2. State instability and economic development

Within the vast literature on state and development, Olson (Reference Olson1982, Reference Olson1986, Reference Olson1993) provides guiding theoretical insights to help explain the rise and decline of economic development due to state duration. Specifically, the idea of roving versus stationary bandits (Olson, Reference Olson1993) describes the incentive structures for state subjects. In the case of roving bandits which occasionally plunder subjects, states are constantly changing, which in turn provide little incentives for the subjects to invest and produce. Short state duration thus corresponds to instability and low levels of economic development.Footnote ²

States of stable order, on the other hand, are achievable when roving bandits turn into stationary bandits and take the form of taxation rather than plunder. The stationary bandits solve the inherent collective action problem in societies looking to transition from anarchies to stable states.Footnote ³ Furthermore, they provide social contracts and protection for their subjects, to the extent that their incomes are maximized by increasing the investment and production by the subjects. But long state duration may also face reduced productivity with the eventual rise of powerful interest groups over time (Olson, Reference Olson1982, Reference Olson1986). These groups benefit from furthering their own agendas and private gains at the expense of outcome for society as whole, and may lead to the process of “institutional sclerosis, slowing its adaptation to changing circumstances and technologies” (Olson, Reference Olson1982).Footnote ⁴

The insights from Olson's works have been supported by others in the literature. For example Cox et al. (Reference Cox, North and Weingast2015) show that in the presence of frequent conflicts, rulers have little incentive to invest in state-building and strong states fail to emerge, while Alesina and Perotti (Reference Alesina and Perotti1996); Barro (Reference Barro1996) make the now well-established claim that political instability is harmful to development. Numerous works also discuss cases in which long and stable states perform poorly as they become resistant to beneficial changes. Lagerlof (Reference Lagerlof2016) presents an institutional mechanism in which autocracies invest in extractive capacity and become more resistant to democracy, while Hariri (Reference Hariri2012) presents similar arguments in the case of former colonies. The longer a state persists in a society, the more a particular group of economic actors become beholden to that institution. As in Olson (Reference Olson1982, Reference Olson1986), such special interest groups contribute to institutional stagnation that could, in turn, result in lowering economic growth over time (North, Reference North1971; Horgos and Zimmermann, Reference Horgos and Zimmermann2009; Coates et al., Reference Coates, Heckelman and Wilson2010, Reference Coates, Heckelman and Wilson2011).

In the case of stagnant states, institutional modifications can impact local economic development through increases in state capacity, wealth and redistribution, even when such changes are violent and destructive (Tilly, Reference Tilly1992; Stasavage and Scheve, Reference Stasavage and Scheve2012; Dincecco and Katz, Reference Dincecco and Katz2016). Given that state overthrows by subjects are difficult due to collective action problems, rulers are often replaced by succession crises or external invasions (Olson, Reference Olson1993). These “creative destructions” may lead to strong states in regions with no prior state foundation, or new ones replacing the old and inefficient, varying by natural surroundings and exogenous shocks (Acemoglu et al., Reference Acemoglu, Johnson and Robinson2005, Reference Acemoglu, Cantoni, Johnson and Robinson2011). However, these findings are not usually based on multitudes of state turnovers. Rather, a period of instability is often followed by establishment of stronger institutions, and the transition is complete after few turnovers.

3. Data sources and description

In constructing our mean duration of state rule variable and other controls, we use two main data sources in this paper: (1) Euratlas, which provides historical maps of Europe, and (2) Eurostat, which provides GDP data for the entire continent.Footnote ⁵ The historical maps from Euratlas contain geographic boundaries of all political entities in Europe for every 100 years starting from the year 1 until 2000 AD. These maps provide us with information on the sovereign entity that ruled a given region in Europe at the turn of the century over the past 2000 years. While the time span that we look at is limited by the absence of similar maps in earlier periods, the existing maps are particularly useful since they allow us to track political changes for a particular region over time. By overlaying these maps on top of each other, we can obtain a list of different sovereign entities that governed a given area.Footnote ⁶ As an example, consider different parts of present-day Germany; many sovereign entities have ruled parts of these areas, and the major ones include Kingdom of Austrasia, Saxony, Bavaria, Kingdom of the East Franks, Kingdom of the Holy Roman Empire, Brandenburg, Bohemia, Prussia, Hanover, the German Empire and the Federal Republic of Germany. Such a listing of sovereign rulers allows us to count the number of centuries for which each region was occupied by a state, but also identify unique sovereign states that occupied these areas.

Figure 1 presents the overlap of the political boundaries in Europe using all of 21 Euratlas maps. Europe provides an apt case study for us, since there were extended periods of fragmentation of former empires but also consolidation of nation states.Footnote ⁷

Figure 1. Overlap of Sovereign Rulers (1–2000 AD). The above figure presents the overlapped boundaries of all sovereign rulers in Europe based on Euratlas for the period 1–2000 AD. The boundary in black refers to present-day (year 2000) state boundaries.

In order to identify what we mean by sovereign states, we first require a working definition. Since the study covers 1 to 2000 AD, our definition of “sovereignty” should be one that is consistent over this time period. Admittedly this task is difficult, because the concept of sovereignty likely has changed and there have been both institutional and jurisdictional fragmentation of states in the European history (Brewer, Reference Brewer1989; Epstein, Reference Epstein2000; Drelichman and Voth, Reference Drelichman and Voth2014). The goal of our paper is not to define a new, encompassing terminology that satisfies all the different definitions of sovereignty over different regions and time periods. That said, it is possible to identify sovereign states in Europe as long as we are consistent in the manner through which we single out such entities. In this paper, we identify a sovereign state as an independent entity that possesses four features as outlined in Euratlas (http://www.euratlas.net/): (1) a territory delimited by borders, (2) a population, (3) an authority exercising the effective public power on population and territory, (4) supremacy, that is with capacity to control absolutely the territory and the population.Footnote ⁸

Both big empires and small city-states (e.g., the Roman Empire and Venice) are classified as sovereign states in Euratlas, since they satisfy all of the said criteria. On the other hand, Euratlas inevitably excludes some entities that could be described as independent polities but did not satisfy one or more of these criteria. For example, it describes some of the excluded entities as autonomous peoples: generally nomadic, semi-nomadic or not well-known populations without evident central authority. This approach necessarily limits the number of entities in consideration, especially in the early periods. Nevertheless, we only focus on the sovereign states as defined above, because our main interest lies on the effect of changes in both defined territorial borders and central authorities that have become fundamental to the modern states today. There are clear advantages when identifying sovereign states in this manner. First, the definition can be consistently applied over the last 20 centuries. Second, this conception of a sovereign state is different and more general from the other similar terms such as “regimes” that are used primarily to distinguish between democracies and autocracies (Przeworski et al., Reference Przeworski, Alvarez, Cheibub and Limongi2000; Jong-A-Pin and Haan, Reference Jong-A-Pin and Haan2011). Our use of the term “sovereign state” and “sovereign entity” (used interchangeably in this paper) is preferred, because we cover a vast time span (1–2000 AD) where typologies such as democracy and autocracy are less relevant, especially in the early time periods. Lastly, it is consistent with our empirical data source, ensuring that our theoretical conception of a “sovereign state” matches the empirical strategy in this paper.Footnote ⁹

Given this definition of a sovereign state, a change in sovereignty of a particular region refers primarily to a change in the authority that governs the region and its population (Nussli, Reference Nussli2011). Typically, dynastic changes or ruler turnovers are not considered as state turnovers in the data; we therefore focus predominantly on conquests that involve domination by foreign entities. According to this approach, mergers and turnovers following implosion of a state would also not be considered state turnovers, so long as the region retains a common entity as listed in Nussli (Reference Nussli2011)'s references.Footnote ¹⁰

Combining state presence and turnovers, we construct our measure of mean duration of state rule as follows:

(1)$${\rm MeanStateDuration}_{i} = {{1} \over {\#{\rm States}_{i}}} {{\#{\rm Centuries\, of\, Rule}_{i}} \over {20 {\rm Centuries}}} $$

where subscript i refers to grid-cell i, #States refers to the number of unique sovereign states that ruled over the grid-cell from 1 to 2000 AD, and #Centuries of Rule refers to the number of centuries (out of a total of 20) for which the grid-cell was under state rule. The above formula gives a composite measure of mean duration of state rule, capturing both the total number of states that ruled over the region, as well as the number of centuries of state presence within the 21 century time frame in our data. One could interpret the measure as a proxy for state stability, similar to Blaydes and Chaney (Reference Blaydes and Chaney2013))'s ruler duration.Footnote ¹¹ The difference here is that our measure captures the level of instability through state turnovers rather than ruler changes. The measure also incorporates the fraction of time for which region i witnessed the presence of a sovereign state. The importance of state presence and its antiquity are discussed at length in both Bockstette et al. (Reference Bockstette, Chanda and Putterman2002) and Borcan et al. (Reference Borcan, Olsson and Putterman2018)'s work, which also present findings that are in line with our empirical findings below. Our approach to understanding the impact of state history differs from these authors, in that we use a fixed geographic unit of observation (grid-cell) to count the number of unique sovereign states, and calculate the mean duration of state based on this number. Unlike the existing works, we focus on the outcome of state changes at the local level instead of the state level in the current period. The geographical boundaries of these states have changed over time and created regions with different state histories, and are now contained within the same current state.

Our approach follows existing works that have similarly used either century or half-century intervals to track political situations (Bockstette et al., Reference Bockstette, Chanda and Putterman2002; Stasavage, Reference Stasavage2014; Borcan et al., Reference Borcan, Olsson and Putterman2018). There is arguably no objective and measurable criterion to define the relative importance of specific years, or correct time intervals for analysis. Euratlas captures state presence at fixed moments in history; ideally we would obtain a map of Europe for every year since there could be cases where state changes took place within a given century. In presenting only the states at the beginning of each century, Euratlas maps potentially leave important information out, but do show the result of every time sequence within the 2000 years. We argue that while events in between centuries are not captured in the Euratlas data, it is also unlikely that state changes took place systematically by the imposed century-intervals, in anticipation of the beginning of a new century. That is, we find no reason to believe that our observations in the beginning of each century would lead to a bias with an outcome different from those observed over some other time intervals. On the other hand, there still remain potential issues with measuring the mean state duration, as we can only track state presence at the beginning of each century. As explained in the following sections, we attempt to address potential systematic measurement issues in certain regions by introducing a set of controls, including standard geographic covariates as well as regional and development indicators.

The other data that we use in this paper are Gross Domestic Product (GDP) and the population in the year 2010 from the NUTS-3 classification, the most disaggregated administrative level in Eurostat.Footnote ¹² The data construction involves merging of the Euratlas and Eurostat data. In order for us to obtain comparable units, we again use grid-cells to cover the continent. This approach follows existing works in the literature that make use of grid-cells as units of analysis, and potentially run into methodological tradeoffs (Alesina et al., Reference Alesina, Easterly and Matuszeski2011). In our case, we need to determine how to interpret several intersections of land within a grid-cell, in which each different slice of land has a different history of state changes. We also note, however, that such decisions are even more problematic for larger geographic units of analysis, and that the smaller grid-cells instead have the attractive feature of allowing only some of the border changes at a time. Given the various state changes over different centuries, it is otherwise difficult to assess changes using, for example, the entire boundary.Footnote ¹³ Moreover, this approach allows for changes in a sovereign state's territorial size over time. For example, France has been a sovereign state for many centuries but its size has expanded and contracted over time (e.g., expansion during the Napoleon period and contraction thereafter). By using the grid-cell approach, we are able to account for such territorial changes at the borders of the main sovereign state. Similar works have looked at historical conflicts and city locations in Europe (Dincecco and Onorato, Reference Dincecco and Onorato2016) and more recent waves of conflict and global income levels (Tollefsen et al., Reference Tollefsen, Strand and Buhaug2012).

Our grid comprises 10,000 cells, each approximately 77 km × 62 km (see Figure 2). Out of the 10,000 cells, we restrict our analysis to about 2,400 grid-cells that contain some land mass, and to sovereign entities that govern at least 0.1 km of the grid-cell area.Footnote ¹⁴ This ensures that grid-cells only containing water and sovereign rulers that only rule a very small section of the grid-cell are excluded in the analysis. We then identify the sovereign entity that governed a particular grid-cell at the turn of each century. If a grid-cell is part land and part water (e.g., coastal grid-cells), we identify the sovereign entity that govern the land mass in that grid-cell. In addition, if there is more than one sovereign entity that rules a grid-cell in a given year, we use the sovereign entity that governs the maximum area of the grid-cell.

Figure 2. Overlap of Sovereign Entities (1–2000 AD). The above figure presents the overlapped boundaries of all sovereign rulers in Europe based on Euratlas for the period 1–2000 AD along with the grid that comprises 10,000 grid-cells. The boundary in black refers to present-day (year 2000) state boundaries.

Next, we overlap the grid with NUTS-3 Eurostat maps and calculate the year 2010 GDP per capita for each grid-cell. As before, we restrict the analysis to grid-cells that have some land mass, and to NUTS-3 areas that occupy at least 0.1 km of the grid-cell area. If a particular grid-cell is part of two or more NUTS-3 regions, we use the GDP per capita data based on the NUTS-3 region that occupies the maximum area of the grid-cell. These steps ensure that the process of calculating the regional income data is consistent with that of the sovereign rulers (we also calculate the “average” and “area-weighted” income data for each grid-cell for robustness checks in the Appendix). Figures 3 and 4 show the distribution of the mean state duration and year 2010 GDP per capita for each grid-cell, respectively.

Figure 3. Mean State Duration. The above figure presents the mean state duration based on Euratlas from the year 1 to 2000 AD along with the grid that comprises 10,000 grid-cells. The data are restricted to only those grid-cells that have some land mass. The boundary shown refers to the present-day (year 2000) state boundaries.

Figure 4. GDP per Capita Distribution. The above figure presents the distribution of GDP per capita based on Euratlas in the year 2000 along with the grid that comprises 10,000 grid-cells. The data are restricted to only those grid-cells have some land mass. The boundary shown refers to the present-day (year 2000) state boundaries.

In sum, the unit of analysis we use in this paper is the grid-cell. This is a preferable option compared to that of present-day states because the state boundaries have changed over time and there are only a limited number of countries in Europe. The use of the grid-cell as the unit of analysis also ensures that the size of the different units are the same over time and makes it possible to conduct a large-N analysis. The main outcome variable is the logged GDP per capita in the year 2010 for each grid-cell (calculated in Purchasing Power Standards thousands of Euros per capita),Footnote ¹⁵ and the main predictor variable is the mean state duration associated with the grid-cell over the period from 1 to 2000 AD.

For each grid-cell, we calculate the number of centuries during which there was a sovereign state in power, and the total number of unique sovereign entities that governed the same area. We then construct the mean duration of state rule as laid out in Equation 1 above. In addition to the state duration measure, we consider four different alternative explanations of economic development. First, the geographic environment of a region influences its economic development (Sachs, Reference Sachs2001). For instance, agriculturally productive areas are associated with higher levels of production and income (Johnston and Mellor, Reference Johnston and Mellor1961). In particular, Diamond (Reference Diamond1999) and Ashraf and Michalopoulos (Reference Ashraf and Michalopoulos2015) argue that the geographical location of Europe made it more suitable for agriculture and domestication of animals that eventually led to the Neolithic Revolution. Especially in periods prior to industrialization, the share of agriculture of the GDP was very high, making agricultural suitability an important determinant of economic development and a coveted resource for sovereign states. We obtain agricultural suitability data from Ramankutty et al. (Reference Ramankutty, Evan, Monfreda and Foley2008) and calculate the suitability index for each grid-cell measured as the fraction of land suitable for agriculture. Agricultural suitability is also related to the elevation of a particular region, since high-altitude areas are more conducive to growing certain types of crops. We obtain the mean elevation data from ESRI 2008 GIS Maps.

Second, the European economic growth is associated with the maritime trade, especially in the post-1500 period (Acemoglu et al., Reference Acemoglu, Johnson and Robinson2005). Trade between European countries and their colonies increased the power of traders and merchants who demanded better property rights. In turn, these institutions are associated with higher levels of economic growth in those regions. In addition to arable land, regions with access to water passages would have added more value to sovereign states and potentially witnessed more contestations over them. We control for this mechanism using the distance from the center of a grid-cell to the nearest body of water.

Third, if the frequency of state changes is related to current economic outcomes, then the existence of states before 1 AD may also have lasting influence. While we necessarily limit our calculation of state changes back to 1 AD, one way to control for additional historical state presence is by including the timing of agricultural adoption (Borcan et al., Reference Borcan, Olsson and Putterman2018)Footnote ¹⁶. That is, we use the number of years from the adoption date up to the year 2000 (Pinhasi et al., Reference Pinhasi, Fort and Ammerman2005; Olsson and Paik, Reference Olsson and Paik2013), as a proxy for the pre-1 AD state influence.Footnote ¹⁷

Fourth, the growth of cities is correlated with places moving from an agriculture-based economy to more industrialized forms of production. Cities bring together manufacturing and service arms of different sectors, making the transport of goods, dissemination of information and procurement of labor more efficient (Henderson, Reference Henderson1988; Glaeser et al., Reference Glaeser, Kallal, Scheinkman and Shleifer1992; Stasavage, Reference Stasavage2014). Moreover, high population densities in cities are seen as essential to increases in labor productivity and economic growth of that region (Ciccone and Hall, Reference Ciccone and Hall1996). These urban centers are also often targets for invasion due to their wealth and strategic locations, which may lead to more state changes. Hence we control for the distance to the nearest city to account for the impact of urbanization on GDP output.Footnote ¹⁸

Finally, we include the geographic coordinates, as well as the year 2000 country indicators, further demanding tests that help control for location and current border shocks. Certain locations, even controlling for geographical endowments and historical development, may witness frequent turnovers simply due to their proximity to the core of Europe, while the peripheries experience less (Merriman, Reference Merriman2003). In addition, we expect that local income levels are likely driven by the state-level unobservables that vary across the modern state boundaries.

Table 1 presents the summary statistics of all the variables used in our analysis. There is considerable variation in the number of unique sovereign states that ruled a given grid-cell. Less than 1 percent of the grid-cells, all of which lie in the peripheral stretches of land by the sea or remote islands, are coded as having been ruled by just one sovereign state, and only one grid-cell was governed by the maximum number of 12 different sovereign states (in Sardinia, Italy). On average, a grid-cell was ruled by five unique sovereign entities over the 21 centuries. As described above, the mean state duration is the fraction of time for which a region was ruled by a sovereign state. This variable ranges between 0.06 and 0.75 with a mean of 0.19, and from Equation 1 this translates to the mean ratio of centuries of rule over the number of unique states being 3.8. Grid cells with mean levels of state rule are located in present-day UK, Ireland and Germany. As shown in the maps earlier, there are also expected variations in the GDP per capita across the continent. On average, the GDP per capita of a grid-cell was about 19,000 PPS euros per thousand people with a fairly high standard deviation of 10,600. The distribution is similar when we use other variations of the logged GDP per capita measure.Footnote ¹⁹

Table 1. Summary statistics.

GDP per capita variables are measured in PPS Euros/1000 ppl.

4. Mean state duration on development

How is the mean state duration related to the current local income level? On the one hand, Borcan et al. (Reference Borcan, Olsson and Putterman2018) show that both short and long state antiquities lead to low productivity relative to those in the middle range. The former inhibit productivity because of their inability to increase fiscal capacity, while the latter do so as centralized powers become extractive and entrenched in power struggles. But the relationship between the mean duration of state and development may also be non-linear and inverse-U shaped, as the general findings discussed in the literature suggest that turnovers may be associated with state performance. As in Equation 1, the mean duration of state is calculated not only by the length of state presence in a region, but also the number of state turnovers. Holding other things equal, more turnovers imply shorter mean durations, a critical distinction that separates a stable state presence from an unstable one. This means that both regions where states had short mean durations and those where they had long durations suffer from lower development. The former has either a short state history, frequent turnovers, or both, while the latter has a long state history, low turnovers, or both.

To see whether the mean duration of state rule is related to the local income level as predicted, we use the following estimation equation for our analyses:

$$ \eqalign{ Income_{i} = & \beta_{0} + \beta_{1} \ast [{\rm Mean\, State \, Duration}]_{i} + \beta_{2} \ast [{\rm Mean\, State\, Duration}]^{2}_{i} + \cr & {\bf\,{X}_{i}}\beta_{3} + \gamma_{\,j} + \epsilon_{ij}} $$

where Income _i is the logged GDP per capita in the year 2010 in grid-cell i, Mean State Duration_i is the mean duration of state rule, and Mean State Duration² is its square. This specifically allows us to examine a non-linear relationship between the mean duration of state and GDP per capita. X_i contains the list of control variables for grid-cell i including the geographic and urbanization controls. As discussed earlier, geographic controls include the agricultural adoption date, agricultural suitability, distance to water, elevation, and a polynomial of latitude and longitude for each grid-cell. We use the distance to the nearest city as our measure for urbanization. Lastly, we also include country fixed effects in the year 2000 (γ_j) to account for any unobservables that vary across current state borders. The coefficients of interest are β₁ and β₂, both related to the mean state duration variable.

5. Main findings

We present our main findings in Table 2. Column one is the baseline model where the logged GDP per capita in the year 2010 is regressed on the mean duration of state rule between 1 and 2000 AD and its squared term. The mean duration of state rule has a positive coefficient estimate whereas the squared term has a negative coefficient estimate, and both estimates are statistically significant at the one percent level. This suggests that there is a non-linear relationship between mean duration of state rule and GDP per capita. The signs on the two coefficient estimates suggest an inverse-U relationship: the mean duration of state rule is associated with higher income levels but this relationship tapers off and ultimately becomes negative as state rule continues for more centuries.

Table 2. Mean state duration on logged 2010 GDPPC.

The outcome variable in all of the above models is the logged GDP Per Capita in the year 2010 (in PPS thousands EUR per thousand people). All models include country fixed effects in the year 2000, and robust standard errors are shown in parenthesis. *p<0.10, **p<0.05, ***p<0.01.

Columns 2–6 progressively introduce different control variables. In all models, the positive and statistically significant coefficient estimate of the mean duration of state rule and the negative and statistically significant coefficient estimate of its squared term are present at the one percent significance level. All seven models include country fixed effects in the year 2000 to account for possible unobservables in current state boundaries that could affect the GDP per capita of a region. The sign and statistical significance of the mean state duration variable and its squared term remain consistent across all models, providing us with robust evidence on the inverse-U shaped relationship.

Figure 5 presents the inverse-U relationship between the mean state duration and the GDP per capita based on the full model in Table 2. It plots the predicted values of GDP per capita (along with the standard errors) as the mean state duration varies between values of 0 and 1 . The figure shows that the logged 2010 GDP per capita increases with the mean state duration but does so only up to point. As the mean state duration goes beyond 0.3, its association with the logged GDP begins to decrease. To substantively make sense of these findings, we present the marginal association of mean state duration on the change in the 2010 logged GDP per capita levels in Figure 6, holding all other controls at their mean levels. The figure shows that the marginal effect of states has a positive association on income initially but that this effect is decreasing and ultimately turns negative.

Figure 5. Quadratic Relationship between Mean State Duration and logged GDP per capita.

Figure 6. Linear Marginal Effect of Mean State Duration on logged GDP per capita.

In the Appendix, we provide further support for our results through additional robustness tests. First, we include state presence as a separate control from the mean state duration variable. We also include indicators for specific empires, whose rule determined long-term economic growth (ex. Roman Empire), as well as certain regions that experienced different levels of state capacity (ex. the European “city belt”). Next, we address issues on institutions and violence; some European cities developed parliaments while others were plundered during the last 2000 years. Finally, we explore alternative measures of state turnovers and development, as well as potential spatial correlations between different grid cells. Our results remain robust under these additional checks.