1. Leveraging exogeneity: Nunn & Qian's potato productivity shock

We start by relating how Reference Nunn and QianNunn and Qian (NQ) conceive of a potato productivity shock, the variable that we employ as an instrument of economic growth. As NQ tell it, the widespread adoption of potatoes in the diet of populations outside the Americas caused a qualitative demographic and economic “jump” during the eighteenth and nineteenth centuries. Areas of the Old World that were particularly well-suited to the cultivation of potatoes benefited more upon introduction of this tuber. Because the suitability of land in the Old World to potato cultivation was essentially unaffected by economic and political development, we can see this “jump” as exogenously assigned under some plausible assumptions.

In fact, NQ insist that only the interaction between potato suitability Footnote ³ and its timing of introduction to the Old World generates a productivity shock that can be interpreted as exogenous to population size and urbanization. To build the potato productivity shock variable, NQ interact suitability to potato cultivation, which is a continuous variable, with dummy indicators for a number of time periods before and after the presumptive widespread adoption of potatoes in the Old World, which occurred at some point between the late seventeenth century and the early eighteenth century.Footnote ⁴ When used as a predictor of urban population shares in a regression model, the coefficient estimates for these interactions capture the additional growth in urbanization in countries with high potato suitability (relative to countries with low potato suitability) in post-adoption periods (relative to pre-adoption periods). NQ's claim — and ours by extension — is that this “additional growth” in urbanization is exogenously assigned; our claim is that we can use the potato productivity shock as an instrument for urbanization, and therefore exploit this as an instrument to estimate the effect of economic growth on democracy.Footnote ⁵

Note that capturing the interaction between timing of introduction and suitability is similar to employing a difference-in-differences (DD) strategy that controls for both time-invariant differences across countries and secular changes over time. As in any DD design, inferential threats remain if other productivity shocks occur around the time of introduction of potatoes and if these shocks are correlated with potato suitability. To assuage this threat, NQ include controls for interactions between time periods and suitability of land to other staple crops imported from the Americas (to control for potential simultaneous shocks), along with country and year fixed effects (FE) to capture potential country-specific confounders and potential common temporal shocks to urbanization.

Equation (1), which is the exact same model that NQ estimate (see Nunn and Qian, Reference Nunn and Qian2011, Equation 4, p. 618), corresponds to the first-stage regression in our instrumental-variable model:

(1)$$\eqalign{ {\rm \it{Y}}_{it}& =\sum_{\,j=1000}^{1900} \beta_j \underbrace{{\rm Potato suitability}_i{\cdot} {\rm \it{I}}_t^j}_{{\rm \text{Potato productivity shock}}} + \overbrace{\sum_{\,j=1000}^{1900}{\boldsymbol {X}}'_i{\boldsymbol {I}}_t^j\phi_j}^{{\rm \text{Simultaneous shocks}}} \cr & \quad + \underbrace{\sum_c \gamma_c {\rm \it{I}}_i^c}_{{\rm \text{Country FE}}}+\overbrace{\sum_{\,j=1000}^{1900}\rho_j {\rm \it{I}}^j_t}^{{\rm \text{Year FE}}} + \varepsilon_{it} }$$

In Equation (1), outcome Y corresponds to urbanization rates indexed by country i and period t. I_t is a period indicator (for periods 1000 to 1900, see fn. 6), I_i is a country indicator, and X includes geographic controls — average elevation, tropical surface, rugged terrain area, and suitability to “old world” crops — interacted with the full set of period indicator dummies to ensure feasibility of the exclusion restriction. The model includes fixed effects that capture country-specific traits and period-specific shocks common to all countries. Consequently, only the interaction of potato suitability and year — the potato productivity shock — contains information that could be interpreted as exogenously assigned.

In order to use the potato productivity shock as an instrument for urbanization, we must further assess the verisimilitude of a number of additional assumptions (Sovey and Green, Reference Sovey and Green2011). To do so, we illustrate in Figure 1 the implied causal chain behind our identification strategy. The outcome of interest is democracy, and the quantity of interest is the effect of urbanization on democracy. We follow ample precedent in thinking of urbanization as a proxy for economic development (see, inter alia, De Long and Shleifer, Reference De Long and Shleifer2011; Acemoglu et al., Reference Acemoglu, Johnson and Robinson2002).

Figure 1. Representation of identification mechanism: arrows identify causal pathways, a dashed arrow identifies a potential violation of the exclusion restriction.

Consider first the theoretical path through which a potato productivity shock might affect urbanization, which NQ lay out with great care. Indeed, the first three stages in Figure 1 (instrumental variable, intervening mechanisms, and instrumented variable) correspond to NQ's argument. In their view, the potato productivity shock had effects on population and urbanization levels. NQ provide ample historical evidence that widespread adoption of potato cultivation led to population growth through increases in living standards, health, fertility, and life expectancy starting in the mid-1700s. Population growth translated into increases in labor, and increased labor had a direct positive impact on the speed at which countries produced economic output. In addition, NQ suggest that an increase in agricultural productivity from widespread potato adoption decreased the need for labor in farming, pushing surplus population out of the countryside and into cities, which explains why population size has a direct effect on urbanization in Figure 1. Finally, NQ suggest that the potato productivity shock directly increased urbanization levels through increases in per capita income, a causal path captured in Figure 1 by the solid curved arrow going from the instrumental to the instrumented variable. NQ then estimate a positive effect of the potato productivity shock on urbanization starting in 1800. They ascertain, and we corroborate, 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.Footnote ⁶ The potato productivity shock is thus a relevant instrument.

2. Exogenous urbanization and democracy

In addition to ensuring that the potato productivity shock is a relevant instrument of urbanization, we must also assess whether the instrument is independent of potential outcomes. Beyond productivity shocks produced by other crops imported from the Americas, one could be concerned about potential correlation between the potato productivity shock and simultaneous political development experiences, which would also violate the exclusion restriction. This potential violation is captured by the dashed curved arrow going from instrumental variable to the outcome in Figure 1. One such experience could be colonization, a development that presumably affected 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, which could introduce concerns about biased estimates of the causal effect of urbanization produced by the fact that countries not suitable to potato crops became colonies that, say, saw potential paths toward democracy blocked by their colonizers. Recall, however, that our instrument is not potato suitability alone, but a potato productivity shock that we estimate based on a DD estimator for the first stage regression. That is, we use within-country differences in urbanization from pre- to post-adoption periods for causal identification, which removes time-invariant cross-country heterogeneity. Thus, factors such as geographic location or a history of colonization pre-dating or post-dating widespread adoption of potatoes have no bearing on our estimate of the effect of urbanization on democracy.

Other than through its effect on urbanization, it is difficult to imagine alternative paths through which the potato productivity shock could promote democracy, which suggests that violations of the exclusion restriction assumption are unlikely. Arguably, a sudden increase in potato cultivation could lead to the rapid entrenchment of a particular social class, say land-owners, that would become inimical to the adoption of democracy; however, 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 nutritious tubers may have invited foreign invasion, thus increasing the incidence of war and postponing democratization. Yet, we think it more likely that war followed nourishment-induced shocks on urbanization. In other words, we acknowledge that war may be a consequence of urbanization, but we doubt it followed directly from a potato productivity shock. Mechanisms such as war may appear in the causal pathway tying urbanization to democracy (i.e., they are post-treatment), but not necessarily in the alternative causal pathways connecting potato productivity shocks to democracy.Footnote ⁷ Be this as it may, we report below on sensitivity tests of potential violations of the exclusion restriction.

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, 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 “urbanization effect on democracy.” Instead, our design estimates a “local” effect: the average causal effect of urbanization on democracy among “compliers.” In this setup, “compliers” include both countries with high values of potato suitability that produce high volumes of potatoes upon their 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-takers” would have failed to receive the potato productivity shock though they had high potato suitability). Cross-over from “always-takers” is unlikely: it would be impossible to produce high volumes of potatoes on unsuitable land. The DD design that we exploit thus produces an estimate of the average causal effect for compliers (Dunning, Reference Dunning2012, 141). In our view, the potato productivity shock is an instrument that identifies a homogeneous effect. In other words, we have no reason to expect the effect to be larger in countries that were relatively more advanced.

We estimate the effect of urbanization — where urbanization 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; all Old World areas before 1800 receive a Polity score of − 10.Footnote ⁸ Because 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; justification of all our coding decisions appears in Online Appendix B. The second-stage regression model is:

(2)$${\rm Polity}_{i, {t+1}} = \alpha {\vskip -1.9pt \hskip 2.3pc \hscale 750% \widehat \vskip 1.9pt \hskip -2.3pc} {{\rm Urbanization}}_{it} + \sum_{\,j=1000}^{1900}{\bf \it{X}}'_i{\bf \it{I}}_t^j\lambda_j + \sum_c \xi_c {\rm \it{I}}_i^c + \sum_{\,j=1000}^{1900}\phi_j {\rm \it{I}}^j_t + \nu_{it}$$

In Equation (2), we use fitted values of a country's urban population share at time t based on coefficient estimates from the first-stage regression — i.e., ${\vskip -1.9pt \hskip 2.3pc \hscale 750% \widehat \vskip 1.9pt \hskip -2.3pc} {{\rm Urbanization}}_{it}$ are the instrumented values of urban population share — to predict Polity scores 50 years later. The second-stage regression includes the same controls that we had in the first-stage model, including fixed effects that preempt confounding from unidentified country- and time-specific factors. The quantity of interest is parameter α, which captures the causal effect of urbanization on democratization.

3. The causal effect of development on democracy

Table 1 presents 2sls estimates of the effect of productivity-induced changes in urbanization on a country's Polity score in three model specifications (for the sake of space, estimates of the baseline control parameters in both the first and second stages are relegated to Online Appendix F). As expected, and consistent with the parallel trends assumption, the first stage regressions show null effects for all periods prior to the widespread adoption of potatoes at some point in the eighteenth century, and then bigger and growing effects for subsequent periods, especially after 1850, across all models — this can be seen in estimates of β_j that are statistically different from 0 in 1850 (or even in 1800 in a couple of models).Footnote ⁹ We also show that the instrument is strongly associated with urbanization, with F-statistics that far exceed the rule-of-thumb value of 10.

Table 1. IV estimation of the effect of urbanization on democracy, 1100–1950.

Notes: ***p < 0.01; **p < 0.05; *p < 0.10, one-tailed. Cells report estimated coefficients with t-statistics in parentheses. Standard errors used for the computation of the t-test are clustered at the country level. See Online Appendix C for descriptive statistics. $^\dagger$F-statistic is based on variation in the post-1750 treatment period.

In the second stage regressions, exogenous increases in urbanization positively affect a country's Polity score 50 years later. In Model 1, which only includes period and country FE, instrumented values of urbanization are significantly associated with democratic political institutions at the 99% confidence level. Even after controlling for potential simultaneous shocks in Model 2 and, additionally, for geographic characteristics in Model 3, the estimated effects on Polity remain significant at least at the 95% confidence level.

The estimates in the first row of Table 1 show the expected change in Polity due to a change in the share of urban population from a country with no cities (a share of urban population of 0) to a country with everyone living in cities (a share of urban population of 1). Based on estimates from Model 3, this means in substantive terms that an increase over 50 years of the share of the urban population from the sample average — 1.9% of the total population — to one standard deviation above the average — i.e., 6.8% of the total population — would increase the expected Polity score by almost two full points, from 2.11 to 4.02.

Online Appendix D demonstrates that this estimated effect is robust to excluding all observations that, by design, have no between- and within-unit variation in our outcome of interest (i.e., all observations prior to 1800). Moreover, formal sensitivity analyses presented in Online Appendix E show that the effect of development on democracy is robust even if the exclusion restriction is violated to a substantial extent (Conley et al., Reference Conley, Hansen and Rossi2012). The estimated effect would still be significant at least at the 90% significance level even if up to 60% of the reduced-form effect of the potato productivity shock on democracy occurred through some unforeseen channel different than urbanization. In short, a potential violation of the exclusion restriction assumption would need to be extremely severe for our estimate to become indistinguishable from zero.