the causal effect of majoritarian us to conclude that one specific institution de05_Article3 11/16/05 3:59 PM Page 1044

1044 Journal of Economic Literature, Vol. XLIII (December 2005)

is more important than another. tions at the colonies on institutional devel- However, what we want to know in prac- opment together with the strategy of La tice is not only that “institutions” (defined Porta, Lopez-de-Silanes, Shleifer, and as a broad cluster, and therefore almost nec- Vishny (1998), which exploits differences essarily as a black box) matter, but which in the identity of the colonizer. While local specific dimensions of institutions matter conditions, in particular the disease envi- for which outcomes. It is only the latter type ronment and the indigenous population of knowledge that will enable better theo- density at the time of colonization, affect ries of institutions to be developed and the property rights institutions, the identi- practical policy recommendations to ty of the colonizer is important for the emerge from this new area. Consequently, legal origin and therefore for the con- the issue of “unbundling institutions,” that tracting institutions. We show that there is, understanding the role of specific com- are almost perfectly “separable” first- ponents of the broad bundle, is of first stages whereby the identity of the coloniz- order importance. er has almost no effect on property rights We therefore need to find other strategies, institutions, while local conditions have no even more clever instruments, or other, per- effect on contracting institutions. This haps new, econometric techniques to decide enables a multiple IV strategy to separate- which specific dimensions of these institu- ly identify the effects of both types of tions matter. I believe that there is going to institutions. The results suggest that while be a lot of exciting research in this area in property rights institutions have a first- the next ten years. Let me discuss a couple order impact on all aspects of economic of potential questions, and in some cases, and financial development, contracting potential avenues for investigation. institutions mainly affect the form of 1) Contracting institutions versus property financial intermediation (and have no rights institutions: a key question that effect on long-run economic growth, on emerges from works that emphasize the investment and on the overall amount of importance of institutions concerns the financial development). relative importance and specific effects of 2) Financial development versus contract- institutions that regulate interactions ing institutions: a similar unbundling between private citizens (“contracting issue arises in evaluating the role of finan- institutions”) and those that constrain the cial development (and financial institu- behavior of political and economic elites tions) and economic development. A (“property rights institutions”). Evidence large literature surveyed in Ross Levine that a broad cluster of institutions matters (1997, 2004) documents a robust correla- for long-run economic development is tion between financial development and consistent with the primary role of both economic growth. But is this the effect of types of institutions. Moreover, there are financial development, or the direct many reasons to expect societies that effect of better contracting institutions have better property rights institutions to that enable financial development in the also have better contracting institutions, first place? More explicitly, better con- so these two types of institutions are like- tracting institutions will lead to better ly to covary in practice. So unbundling in nonfinancial contracts (such as between this context is very important. downstream and upstream firms, firms Acemoglu and Johnson (2004) attempt and workers, etc.) as well as to improved this type of unbundling. The idea is to financial contracts, so in the data when combine the strategy in AJR (2001, 2002), we see a society with greater financial which exploits the effect of local condi- development, it will also have better de05_Article3 11/16/05 3:59 PM Page 1045

Acemoglu: Constitutions, Politics, and Economics: A Review 1045

nonfinancial contracts. This makes it dif- Indian states, by Thomas J. Holmes ficult to identify the effect of financial (1998), who investigates the implications development. of union laws on the location of manu- A potential strategy for unbundling in this facturing plants in the United States, by context might be to investigate the effect Marianne Bertrand and Francis Kramarz of purely financial reforms that do not (2002), who look at labor market effects change the scope for nonfinancial con- of product market regulations, and by tracts. Reforms in capital accounts that Jakob Svensson (2003), who exploits dif- are imposed (or “encouraged”) by the ferences across industries to investigate IMF or other international bodies, with- the amount of bribes that firms in out other reforms, could provide one Uganda have to pay to officials useful source of variation. 5) Formal versus informal political institu- 3) Economic versus political institutions: an tions: finally, returning to the theme of even more formidable task is to separate- PT’s book, how important are formal ly identify the effects of economic and institutions? PT’s work suggests that they political institutions. There is a clear com- are essential. However, in practice, for- plementarity between these two types of mal and informal institutions are highly institutions. For example, an economic correlated, and one may conjecture that system like slavery or forced labor cannot changing constitutions or formal rules of be sustained in a democratic society. political decision-making might have lim- Nevertheless, many relevant theoretical ited effects if society at large expects the and policy questions require the potential changes not to be durable or not to be effects of economic and political institu- obeyed. The important results in this tions to be unbundled. For example, book notwithstanding, this is also an what would be the effects of improving important area for future research. economic institutions under a given set of political institutions, for example as in 10. Weakly Institutionalized Polities China? This is an area for future Another area for future research is what research. might be referred to as politics in “weakly 4) Different types of economic institutions: institutionalized polities.” The body of work the question is which economic institu- that PT have surveyed and developed further tions matter more. Entry barriers? Labor in their two books is largely for the analysis of market regulations? Property rights “strongly institutionalized polities,” where enforcement? Limits on government political institutions make politicians, at least corruption? These are important, but partially, accountable to citizens. It is not sur- also very difficult questions. It seems prising that this has been the first focus of that these questions will be almost Western academics, who almost all live in impossible to answer with cross-country strongly institutionalized polities. But the data alone, and micro data investigations, same is not true for a large fraction of the pop- for example, exploiting differences in ulation of the world. The situation in many regulations across markets and regions countries in Africa, Central America, and the appear to be the most promising avenue. Caribbean corresponds much more clearly to Interesting recent work using within- one of “weakly institutionalized polities,” country variation to look at some of these where state–society relations are fundamen- issues include, among others, work by tally different. Examples of this include, but Timothy Besley and Robin Burgess are not limited to, the extreme kleptocratic (2004), who look at the implications of regimes of Mobutu in the Democratic different labor market regulations across Republic of the Congo (Zaire), Rafael Trujillo de05_Article3 11/16/05 3:59 PM Page 1046

1046 Journal of Economic Literature, Vol. XLIII (December 2005)

in the Dominican Republic, the Duvaliers in owing to the absence of strong institutions, Haiti, the Somozas in Nicaragua, Charles rulers can deploy strategies, in particular Taylor in Liberia, and Ferdinand Marcos in “divide-and-rule,” to defuse opposition to the Philippines. their regime. The logic of the divide-and-rule Much historical evidence suggests that a strategy is to enable a ruler to bribe political- systematic study of the political economy of ly pivotal groups off the equilibrium path, such regimes must depart from some of the ensuring that he can remain in power against modeling approaches of politics in strongly challenges. By providing selective incentives institutionalized polities. While in strongly and punishments, the divide-and-rule strate- institutionalized polities, formal political insti- gy exploits the fragility of social cooperation tutions, such as the constitution, the structure in weakly-institutionalized polities: when of the legislature, or electoral rules, place faced with the threat of being ousted, the constraints on the behavior of politicians and kleptocratic ruler intensifies the collective political elites, and directly influence political action problem and destroys the coalition outcomes, the same does not appear to be the against him by bribing the pivotal groups. case in weakly institutionalized polities. A different and innovative answer is given Instead, the nature of politics appears to be by Gerard Padro-i-Miquel (2004), who also different between strongly and weakly institu- constructs a dynamic model of politics. tionalized polities. Most importantly, when Ethnic divisions are the key feature in his institutions are strong, citizens have the model. Each ethnic group is afraid of replac- power to punish politicians by voting them ing their own leader when in power, because out of power; when institutions are weak, this increases the probability of a switch of politicians pursue clientelistic policies that power from their own ethnic group to a rival punish citizens who fail to support them (see group. This makes the standard method of Robert H. Jackson and Carl G. Rosberg 1982 controlling political elites in strongly-institu- and Acemoglu, Robinson, and Thierry tionalized polities ineffective, and enables Verdier 2004). leaders to not only exploit other ethnic In modeling politics in weakly institutional- groups but also their own ethnic group. ized polities, the first question that emerges is Padro-i-Miquel shows how this framework theoretical: how can we understand policy can account for a puzzling feature of African making and collective decisions in such soci- politics first highlighted in Robert H. Bates’ eties? There are only a few papers that classic study Markets and States in Tropical attempt to develop answers this question. In Africa (1981): the simultaneous use of ineffi- Acemoglu, Robinson, and Verdier (2004), we cient transfers to and taxes on the same construct a dynamic model of politics in group.16 In the logic of Padro-i-Miquel’s weakly institutionalized polities based on the model, this strategy makes sense because idea of “divide and rule.” The key focus is to leaders need to keep their own group happy understand how kleptocratic regimes that to remain in power. This sets a limit on the impoverish their citizens can remain in power amount of net taxes they can impose on their for so long.15 The answer we suggest is that, 16 Bates described the web of inefficient transfers and 15 Models of strongly institutionalized polities would policies in effect in many parts of Africa, but most notably suggest that poorly performing leaders should be replaced in Ghana and Zambia. For example, the Ghanaian govern- more often. In contrast, many disastrous kleptocracies last ment heavily taxed cocoa producers, while at the same for long periods; Mobutu ruled for thirty-two years, time subsidizing their inputs of seeds and fertilizers. While Trujillo for thirty-one, and the Somozas for forty-two the simultaneous use of taxes and subsidies may be to alle- years. This longevity is made even more surprising by the viate the negative effects of taxes on investments, the fact that many kleptocratic regimes lack both a core interpretation in the literature is that the extent of these constituency of supporters and a firm command of the practices in sub-Saharan Africa is beyond what can be jus- military. tified as an optimal mix of taxes and subsidies. de05_Article3 11/16/05 3:59 PM Page 1047

Acemoglu: Constitutions, Politics, and Economics: A Review 1047

own group. However, taxes across groups are and those documented by PT appear to be linked, since much lower taxes on one group highly robust and of central importance for will encourage other groups to switch eco- our theoretical understanding. Few political nomic activity. This then motivates leaders to economy papers can be written from now inefficiently subsidize their own ethnic group on that do not take this book, both its so that they can increase the tax rate on their methodological and empirical contributions, own ethnic group, and consequently tax seriously. other groups more intensively. Equally important, with the addition of A second crucial area is to construct this book, (comparative) political economy models to understand how weak institu- has taken one more step toward establishing tions can be strengthened. Although there itself as a major field of economics, and it is now a number of formal models of the offers exciting and important research areas creation and consolidation of democracy for further inquiry. (see the literature review in Acemoglu and REFERENCES Robinson 2005), many of these models are Acemoglu, Daron, and Joshua D. Angrist. 2000. “How still motivated by the practice of strongly Large Are Human-Capital Externalities? Evidence from Compulsory Schooling Laws,” in NBER institutionalized polities. Macroeconomics Annual 2000. Ben S. Bernanke and Finally, there is almost no work on how Kenneth Rogoff, eds. Cambridge: MIT Press, 9–59. changes in formal political institutions affect Acemoglu, Daron, and Simon Johnson. 2003. “Unbundling Institutions.” NBER Working Paper policy and economic outcomes in weakly 9934. institutionalized polities. These are all fruitful Acemoglu, Daron, Simon Johnson, and James A. areas for future research. Robinson. 2001. “The Colonial Origins of Comparative Development: An Empirical 11. Conclusion Investigation.” American Economic Review, 91(5): 1369–1401. This book is an important landmark in Acemoglu, Daron, Simon Johnson, and James A. Robinson. 2002. “Reversal of Fortune: Geography political economy. It takes a body of theoreti- and Institutions in the Making of the Modern World cal literature seriously and carefully confronts Income Distribution.” Quarterly Journal of their predictions with data. PT set themselves Economics, 117(4): 1231–94. Acemoglu, Daron, and James A. Robinson. 2005. the very ambitious goal of estimating the Economic Origins of Dictatorship and Democracy. causal effects of constitutions, in particular, Cambridge: Cambridge University Press. the form of government and electoral rules. Acemoglu, Daron, James A. Robinson, and Thierry Verdier. 2004. “Kleptocracy and Divide-and-Rule: A Bearing in mind the difficulties of estimat- Model of Personal Rule.” Journal of the European ing causal effects in social sciences, especial- Economic Association, 2(2): 162–92. ly in cross-country data, the book is a Angrist, Joshua D., and Guido W. Imbens. 1995. “Two- Stage Least Squares Estimation of Average Causal tremendous success. It documents impor- Effects in Models with Variable Treatment tant and robust correlations between the Intensity.” Journal of the American Statistical form of government and electoral systems, Association, 90(430): 431–42. Angrist, Joshua D., Guido W. Imbens, and Donald B. on the one hand, and various policy and eco- Rubin. 1996. “Identification of Causal Effects Using nomic outcomes, on the other. It also uses Instrumental Variables.” Journal of the American state-of-the-art econometric techniques to Statistical Association, 91(434): 444–55. Angrist, Joshua D., and Alan B. Krueger. 1991. “Does estimate causal effects. Compulsory School Attendance Affect Schooling I argued above that these methods may and Earnings?” Quarterly Journal of Economics, not have been sufficient to arrive at causal 106(4): 979–1014. Angrist, Joshua D., and Alan B. Krueger. 1999. effects, both because of certain conceptual “Empirical Strategies in Labor Economics,” in and practical problems. Nevertheless, the Handbook of Labor Economics Volume 3A. Orley achievement here should not be underesti- Ashenfelter and David Card, eds. Amsterdam: North-Holland, 1277–1366. mated. Even noncausal robust relationships Arrow, Kenneth J. 1951. Social Choice and Individual are rare in comparative political economy, Values. New York: Wiley. de05_Article3 11/16/05 3:59 PM Page 1048

1048 Journal of Economic Literature, Vol. XLIII (December 2005)

Austen-Smith, David, and Jeffrey S. Banks. 1999. Holmes, Thomas J. 1998. “The Effect of State Policies Positive Political Theory I: Collective Preferences. on the Location of Manufacturing: Evidence from Ann Arbor: University of Michigan Press. State Borders.” Journal of Political Economy, 106(4): Bates, Robert H. 1981. Markets and States in Tropical 667–705. Africa. Berkeley: University of California Press. Hotelling, Harold. 1929. “Stability in Competition.” Barro, Robert. 1997. Determinants of Economic Economic Journal, 39: 41–57. Growth. Cambridge: MIT Press. Jackson, Robert H., and Carl G. Rosberg. 1982. Bertrand, Marianne, and Francis Kramarz. 2002. Personal Rule in Black Africa. Berkeley: University “Does Entry Regulation Hinder Job Creation? of California Press. Evidence from the French Retail Industry.” La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Quarterly Journal of Economics, 117(4): 1369–1413. Shleifer, and Robert W. Vishny. 1998. “Law and Beard, Charles A. 1913. An Economic Interpretation of Finance.” Journal of Political Economy, 106(6): the Constitution of the United States. New York: The 1113–55. Free Press. Levine, Ross. 1997. “Financial Development and Besley, Timothy, and Robin Burgess. 2004. “Can Labor Economic Growth: Views and Agenda.” Journal of Regulation Hinder Economic Performance? Economic Literature, 35(2): 688–726. Evidence from India.” Quarterly Journal of Levine, Ross. Forthcoming. “Finance and Growth: Economics, 119(1): 91–134. Theory, Evidence, and Mechanisms,” in Handbook Black, Duncan. 1948. “On the Rationale of Group of Economic Growth. Philippe Aghion and Steven Decision-Making.” Journal of Political Economy, Durlauf, eds. Amsterdam: Elsevier. 56(1): 23–34. Lindbeck, Assar, and Jörgen Weibull. 1987. “Balanced- Bourguignon, François, and Thierry Verdier. 2000. Budget Redistribution as the Outcome of Political “Oligarchy, Democracy, Inequality and Growth.” Competition.” Public Choice, 52(3): 272–97. Journal of Development Economics, 62(2): 285–313. Padro-i-Miquel, Gerard. 2004. “The Control of Dixit, Avinash, Gene M. Grossman, and Faruk Gul. Politicians in Divided Societies: Politics of Fear.” 2000. “The Dynamics of Political Compromise.” MIT. Mimeo. Journal of Political Economy, 108(3): 531–68. Persson, Torsten, and Guido Tabellini. 2000. Political Downs, Anthony. 1957. An Economic Theory of Economics: Explaining Economic Policy. Cambridge: Democracy. New York: Harper and Row. MIT Press. Engerman, Stanley L., and Kenneth L. Sokoloff. 1997. Persson, Torsten, and Guido Tabellini. 2003. The “Factor Endowments, Institutions, and Differential Economic Effects of Constitutions: What do the Data Paths of Growth among New World Economies: A Say? Cambridge: MIT Press. View from Economic Historians of the United Persson, Torsten, Gerard Roland, and Guido Tabellini. States,” in How Latin America Fell Behind: Essays 1997. “Separation of Powers and Political on the Economic Histories of Brazil and Mexico, Accountability.” Quarterly Journal of Economics, 1800-1914. Stephen Haber, ed. Stanford: Stanford 112(4): 1163–1202. University Press, 260–304. Persson, Torsten, Gerard Roland, and Guido Tabellini. Ferejohn, John. 1986. “Incumbent Performance and 2000. “Comparative Politics and Public Finance.” Electoral Control.” Public Choice, 50(1–3): 5–25. Journal of Political Economy, 108(6): 1121–61. Hall, Robert E. 1978. “Stochastic Implications of the Rokkan, Stein. 1970. Citizens, Elections, Parties: Life Cycle-Permanent Income Hypothesis: Theory Approaches to the Comparative Study of the and Evidence.” Journal of Political Economy, 86(6): Processes of Development. New York: McKay. 971–87. Staiger, Douglas, and James H. Stock. 1997. Hall, Robert E., and Charles I. Jones. 1999. “Why Do “Instrumental Variables Regression with Weak Some Countries Produce So Much More Output Per Instruments.” Econometrica, 65(3): 557–86. Worker Than Others?” Quarterly Journal of Stigler, George J. 1970. “Director’s Law of Public Economics, 114(1): 83–116. Income Redistribution.” Journal of Law and Hansen, Christian, Jerry Hausman, and Whitney Economics, 13(1): 1–10. Newey. 2004. “Weak Instruments, Many Instruments Stigler, George J. 1972. “Economic Competition and and Microeconomic Practice.” MIT. Mimeo. Political Competition.” Public Choice, 13: 91–106. Hansen, Lars Peter, and Kenneth J. Singleton. 1982. Stock, James H., Jonathan H. Wright, and Motohiro “Generalized Instrumental Variables Estimation Yogo. 2002. “A Survey of Weak Instruments and of Nonlinear Rational Expectations Models.” Weak Identification in Generalized Method of Econometrica, 50(5): 1269–86. Moments.” Journal of Business and Economic Heckman, James J. 1976. “The Common Structure of Statistics, 20(4): 518–29. Statistical Models of Truncation, Sample Selection Svensson, Jakob. 2003. “Who Must Pay Bribes and How and Limited Dependent Variables and a Simple Much? Evidence from a Cross Section of Firms.” Estimator for Such Models.” Annals of Economic Quarterly Journal of Economics, 118(1): 207–30. and Social Measurement, 5(4): 475–92. Wooldridge, Jeffrey M. 2002. Econometric Analysis of Heckman, James J. 1979. “Sample Selection Bias as a Cross Section and Panel Data. Cambridge and Specification Error.” Econometrica, 47(1): 153–61. London: MIT Press.