Endogenous Democracy: Causal Evidence from the Adoption of Potatoes in the Old World ∗

Joan Barcel´o † Guillermo Rosas ‡

∗We thank Matthew Gabel, Jacob Montgomery, and Betsy Sinclair for helpful comments on earlier versions of this article. †Ph.D. Candidate. Department of Political Science. Washington University in St. Louis. joan- [email protected].. Campus Box 1063, One Brookings Drive, St. Louis, MO 63130. Corresponding author. ‡Associate Professor. Department of Political Science. Washington University in St. Louis. tmu- [email protected]. Campus Box 1063, One Brookings Drive, St. Louis, MO 63130. Endogenous Democracy: Causal Evidence from the Adoption of Potatoes in the Old World

Abstract Despite a strong cross-country correlation between development and democracy, it is difﬁcult to gauge the impact of economic development on the likelihood of autocracies to transition toward democratic regimes because of endogeneity, especially due to reverse causation and omitted variable bias. Hence, whether development causes democratization remains a contested issue. We exploit exogeneity in the regional variation of potato cultivation along with the timing of introduction of potatoes to the Old World to identify a causal effect of economic development on transition to democracy. Our results suggest the existence of a causal effect of economic development on democratization with a magnitude that is larger than previously quantiﬁed. This constitutes the best possible evidence to date in favor of endogenous democratization.

Keywords: Historical Experiment, Potato, Economic development, Political Institutions, De- mocratization, Regime Type Does economic development cause democratization? The correlation between levels of democracy and development is a consistent empirical regularity in the field of political economy (Lipset 1959, Przeworski, Alvarez, Cheibub and Limongi 2000). Whether development causes democracy, however, remains a contested issue. Some argue that this correlation arises not because development promotes democracy, but because development simply makes democratic regimes less likely to re- vert to authoritarianism (Przeworski and Limongi 1997, Przeworski et al. 2000). Others suggest that development causes democratization (Boix 2011, Boix and Stokes 2003, Lipset 1959). Yet a third group of scholars conjectures that democracy affects development or that both are the product of events in the process of state formation over the past five hundred years (Acemoglu, Johnson, Robinson and Yared 2008, Weingast 1995). Lacking a plausible way to randomize assignment of countries into different levels of development, political economists have searched for clean identification strategies to estimate the causal effect of development on democracy. The literature includes many correlational studies that control for potential confounders to approximate the conditional independence assumption; these studies exploit variation in cross-sectional time-series data along with a number of flexible model specifications and statistical estimators.1 Scholars have also embraced quasi-experimental designs that exploit instrumental variables for economic development, such as past saving rates (Acemoglu et al. 2008), changes in the incomes of trading partners (Acemoglu et al. 2008, Boix 2011), genetic dis- tances among populations and ratios of domestic-to-world income in 1850 (Boix 2011), or variables from the prehistoric past (presence of domesticable big mammals and annual perennial wild grasses) combined with biogeographic characteristics (climate, number of frost days in winter, length of coastline) (Gundlach and Paldam 2009). However, it is not at all obvious that these instruments approximate “as if random” assignment of countries into different levels of development.2 We exploit exogeneity produced by the interaction between cross-regional variation in potato cultivability and longitudinal variation in the introduction of potatoes to the Old World — i.e., all regions outside the Americas — between 1700 and 1750 to identify a causal effect of economic development on democratization. We borrow our instrument for economic development from Nunn and Qian(2011) (hereafter, we refer to them simply as NQ), who have studied this same interaction between potato cultivability and timing of introduction as a driver of population growth and

1See, inter alia, Benhabib, Corvalan and Spiegel(2013), Brunk, Caldeira and Lewis-Beck(1987), Burkhart and Lewis- Beck(1994), Cervellati, Jung, Sunde and Vischer(2014), Epstein, Bates, Goldstone, Kristensen and O’Halloran(2006), Heid, Langer and Larch(2012), Jackman(1973), Londregan and Poole(1996), Lundberg, Huynh and Jacho-Ch avez´ (2016), Moral-Benito and Bartolucci(2012). 2Online Appendix A contains a summary of these instruments, as well as our assessment of their compliance with the exclusion restriction, independence of potential outcomes, monotonicity, and stable unit treatment value assumptions.

1 economic development in the Old World. Crucially, our identification strategy, which is borrowed verbatim from NQ (2011), relies on a differences-in-differences (DD) design with the degree of suitability to potato cultivation as a continuous treatment and flexible time trends. Our results suggest that the effect of economic development on democratization is larger than previously quantified. Because the instrument is plausibly exogenous to the choice of political regimes, our estimate of the “development effect on democracy”, the quantity of interest in our study, provides the best available evidence to date in favor of endogenous democratization.

Potato productivity shocks, population size, and development

Following NQ (2011), we define potato suitability as the natural log of the total amount of a country’s land that is suitable for potato cultivation. With NQ, we argue that the interaction — and only the interaction — between the geographic distribution of potato suitability and the timing of its introduction generates a productivity shock that is exogenous to the adoption of political regimes. Hence, we exploit the interaction between potato suitability and the timing of the introduction of potatoes as an instrument for economic development, and we use this instrument to quantify the effect of an agricultural productivity-induced change in income on a country’s probability of adopting democracy. Our identification strategy relies on a difference-in-difference (DD) design for the instrument; following NQ, this design has the distinct feature that treatment assignment, the degree of suitability to potato cultivation, is continuous, and the post-treatment period is flexibly assigned. In practice, NQ’s coefficient estimates capture “the additional growth in population levels experi- enced by countries that are suitable for potatoes (relative to those that are not) after potatoes were introduced in 1700 (relative to before)” (NQ 2011, 617). After including controls for other sta- ple crops imported from the Americas, along with country and year fixed effects (FE) to capture potential country-specific confounders and potential common temporal shocks to population and propensity toward democratization, it is this “additional growth” — which we refer to as the “potato productivity shock” — that is exogenous to democratization. NQ compare changes in population from pre-adoption to post-adoption periods across countries that vary in terms of their potato suitability. Though the timing of widespread adoption of potato crops in the Old World is not precisely known, NQ narrow this date to a point between 1700 and 1750. Their work inspects potato suitability and timing of introduction — the potato productivity shock — as a cause behind a country’s population growth and, subsequently, a predictor of eco-

2 nomic development. The logic behind our instrument is based on this same argument, as can be seen in the functional form of the first-stage regression below. Still, to consider this interaction a high-quality instrument to identify the effect of development on democracy, we must assess the verisimilitude of a number of assumptions (Sovey and Green 2011). Consider first the path through which a potato productivity shock might affect economic growth. NQ provide ample historical evidence that widespread adoption of potato cultivation led to population growth through increases in living standards, fertility, and life expectancy. Throughout the late eighteenth, nineteenth, and early twentieth centuries, population growth translated into increases in labor, and the rate of growth of labor had a direct positive impact on the speed at which countries produced economic output.3 In addition, NQ show that an increase in agricultural productivity from widespread potato adoption could decrease the need for labor in farming, pushing population excedents out of the countryside and into cities, and unleashing radical change in innovation and wealth accumulation.4 For our purposes, however, it is enough to document that the potato productivity shock jolted the economy through its effect on population growth. We do need to ascertain that this shock was large enough; a small shock would imply a weak instrument that could produce inflated estimates of the causal effect of growth on democratization.5 In any case, because labor was an indispens- able factor of production in both industrializing and agrarian economies, we anticipate the causal effect of this agricultural productivity-induced economic development on democratic adoption to be homogeneous across countries at different levels of development. The potato productivity shock is thus a relevant instrument that identifies a homogenous effect. We must also ascertain whether the instrument is independent of potential outcomes. Among potential confounders, NQ control, as we do here, for a country’s suitability to other crops imported from the Americas that could have produced a similar productivity shock somewhere around the early eighteenth century. Countries with land suitable to potatoes presumably also have land suitable to other high-yield crops. Beyond these obvious confounders, one could be concerned about potential correlation between the potato productivity shock and political development experiences, particularly experiences with colonization, that would affect a country’s propensity to adopt democracy. Two overlapping sets of countries — those in Africa, Asia, and Oceania, and those that became Spanish and French colonies — have significantly lower potato suitability than countries in Europe,

3Economists typically model production as an interaction among labor, land, and capital. Even if these factors exhibit diminishing returns (negative second derivative), their first-derivative contribution to growth is positive. 4It is difficult to decide empirically “whether potatoes only affected steady state levels or whether they also affected steady state growth rates” (NQ 2011, 609), but this is inconsequential for our purposes. 5F -statistics of all first-stage regressions are larger than 10, satisfying Staiger and Stock’s rule-of-thumb test for weak instruments (Staiger and Stock 1997); we rely, however, on more precise tests developed by Stock and Yogo(2005).

3 which could introduce concerns about biased estimates. Recall, however, that our instrument is not potato suitability, but a potato productivity shock that we estimate based on a DD estimator for the first stage regression. Therefore, we use within-country differences in population from pre- to post-adoption periods for causal identification, which removes time-invariant cross-country hetero- geneity and, thus, factors such as geographical location or a history of colonization have no bearing on our estimates of effect of development on democracy. The exclusion restriction assumption is similarly feasible. Other than through its effect on population, it is difficult to imagine alternative paths through which the potato productivity shock could promote democratization. Arguably, potato cultivation could perhaps lead to the entrenchment of a particular social class, say land-owners, but we do not see indications in NQ’s historical review that the introduction of potatoes abruptly changed land-holding patterns in countries that were more suitable to potato cultivation. Alternatively, the sudden availability of land with potentially high nutritional yield may have invited foreign invasion, thus increasing the incidence of war and post- poning democratization. Yet, we think it more likely that war followed nourishment-induced shocks on growth; in other words, we acknowledge that war may be a consequence of economic growth, but we doubt it followed from a potato productivity shock. Consistent with this view of war as post-treatment to productivity shocks, Iyigun, Nunn and Qian(2015) extend NQ (2011) to estimate a negative long-run effect of a permanent rise in agricultural productivity on conflict. Finally, the instrument is unlikely to violate the assumption of monotonicity In a world in which agricultural techniques remain relatively underdeveloped and in which peasants and landed gentry seek to maximize yield or profit, planting potatoes on land not suitable to its cultivation would lead to lost earnings. We would not expect countries with limited potato suitability to spend resources into potato cultivation, therefore “defying” their assignment into the “control arm”. Strictly speaking, our instrumental-variable design estimates a more limited quantity than the “development effect on democracy.” Instead, our design estimates a “local” effect, the average causal effect of development on democracy among “compliers.” In this setup, “compliers” would be countries with high values of potato suitability that produce high volumes of potatoes upon its introduction and countries with low values of potato suitability that produce low volumes. In this setting, we likely have single crossover from countries with high values of potato suitability that do not cultivate potatoes upon their introduction (these “never treats” fail to receive the potato productivity shock even when their potato suitability is high.) Cross-over from “always treats” is unlikely: It would be impossible to produce high volumes of potatoes on unsuitable land.

4 The causal effect of population size on democracy

We estimate the effect of population size — where population size is instrumented by the potato productivity shock — on the level of democracy of a country. As in NQ’s contribution, we consider contemporary states to be the unit of analysis. We observe Polity scores for current countries in 1800, 1850, 1900, and 1950.6 Recall as well that our design includes both country and year FE that eliminate confounding from unidentified country- and time-specific factors. We employ the interaction between potato suitability and flexible time trends as an instrument of a country’s population size (log scale), then use the instrumented values of population size to estimate the effect of an exogenous agricultural productivity-induced change in population on a country’s probability of adopting democracy.7 The first-stage regression is

1900 X Population sizeit = αt (Potato suitabilityi × It) +Xitγ + ψi + δt + εit, t=1000 | {z } potato productivity shock where i and t index country and period, respectively, It is a period indicator (for years 1000 to 1900), X includes geographic controls — average elevation, tropical surface, rugged terrain area, and suitability to “old world” crops — interacted with the full set of year dummies to ensure fea- sibility of the exclusion restriction, and ψ and δ are FE to capture country-speciﬁc traits and any period-speciﬁc common shocks to all countries. Recall that only the interaction of potato suitability and year contains information that can be used as an instrument for population size. The second stage regression is

V Polityi,t+1 = β · Population sizeit + Xitλ + ξi + φt + νit, where ξ and φ are country and period FE and where we use fitted values of a country’s population size at time t based on coefficient estimates from the first stage regression — i.e., instrumented values of population size — to predict Polity scores 50 years later. Table1 presents two-stage least squares estimates of the effect of productivity-induced changes

6Because not all contemporary states existed as sovereign units at the beginning of the modern era, we have developed a number of coding rules to award them a “political regime” value. See the justiﬁcation of coding decisions in Online Appendix B. We proceed in this way to avoid excessive data attrition. One could argue that there is some risk of invalidating the instrumental-variables design if the potato productivity shock affected the chances that a non-state region would turn into a sovereign state. We counter that to the extent exogenous population increases augment the possibility that a region incorporates into a sovereign state, then state formation is post-treatment to population increase, and therefore not a confounder that needs to be blocked. 7In addition to population size, we also consider instrumented values of urban population share in Online Appendix D — the arguments in NQ (2011) suggest that the potato productivity shock can also be used as an instrument for that variable. Our substantive results are virtually identical.

5 Table 1: IV estimation of the effect of population on democracy, 1100–1950.

(1) (2) (3) (4) (5) (6)

V Population size 7.44∗∗∗ 7.55∗∗∗ 4.58∗ 7.44∗∗∗ 7.80∗∗∗ 4.86∗ (4.97) (3.36) (1.58) (4.97) (3.43) (1.64) Baseline Controls Old World crops × Year FE N Y Y N Y Y Elevation × Year FE N N Y N N Y Ruggedness × Year FE N N Y N N Y Tropical Area × Year FE N N Y N N Y

Cutoff treatment period† 1800 1800 1800 1850 1850 1850 Cragg- Donald F-statistic 56.5 24.4 13.4 84.6 36.5 20.0 Stock-Yogo test‡ 10% 10% 15% 10% 10% 10%

Note: ∗∗∗p<0.01; ∗∗p<0.05; ∗p<0.10, one-tailed tests. N = 1552, in 130 countries. Cells report estimated coefficients with t-statistics in parentheses. The outcome variable in the first stage regressions is population size, with the interaction of potato suitability and years as instrument. The outcome variable in the second stage regression is a country’s Polity score. All models include year and country FE. See Online Appendix C for descriptive statistics and Online Appendix D for first stage regressions. †Cutoffs of the treatment periods maximize the strength of the instruments, see Online Appendix F. ‡Confidence level at which we can reject the null hypothesis that the instrument is statistically weak. in population on a country’s Polity score in six model specifications. Across all models, the first stage regressions show that the instrument is strongly associated with population size, with F - statistics that far exceed the rule-of-thumb value of 10. In the second stage regressions, exogenous increases in population size positively affect a country’s Polity score 50 years later. In Models 1 and 4, which only include period and country FE, instrumented values of population size are significantly associated with democratic political institutions at the 99% confidence level. Even after controlling for geographic characteristics, the estimated effects on Polity remain significant at least at the 90% confidence level.8 Based on the estimates of Model 6, an increase over 50 years of about one mil- lion inhabitants in a country with average population size (4,726,039 inhabitants) would increase the expected Polity score by almost a full point, from 4.48 to 5.42.

The causal effect of economic development on democracy

As we argued before, potato suitability can also be considered an instrument for economic development. This follows theoretically from considering that the potato productivity shock to population and urbanization in turn produces an exogenous increase in economic development. NQ argue that a positive productivity shock from the adoption of potatoes can increase per capita income; we refer the interested reader to their paper to see an elaboration of this argument.

8In Column 3, we can only reject at the 95% conﬁdence level the possibility that the worst case relative bias of the 2SLS estimate is 15% or less (rather than 10% or less). See Online Appendix F for further details on the relative bias of the instrument vis-a-vis` the OLS estimate across each cutoff year.

6 According to the endogenous democratization hypothesis, this exogenous development shock should increase a country’s chances of adopting democracy. To test this link, we consider here a second identification strategy based on Boix(2011). In a manner similar to Boix we interact potato suitability with a time trend starting in 1850 to estimate the agricultural productivity-induced effects of economic growth on adoption of democracy. By interacting potato suitability with a time counter we capture the effect of the original agricultural productivity shock of potato adoption on economic development, net of the upward trend in economic development (as Boix(2011) shows, income levels, but not democracy levels, show extremely high serial correlation). The outcome variable is again a country’s Polity score.9 The first-stage regression models per capita gross domestic product (log scale) as a function of the interaction between potato suitability and a time trend, plus country FE to control for country-specific traits and year FE to control for common shocks to all countries. In addition, we include two principal components of the covariance matrix of the four geographic controls.10 In the second-stage regression, we use values of instrumented per capita gross domestic product at 5- and 10-year lags to estimate the effect of economic development on democratization in a specification that includes all first-stage controls. Columns 1 and 3 show the estimated effect of 5- and 10-year lags of economic development, respectively, on Polity scores during the 1850–2000 period. The first stage regressions show a strong association between the instrument and per capita gross domestic product, even when we include country- and year-FE.11 In the second stage regressions, the exogenous increases in economic development exert a positive and substantively large influence on democracy levels. Though Boix(2011) and Acemoglu et al.(2008) disagree about the existence of an effect of development on democratization in the nineteenth century, they agree that the effect is weak to non-existent in the postwar period (1950–2000). Columns 2 and 4 test this proposition with our instrument. Contrary to them, we find that instrumented values of economic development exert a positive influence on democratization across all specifications during the postwar period (1950– 2000), with two caveats: First, the estimate of 5-year lagged economic development with year FE (Column 2) is positive and large but only significant at the 90% confidence level. Second, the

9We use the same historical estimates of per capita gross domestic product as Boix(2011), who relies on Heston, Summers and Aten(2002) for post-World War II data and on Bourguignon and Morrisson(2002) for pre-war estimates. The analysis is based on all available annual data for a minimum of 77 and a maximum of 94 countries from 1850 to 2000; these are the same countries in Boix’s sample, minus New World countries. 10Inclusion of geographic controls interacted with a time trend produces very unstable estimates, a consequence of multicollinearity of these confounders with potato suitability×time trend. To alleviate this problem, we include two principal components. See the justification and the results of our PCA analysis in Online Appendix G. 11The coefficient estimate for the instrument is statistically significant and substantively large across the four models, and the F -statistic in all specifications far exceeds standard thresholds (Stock and Yogo 2005).

7 Table 2: IV estimation of the effect of economic development on democracy, 1850-2000

Second stage outcome: Polity score

(1) (2) (3) (4)

Log GDPpc\t − 5 3.31∗∗∗ 3.55∗ (2.48) (1.61) Log GDPpc\t − 10 0.47∗∗∗ 11.66∗∗∗ (2.78) (2.37) Democracy t − 5 0.63∗∗∗ 0.45∗∗∗ (19.39) (9.30) Democracy t − 10 0.25∗∗∗ −0.15∗ (3.53) (1.58)

First stage outcome: GDP per capita (log)

ln Potato Suitable Area 0.07∗∗∗ 0.14∗∗∗ 0.63∗∗∗ 0.13∗∗∗ × Time Trend (10.20) (6.92) (2.82) (3.96) Democracy t − 5 0.01∗∗∗ 0.01∗∗ (5.90) (1.84) Democracy t − 10 0.01∗∗∗ 0.003 (4.05) (0.67)

Time period 1850–2000 1950–2000 1850–2000 1950–2000 N observations (countries) 1151 (94) 655 (90) 508 (81) 274 (77) CD F -statistic 103 48 44 16 Stock-Yogo Test‡ 10% 10% 10% 15%

Note: ∗∗∗p<0.01; ∗∗p<0.05; ∗p<0.10, one-tailed tests. Cells report estimated coefficients with t-statistics in parentheses. All models include country and year FE, and two principal components to block the confounding effect of geographic variables. ‡Confidence level at which we can reject the null hypothesis that the instrument is statistically weak. estimate of the 10-year lagged effect of instrumented development on democracy is inordinately large (Column 4), which may be a consequence of a weak instrument.12 But for these caveats, our models consistently indicate a statistically significant, positive causal effect of development on democratization for the entire period (1850–2000) and for the postwar period (1950–2000).

Conclusion

Social scientists have long sought to corroborate whether economic wealth engenders political democracy. Because the experimental manipulation of economic development is impossible, scholars have sought to exploit natural experiments to cast light on this question. Unfortunately, evidence collected from natural experiments is only as strong as the verisimilitude of the assumptions behind the use of instrumental variables, and instruments used previously have not always satisﬁed these assumptions. NQ’s discovery that a country’s suitability to the cultivation of potato strongly pre-

12The ﬁrst stage F -statistic is 16; at this value, we can only reject at the 95% conﬁdence level the possibility that the worst-case relative bias of the 2SLS estimate is 15% or less, rather than 10% or less.

8 dicts population size, urban population share, and economic development adds a new instrument to the toolbox of political economists. Employing potato suitability as a potential instrument for economic development allows us to address the endogenous democratization hypothesis with far more conﬁdence than has been previously possible. Our design gauges how agricultural productivity- induced changes in population and economic development affect a country’s democratization, and our results point to a causal effect of economic development on democratization that is larger than previously quantiﬁed. This constitutes the best available evidence to date in favor of endogenous democratization.

References

Acemoglu, Daron, Simon Johnson, James A Robinson and Pierre Yared. 2008. “Income and Democ- racy.” The American Economic Review 98(3):808–842. Benhabib, Jess, Alejandro Corvalan and Mark M Spiegel. 2013. “Income and Democracy: Evidence from Nonlinear Estimations.” Economics Letters 118(3):489–492. Boix, Carles. 2011. “Democracy, Development, and the International System.” American Political Science Review 105(4):809–828. Boix, Carles and Susan C. Stokes. 2003. “Endogenous Democratization.” World Politics 55(4):517– 549. Bourguignon, Franc¸ois and Christian Morrisson. 2002. “Inequality Among World Citizens: 1820- 1992.” American Economic Review 92:727–744. Brunk, Gregory G, Gregory A Caldeira and Michael S Lewis-Beck. 1987. “Capitalism, Socialism, and Democracy: An Empirical Inquiry.” European Journal of Political Research 15(4):459–470. Burkhart, Ross E and Michael S Lewis-Beck. 1994. “Comparative Democracy: The Economic De- velopment Thesis.” American Political Science Review 88(04):903–910. Cervellati, Matteo, Florian Jung, Uwe Sunde and Thomas Vischer. 2014. “Income and Democracy: Comment.” The American Economic Review 104(2):707–719. Epstein, David L, Robert Bates, Jack Goldstone, Ida Kristensen and Sharyn O’Halloran. 2006. “Democratic Transitions.” American Journal of Political Science 50(3):551–569. Gundlach, Erich and Martin Paldam. 2009. “A Farewell to Critical Junctures: Sorting Out Long-Run Causality of Income and Democracy.” European Journal of Political Economy 25(3):340–354. Heid, Benedikt, Julian Langer and Mario Larch. 2012. “Income and Democracy: Evidence from

9 System GMM Estimates.” Economics Letters 116(2):166–169. Heston, Alan, Robert Summers and Bettina Aten. 2002. “Penn World Table Version 6.1.” Center for International Comparisons at the University of Pennsylvania. Iyigun, Murat, Nathan Nunn and Nancy Qian. 2015. “Agricultural Productivity and Conflict: Evi- dence from Potatoes, 1400–1900.” Unpublished manuscript. Jackman, Robert W. 1973. “On the Relation of Economic Development to Democratic Performance.” American Journal of Political Science 17(3):611–621. Lipset, Seymour Martin. 1959. “Some Social Requisites of Democracy: Economic Development and Political Legitimacy.” American Political Science Review 53(01):69–105. Londregan, John B and Keith T Poole. 1996. “Does High Income Promote Democracy?” World Politics 49(01):1–30. Lundberg, Alexander L, Kim P Huynh and David T Jacho-Chavez.´ 2016. “Income and Democracy: A Smooth Varying Coefficient Redux.” Journal of Applied Econometrics . Moral-Benito, Enrique and Cristian Bartolucci. 2012. “Income and Democracy: Revisiting the Evi- dence.” Economics Letters 117(3):844–847. Nunn, Nathan and Nancy Qian. 2011. “The Potato’s Contribution to Population and Urbanization: Evidence from a Historical Experiment.” The Quarterly Journal of Economics 126(2):593–650. Przeworski, Adam and Fernando Limongi. 1997. “Modernization: Theories and Facts.” World Poli- tics 49(02):155–183. Przeworski, Adam, Michael E. Alvarez, Jose Antonio Cheibub and Fernando Limongi. 2000. Democ- racy and Development: Political Institutions and Well-Being in the World, 1950-1990. Vol. 3 Cambridge University Press. Sovey, Allison J. and Donald P. Green. 2011. “Instrumental Variables Estimation in Political Science: A Readers’ Guide.” American Journal of Political Science 55(1):188–200. Staiger, Douglas and James H. Stock. 1997. “Instrumental Variables Regression with Weak Instru- ments.” Econometrica 65(3):557–586. Stock, James H and Motohiro Yogo. 2005. Testing for Weak Instruments in Linear IV Regression. In Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, ed. Jan Fagerberg, David C. Mowery and Richard R. Nelson. New York: Cambridge University Press pp. 80–108. Weingast, Barry R. 1995. “The Economic Role of Political Institutions: Market-Preserving Federal- ism and Economic Development.” Journal of Law, Economics, & Organization pp. 1–31.

10 Endogenous Democracy: Causal Evidence from the Adoption of Potatoes in the Old World Online Appendix (Supplemental Material)

Contents

A Variables previously used as instruments for income2

B Country observations4

C Descriptive statistics8

D First-stage regressions 10

E Estimation of the effect of growth on democracy, 1100–1950. Instrumented value is share of urban population. Second stage regressions of Polityt+50 as outcome. 11

F The relative bias of the instrument across post-treatment periods 12

G Principal component analysis 13

1 A Variables previously used as instruments for income

Acemoglu, Johnson, Robinson and Yared(2008) • Country’s saving rates from t − 5 to t − 1

– The exclusion restriction fails if “saving rates might be correlated with future anticipated regime changes” (Acemoglu et al. 2008, 810). Research suggests that saving rates reﬂect current conditions but, especially, future anticipated conditions including ongoing shifts in a country’s institutional dynamics (Schmidt-Hebbel, Serven and Solimano 1996). – Furthermore, saving rates at time t − 5 are likely to be endogenous to macroeconomic conditions at t − 5, which are also likely to be associated with economic development at t, and with political institutions at t + 5. As Angrist and Krueger point out, “the use of lagged endogenous variables as instruments is problematic if the equation error or the omitted variables are serially correlated” (Angrist and Krueger 2001, 77).

• Country’s predicted income based on contemporary incomes of trading partners

– Consider the possibility that A trades equally with B and C. Based on this design, A’s income is highly correlated with the incomes of B and C, and this is the only channel by which incomes in B and C could possibly influence country A’s level of democracy. Yet, it follows that income in A also influences incomes in B and C, leading to a feedback relation among countries, especially among those with a small circle of trading partners, that may violate SUTVA. – A number of omitted variables — war, decolonization, diffusion effects through global networks — may easily affect incomes in A, B, and C simultaneously, while simultaneously affecting country A’s political regime. – Boix(2011) cannot reject the null hypothesis that this instrument is weak based on a Stock-Yogo test for a fully-specified model with country and year-FE. – Trade affects income, but it also likely transmits ideas, innovation, and cultural predis- positions that may lead to regime change, violating the exclusion restriction.

Boix(2011) • Country’s genetic distance to economic pace-setter interacted with time trend

– Boix(2011) follows Spolaore and Wacziarg(2009) in building a relative genetic distance measure, which is the distance between the genetic imprint of a country’s population relative to the population of the economic pace-setter, which are the United Kingdom in the nineteenth century and the United States in the twentieth century. – But genetic distance may conﬂate important confounders. (Spolaore and Wacziarg 2009, 495-502) suggests controlling for shared geographic conditions with another country, but this control does not appear in Boix(2011).

• Country’s ratio of domestic-to-world income in 1850 interacted with world median income

2 – This instrument exploits two facts: the correlation between income and democracy does not appear until 1850, and a country’s position in income relative to the world in 1850 is a strong predictor of a country’s position in income in the subsequent century. But the validity of the instrument requires a strong assumption that the conditions that favor economic development are not also conditions that favor democratization in the future and, thus, that the lack of a pre-1850 correlation between income and democracy is the result of sequential stages, and not of processes that have a common root but differential timing.

• Country’s ratio of domestic-to-world income in 1850 interacted with time trend

– See above

Gundlach and Paldam(2009) • Country’s estimate of number of domesticable big mammals and number of annual perennial wild grasses in the prehistoric era interacted with biogeographic characteristics (e.g., climate, number of frost days in winter, proportion of coastline)

– Research shows a strong relationship between historical geographic conditions and past income (CITES), making historical geographic conditions an appealing instrument for current economic conditions. – But the exclusion restriction is violated because historical geographic conditions affected the timing of the Neolithic revolution (Ashraf and Michalopoulos 2015), and this historical transition from foraging to farming that has a strong predictive power on the early development of state institutions (Hariri 2012).

3 B Country observations

We now explicate our coding rules. First, there are many contemporaneous states that appeared after the disolution of long-standing imperial units. For example, a number of states in the Balkans and the Middle East peeled away from the Ottoman Empire in the nineteenth-century; by the end of World War I, which sounded the death knell of this and other empires, some other vassal regions became independent entities, though many others became protectorates of France or Great Britain. Countries that are currently independent but formed part of a more extensive empire or multinational state receive the Polity score of the “tutelar” country at the time of observation (i.e., Serbia in 1800 receives Turkey’s 1800 score, as Turkey was the “tutelar” country in the Ottoman Empire, but Serbia in 1950 receives Yugoslavia’s score). For the most part, this decision is relatively unproblematic, as the tutelar nation in the empire commonly obtains a score of −10 corresponding to a close autocracy. Second, we also distinguish among countries that were protectorates or colonies of a European power and those territories that, while independent from colonial domination, cannot be seen as having stable state structures. Such is the case, for example, of the Congo, which was partitioned among a number of tribal, semi-stable political units before the scramble from Africa that are too far from resembling a modern state to warrant inclusion in our sample. Finally, all units before 1800 receive a Polity score of −10. Table B.1 includes further justiﬁcation for our coding decisions. Table B.1: Coding decisions regarding Polity scores

Polity score at Country 1800 1850 1900 1950 Afghanistan Original polity score Original polity score Original polity score Original polity score Albania Turkey polity score Turkey polity score Turkey polity score Russia polity score Algeria Turkey polity score Turkey polity score Turkey polity score France colony Angola Portugal colony Portugal colony Portugal colony Portugal colony Armenia Turkey polity score Turkey polity score Turkey polity score Russia polity score Australia United Kingdom score United Kingdom score United Kingdom score Original polity score Austria Original polity score Original polity score Original polity score Original polity score Azerbaijan Iran polity score Russia polity score Russia polity score Russia polity score Bahrain Al Khalifa royal family Al Khalifa royal family British protectorate British protectorate Bangladesh British colony British colony British colony India polity score Belarus Russia polity score Russia polity score Russia polity score Russia polity score Belgium France polity score Original polity score Original polity score Original polity score Benin Independent kingdom sub- Independent kingdom sub- French colony French colony servient to Portugal servient to Portugal Bhutan Absolute monarchy Absolute monarchy Absolute monarchy Original polity score Bosnia and Herzegovina Turkey polity score Turkey polity score Turkey polity score Original polity score Botswana No state No state English protectorate English protectorate Bulgaria Turkey polity score Turkey polity score Original polity score Original polity score Burkina Faso No state No state French protectorate French colony Burundi No state No state German colony Belgian protectorate Cambodia No state Joint Thailand-Vietnam French protectorate French protectorate control Cameroon No state No state German colony UK-French protectorate Central African Republic No state No state French colony French colony Chad No state No state No state French colony China Original polity score Original polity score Original polity score Original polity score Comoros No state French colony French colony French colony Congo Brazzaville No state No state French colony French colony Congo, Democratic Re- No state No state Belgian colony Belgian colony public Cote d’Ivoire Native kingdoms French colony French colony French colony Croatia Turkey polity score Turkey polity score Turkey polity score Yugoslavia polity score Continued on next page 4 Table B.1 — continued from previous page Country 1800 1850 1900 1950 Cyprus Turkey polity score Turkey polity score British colony British colony Czech Republic Austria polity score Austria polity score Austria polity score Czechoslovakia polity score Denmark Original polity score Original polity score Original polity score Original polity score Djibouti Turkey polity score Turkey polity score French colony French colony Egypt Turkey polity score Turkey polity score English colony Original polity score Equatorial Guinea Spanish colony Spanish colony Spanish colony Spanish colony Eritrea Turkey polity score, Turkey polity score, Italian colony Ethiopia polity score Turkey controlled Turkey controlled seaboard seaboard Estonia Russia polity score Russia polity score Russia polity score Russia polity score Ethiopia No central state No central state Original polity score Original polity score Fiji British colony British colony British colony British colony Finland Sweden score Russia polity score Russia polity score Original polity score France Original polity score Original polity score Original polity score Original polity score Gabon No state No state French colony French colony Gambia No state British colony British colony British colony Georgia Russia polity score Russia polity score Russia polity score Russia polity score Germany Prussia polity score Prussia polity score Germany polity score West Germany polity score Ghana Ashanti kingdom, Euro- Ashanti kingdom, Euro- Ashanti kingdom, Euro- Ashanti kingdom, Euro- pean colonies pean colonies pean colonies pean colonies Greece Turkey polity score Original polity score Original polity score Original polity score Guinea No unified state No unified state French colony French colony Guinea-Bissau Portuguese colony Portuguese colony Portuguese colony Portuguese colony Hungary Austria polity score Austria polity score Original polity score Original polity score India British colony British colony British colony Original polity score Indonesia Dutch colony Dutch colony Dutch colony Original polity score Iran Original polity score Original polity score Original polity score Original polity score Iraq Turkey polity score Turkey polity score Turkey polity score Original polity score Ireland United Kingdom score United Kingdom score United Kingdom score Original polity score Israel Turkey polity score Turkey polity score Turkey polity score Original polity score Italy Average of Austria and Average of Modena, Papal Original polity score Original polity score Spain States, Parma, Sardinia, Two Sicilies, Tuscany Japan Original polity score Original polity score Original polity score Original polity score Jordan Turkey polity score Turkey polity score Turkey polity score Original polity score Kazakhstan Kazakh Khanate Russia polity score Russia polity score USSR polity score Kenya No unified state No unified state British colony British colony Korea, DPR Original polity score Original polity score for Original polity score for Original polity score Korea Korea Korea, Republic Original polity score for Original polity score for Original polity score Original polity score Korea Korea Kuwait Seems to be independent Seems to be independent British protectorate British protectorate of Persia of Persia Kyrgyzstan Uzbekh score Russia polity score Russia polity score USSR polity score Laos Thai-Burma domination Thai domination French colony French colony Latvia Russia polity score Russia polity score Russia polity score USSR polity score Lebanon No unified state Turkey polity score French protectorate Original polity score Lesotho No state No state British protectorate British colony Liberia No state Original polity score Original polity score Original polity score Libya Turkey polity score Turkey polity score Turkey polity score Original polity score (1951) Lithuania Russia/Prussia polity score Russia polity score Russia polity score USSR polity score Macedonia Turkey polity score Turkey polity score Turkey polity score Yugoslavia polity score Madagascar Independent kingdom Independent kingdom French colony French colony Malawi No state No state British protectorate British protectorate Malaysia British colony British colony British colony British colony Continued on next page

5 Table B.1 — continued from previous page Country 1800 1850 1900 1950 Mali No state No state French colony French colony Mauritania No state No state French colony French colony Mauritius French colony British colony British colony British colony Moldova Turkey polity score Russia polity score Russia polity score USSR polity score Mongolia Manchu rule Manchu rule Manchu rule Original polity score Morocco Original polity score Original polity score Original polity score Interruption Mozambique Portuguese colony Portuguese colony Portuguese colony Portuguese colony Myanmar Unified state British colony British colony Original polity score Namibia No state No state German colony South Africa protectorate Nepal Original polity score Original polity score Original polity score Original polity score Netherlands Batavian republic Original polity score Original polity score Original polity score New Zealand No state No state Original polity score Original polity score Niger No unified state Kanem Empire French colony French colony Nigeria No unified state British protectorate British protectorate British colony Norway Denmark-controlled Original polity score Original polity score Original polity score Oman Original polity score Original polity score Original polity score Original polity score Pakistan British colony British colony British colony Original polity score Portugal Original polity score Original polity score Original polity score Original polity score Papua New Guinea No state No state German and British Australian protectorate colonies Philippines Spanish colony Spanish colony Philippine-American war Original polity score Poland Partitioned Partitioned Partitioned Original polity score Qatar Saudi rule Independent rule Turkey polity score British protectorate Romania Turkey polity score Turkey polity score Original polity score Original polity score Russian Federation Original polity score Original polity score Original polity score USSR polity score Rwanda Kingdom of Rwanda Kingdom of Rwanda German colony Belgian colony Saudi Arabia Saudi kingdom Saudi kingdom Turkey polity score Original polity score Senegal No state French colony French colony French colony Serbia and Montenegro Turkey polity score Original polity score Original polity score Yugoslavia polity score Sierra Leone British colony British colony British colony British colony Singapore Unclear British colony British colony British colony Slovakia Austria polity score Austria polity score Austria polity score Czechoslovakia polity score Slovenia Turkey polity score Turkey polity score Turkey polity score Yugoslavia polity score Solomon Islands No state No state British protectorate British protectorate Somalia Sultanate Sultanate Dervish state British protectorate South Africa British colony British colony British colony Original polity score Spain Original polity score Original polity score Original polity score Original polity score Sri Lanka Sri Raj British colony British colony Original polity score Swaziland Semi-independent king- Semi-independent king- Semi-independent king- British colony dom dom dom Sweden Original polity score Original polity score Original polity score Original polity score Switzerland Original polity score Original polity score Original polity score Syria Turkey polity score Turkey polity score Turkey polity score Original polity score Taiwan Qing rule Qing rule Japanese colony Original polity score Tajikistan Khanate Khanate Russia polity score USSR polity score Tanzania Omani Sultanate Omani Sultanate German colony British protectorate Thailand Original polity score Original polity score Original polity score Original polity score Togo No state No state German colony British-French protectorate Tunisia Turkey polity score Turkey polity score French colony French colony Turkey Original polity score Original polity score Original polity score Original polity score United Arab Emirates Independent emirates Independent emirates Independent emirates Independent emirates Uganda No state No state British colony British colony Ukraine Russia polity score Russia polity score Russia polity score USSR polity score United Kingdom Original polity score Original polity score Original polity score Original polity score Uzbekistan Divided in three khanates Russia polity score Russia polity score USSR polity score Continued on next page

6 Table B.1 — continued from previous page Country 1800 1850 1900 1950 Viet Nam Nguyen rule Nguyen rule French colony French colony Yemen Turkey polity score Turkey polity score Turkey polity score British protectorate Yemen North Turkey polity score Turkey polity score Turkey polity score British protectorate Zambia No central state No central state No central state British protectorate Zimbabwe Rozwi Empire Ndebele Kingdom British colony British colony

7 C Descriptive statistics

Table C.1: Descriptive statistics variables in Table 1 and E.1

Statistic N Mean St. Dev. Min Max

Polity Score 1,552 −8.20 4.87 −10 10 Population (log scale) 1,552 13.32 2.31 3.65 19.98 Share of Urban Population 1,552 0.02 0.05 0.00 0.54 Potato Suitable Area (log scale) 1,552 0.31 1.46 0.00 11.89 Old World Crop Area (log scale) 1,552 0.59 2.13 0.00 12.06 Elevation (log scale) 1,552 0.50 1.68 0.00 8.00 Ruggedness (log scale) 1,552 −0.02 0.31 −3.31 1.91 Tropical Area (log scale) 1,552 0.32 1.74 0.00 12.31

8 Table C.2: Descriptive statistics for variables in Table 2

Statistic N Mean St. Dev. Min Max

Polity Score 1,153 −0.65 7.60 −10 10 Income (log scale) 1,153 7.75 0.94 6.10 10.75 Potato Cultivation Area (log scale) 1,153 5.00 3.59 0.00 11.89 PC 1 1,153 0.54 1.58 -2.15 4.37 PC 2 1,153 0.18 1.30 -4.40 3.99

9 D First-stage regressions

Table D.1 reports the ﬁrst stage regressions from Tables 1 and E.1.

Table D.1: First Stage Regressions from Tables 1 and E.1

Dependent variables: Log Population Urban Population Share

ln Potato Suitable Area × 1100 0.013∗ 0.011 0.012 −0.002 −0.001 −0.001 (1.43) (1.05) (0.93) (1.32) (0.83) (0.31) ln Potato Suitable Area × 1200 0.029∗∗∗ 0.024∗∗∗ 0.024∗∗ −0.001 −0.001 −0.001 (3.3) (2.39) (1.83) (0.79) (0.81) (0.58) ln Potato Suitable Area × 1300 0.039∗∗∗ 0.031∗∗∗ 0.030∗∗ −0.0002 −0.001 0.001 (4.38) (3.01) (2.32) (0.15) (0.33) (0.65) ln Potato Suitable Area × 1400 0.019∗∗ 0.004 0.021∗ 0.001 0.0002 0.001 (2.06) (0.39) (1.63) (0.56) (0.10) (0.49) ln Potato Suitable Area × 1500 0.034∗∗∗ 0.014∗ 0.027∗∗ 0.0003 −0.0002 0.001 (3.78) (1.42) (2.09) (0.24) (0.10) (0.38) ln Potato Suitable Area × 1600 0.041∗∗∗ 0.021∗∗ 0.026∗∗ 0.0002 −0.001 −0.00003 (4.51) (2.04) (2.02) (0.17) (0.64) (0.02) ln Potato Suitable Area × 1700 0.043∗∗∗ 0.018∗∗ 0.024∗∗ 0.002∗ 0.002 0.002 (4.75) (1.75) (1.87) (1.45) (1.09) (1.1) ln Potato Suitable Area × 1750 0.055∗∗∗ 0.030∗∗∗ 0.031∗∗∗ 0.001 0.001 0.001 (6.11) (2.97) (2.42) (1.09) (0.67) (0.66) ln Potato Suitable Area × 1800 0.073∗∗∗ 0.048∗∗∗ 0.041∗∗∗ 0.002∗ 0.002 0.002 (8.07) (4.69) (3.16) (1.49) (1.04) (0.87) ln Potato Suitable Area × 1850 0.095∗∗∗ 0.069∗∗∗ 0.060∗∗∗ 0.002∗∗ 0.002∗ 0.003∗ (10.53) (6.72) (4.67) (1.76) (1.39) (1.55) ln Potato Suitable Area × 1900 0.121∗∗∗ 0.092∗∗∗ 0.080∗∗∗ 0.012∗∗∗ 0.012∗∗∗ 0.010∗∗∗ (13.5) (9.05) (6.16) (8.77) (7.85) (5.0)

Year FE Y Y Y Y Y Y Country FE Y Y Y Y Y Y

Baseline Controls (× Year FE) ln Old World Crop Area × Year FE N N Y N N Y ln Elevation × Year FE N N Y N N Y ln Ruggedness × Year FE N N Y N N Y ln Tropical Area × Year FE N N Y N N Y

N observations 1552 1552 1552 1552 1552 1552 N countries 130 130 130 130 130 130

Note: ∗∗∗p<0.01; ∗∗p<0.05; ∗p<0.10, one-tailed tests. Table reports estimated coefﬁcients with t-statistics in parentheses. Cluster-robust standard errors at the level of the country to model for within-country error correlation are computed.

10 E Estimation of the effect of growth on democracy, 1100– 1950. Instrumented value is share of urban population. Second stage regressions of Polity t+50 as outcome.

Table E.1: Estimation of the effect of growth on democracy, 1100–1950

(7) (8) (9) (10) (11) (12)

V Share of urban population 56.9∗∗∗ 35.0∗∗∗ 35.8∗ 74.5∗∗∗ 53.2∗∗∗ 39.4∗∗ (3.68) (2.46) (1.52) (4.79) (3.66) (1.83) Baseline Controls Old World crops × Year FE N Y Y N Y Y Elevation × Year FE N N Y N N Y Ruggedness × Year FE N N Y N N Y Tropical Area × Year FE N N Y N N Y

Cutoff treatment period† 1850 1850 1850 1900 1900 1900 Cragg- Donald F-statistic 30.0 27.2 12.1 41.3 36.3 20.9 Stock-Yogo test‡ 10% 10% 15% 10% 10% 10%

Note: ∗∗∗p<0.01; ∗∗p<0.05; ∗p<0.10, one-tailed tests. N = 1552, in 130 countries. Cells report estimated coefficients with t-statistics in parentheses. The outcome variable in the first stage regressions is share of urban population, with the interaction of potato suitability (log scale) and year FE as instrument. All models include year and country FE. The outcome variable in the second stage regression is a country’s Polity score. Countries in the New World, where potatoes were consumed in pre-Columbian times, are excluded. †Cutoffs of the treatment periods maximize the strength of the instruments (see text). ‡Confidence level at which we can reject the null hypothesis that the instrument is statistically weak.

11 F The relative bias of the instrument across post-treatment periods

It is well known that weak instruments can produce biased estimates. It is common then to estimate F -statistics corresponding to the added explanatory power of instruments in order to reject this possibility. Stock and Yogo(2005) propose a “weak instrument test” based on the Cragg-Donald F -statistic (?). In our application, it is necessary to determine whether the instrument is weak in post-adoption periods. To do so, Figure F.1 reports the relative bias of the instruments vis-a-vis` the 2SLS estimate for each potential cutoff point to divide the pre-treatment and the post-treatment period in the difference-in-difference design. The red lines plot the Stock-Yogo 95% conﬁdence- level boundaries for biases of 10% and 15% in the size of the 2SLS estimate relative to the OLS estimate, whereas the black lines correspond to the Cragg-Donald F -statistics based on different cutpoints. An instrument is “weak” whenever the red dot appears above the corresponding black dot. For example, based on the 1850 cutpoint, we can say with conﬁdence 95% that the worst case relative bias is approximately 10% or less when we instrument population (log scale). In other words, we can reject under these circumstances the null hypothesis of a weak instrument.

Figure F.1: The relative bias of the instrument across post-treatment periods 40 30

S−Y critical values (10 %) 20

S−Y critical values (15 %) Cragg−Donald F−statistic Cragg−Donald

Population models (F−tests) 10

Urbanization models (F−tests) 0

1750 1800 1850 1900

Cutoff points

12 G Principal component analysis

Though our identiﬁcation strategy requires that we control for the interaction between potential confounders and time trends, these quantities are themselves very highly correlated with the interaction between potato suitability and time trends. This is far from surprising: after all, time enters in our model as a linear trend, which produces multicolinearity. Consider the very high pairwise correla- tions that obtain among these variables:

Table G.1: Correlation matrix of geographic variables ln Potato A. ln Old World ln Elevation ln Ruggedness ln Tropical A. × trend Crop A. × trend × trend × trend × trend ln Potato A. × trend 1 ln Old World Crop A. × trend 0.67* 1 ln Elevation × trend 0.49* 0.76* 1 ln Ruggedness × trend 0.08* -0.18* 0.13* 1 ln Tropical A. × trend -0.08* 0.49* 0.35* -0.25* 1 * p < 0.001

We carry out a principal component analysis of four variables, all of which consist of a country- speciﬁc geographic variable (log scale) times a time trend. Table G.2 displays eigenvalues and eigenvectors of the four principal components:

Table G.2: Principal component analysis of geographic controls Comp. 1 Comp. 2 Comp. 3 Comp. 4 ln Old World Crop Area × trend 0.64 0.06 –0.31 0.71 ln Elevation × trend –0.15 0.86 0.42 0.24 ln Ruggedness × trend 0.57 0.41 –0.25 –0.66 ln Tropical Area × trend 0.50 –0.30 0.81 –0.06 Eigenvalue 2.11 1.14 0.57 0.18

The ﬁrst two principal components account for 81% of total variation in the inputs. These principal components are still highly correlated with the instrument (the correlation coefﬁcients are 0.45 and 0.31), but using them in lieu of the four geographic controls eliminates instability in estimates derived from multicollinearity.

13 References

Acemoglu, Daron, Simon Johnson, James A Robinson and Pierre Yared. 2008. “Income and Democ- racy.” The American Economic Review 98(3):808–842. Angrist, Joshua and Alan B Krueger. 2001. Instrumental variables and the search for identification: From supply and demand to natural experiments. Technical report National Bureau of Economic Research. Ashraf, Quamrul and Stelios Michalopoulos. 2015. “Climatic fluctuations and the diffusion of agri- culture.” Review of Economics and Statistics 97(3):589–609. Boix, Carles. 2011. “Democracy, Development, and the International System.” American Political Science Review 105(4):809–828. Gundlach, Erich and Martin Paldam. 2009. “A Farewell to Critical Junctures: Sorting Out Long-Run Causality of Income and Democracy.” European Journal of Political Economy 25(3):340–354. Hariri, Jacob Gerner. 2012. “The autocratic legacy of early statehood.” American Political Science Review 106(03):471–494. Schmidt-Hebbel, Klaus, Luis Serven and Andres Solimano. 1996. “Saving and investment: paradigms, puzzles, policies.” The World Bank Research Observer 11(1):87–117. Spolaore, Enrico and Romain Wacziarg. 2009. “The Diffusion of Development.” The Quarterly journal of economics 124(2):469–529. Stock, James H and Motohiro Yogo. 2005. Testing for Weak Instruments in Linear IV Regression. In Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, ed. Jan Fagerberg, David C. Mowery and Richard R. Nelson. New York: Cambridge University Press pp. 80–108.