OLD BORDERS AND TERRITORIAL CLAIMS

The vast majority of research on territory and conflict examines either how territorial disputes affect militarized conflict (Hensel Reference Hensel and Vasquez2000; Holsti Reference Holsti1991; Kocs Reference Kocs1995; Luard Reference Luard1986; Vasquez Reference Vasquez1993; Reference Vasquez1995), or how disputants manage and potentially resolve their claims (Hensel Reference Hensel2001; Huth and Allee Reference Huth and Allee2002; Huth, Croco, and Appel Reference Huth, Croco and Appel2011; James, Park, and Choi Reference James, Park and Choi2006; Mitchell Reference Mitchell2002; Mitchell and Hensel Reference Mitchell and Hensel2007). Despite this literature’s clear demonstration of the centrality of territorial disputes to patterns of conflict and cooperation, we have comparatively little evidence about the origins of territorial claims.

As making claims carries significant costs, leaders are quite selective in staking claim to a neighboring state’s territory. Accordingly, the relatively small set of claims made by leaders tend to be the most credible and attractive territorial claims out of the infinite set of potential claims. We argue that the set of potential claims that leaders find attractive and plausible are bounded by two key factors: the opportunity and incentive to make a claim.Footnote ² First, leaders are attentive to how other international actors and their own domestic audience will react to a claim. Claims that are not seen as credible or justified by the leaders of third states or that do not resonate with a leaders’ domestic constituency carry higher costs and are much less attractive. Second, leaders are also attentive to how attractive a potential claim is to prospectively integrate into their territory. For instance, what are the economic, political, or military implications of redrawing a state’s boundary around a given territorial claim?

RESEARCH DESIGN

Assessment of the influence that historical borders have on territorial claims is a nontrivial task, as it requires new data on the location of all relevant historical international borders. Moreover, care is needed in choosing the period to measure historical boundaries in as well as the subsequent time period to examine how these precedents affect the emergence of territorial claims. We focus on historical boundaries and subsequent territorial claims in Europe.

A focus on Europe has several advantages. First, the advent of the modern system of territorial states occurred in Europe in the 17th century and subsequently spread (or was exported) to other regions of the globe. Specifically, scholars typically point to the 1648 Treaty of Westphalia, which ended the Thirty Years War, as marking the advent of the modern sovereign state.Footnote ⁹ A focus on historical borders in Asia, the Western Hemisphere, or Africa during the 17th and 18th centuries would provide less variation in historical boundaries, as the territorial state became the dominant form of political organization significantly later in those regions. Furthermore, Europe’s status as the region that “birthed” the modern territorial state makes the findings of our study relevant to the rest of the world, as we might expect similar patterns elsewhere during later periods.

Accordingly, we measure historical precedents with data on all European borders after 1650, just subsequent to the Peace of Westphalia and up until 1789, which marks the start of the French revolution. A focus on the period from 1650 to 1789 has several advantages. First, it allows us to analyze the role of international borders that existed during the initial periods of state formation under the Westphalian model. Second, it facilitates the analysis of all territorial claims after the Congress of Vienna (1815). For the last several decades, the field of international relations has focused on international conflict, territorial claims, and numerous other topics from 1816 onwards, which facilitates comparison of our results to those in the existing literature.

An additional advantage to measuring historical boundaries prior to 1790 is that we largely sidestep objections that these historical borders are simply picking up the distribution of nationalities or ethnic groups. The earliest historical period commonly used to discuss the beginnings of nationalism is the start of the French Revolution, which is subsequent to our measurement of historical borders. In any event, Weber (Reference Weber1976) shows that even in France, the country most often identified with the emergence of nationalism, a coherent national identity did not fully develop until the spread of railroads and national publications facilitated increased interaction across regions and localities in the 1870s (see also Anderson (Reference Anderson1991) and Sahlins (Reference Sahlins1989)).Footnote ¹⁰ Of course, leaders might often use nationalism and ethnic identity in making territorial claims after 1815 (as discussed above); however, these pre-1789 historical boundaries have little to do with the distribution of nationalities.Footnote ¹¹ We provide several robustness checks in the Online Appendix that help demonstrate this empirically.

DATA

Our research design requires data on the history of European state boundaries between 1650 and 1789 and territorial claims in Europe after 1815. We create our theoretically key independent variables using newly collected data on the existence and shape of European states (Abramson Reference Abramson2016). To measure territorial claims after the Congress of Vienna, we use the Issues Correlates of War (ICOW) data (Hensel Reference Hensel2001; Reference Hensel2013).

Territorial Changes in Europe

Our main variables of interest capture the density and variability of political boundaries after Westphalia and before the French Revolution. To measure these outcomes we use data collected by Abramson (Reference Abramson2016) that describes the existence, location, and boundaries of every European state before 1790. These data define states based upon the Weberian definition as quasi-monopolists of violence and use three criteria to distinguish between states, statelessness, and other forms of political organization. We treat political units as territorial states if they meet all three of the following criteria:

1. 1. No Direct Military Occupation

   If a territorial unit is militarily occupied by a foreign power, according to this coding scheme it ceases to exist. Similarly, if a political unit successfully conquers a piece of territory, this newly occupied territory is treated as a part of the conquering state. For example, when the Ezzelino or Pallavicini families were able to effectively wield military control over several Italian city-states, they are thus coded as the amalgamation of these units into a single state.

2. 2. The Capacity To Tax

   The power to expropriate resources, or the ability to take from another that which she has produced or owns, is the coercive authority central to statehood. Accordingly, the capacity to tax is used as evidence of the state’s monopoly over coercion. So, for example, when Worms (1184) or Lubeck (1226) ceased to pay imperial taxes and (demonstrably) gained rights to collect taxes and tolls within their boundaries, they are subsequently treated as independent states. Moreover, this allows us to distinguish states from other forms of political organization such as Leagues that maintain a limited capacity to generate revenue. With this perspective it is apparent that the Holy Roman Empire was marked by substantial political fragmentation, capturing the proliferation of the political units that maintained “full jurisdiction. . . rights of legislation, privileges of coining money, levying tolls and (collecting) taxes (Bryce Reference Bryce1920, chap. xiv).”

3. 3. A Common Executive

   An important feature of a unified territorial state is sharing a common executive. Many early-modern states were composites of, sometimes, geographically and institutionally disparate groups. That is, although they shared a common executive many subnational groups maintained distinct parliaments and other subnational institutions granting particular rights and prerogatives to certain regions or groups and not to others. For example, although they belonged to the Spanish state Castile, Aragon, Valencia, and Navarre maintained distinct parliamentary bodies each with separate and distinct political and economic rights. When there exist semiautonomous groups like this that share a common executive we treat them as a single state, as long as they also satisfy the prior two criteria.

We use this definition of what political units are considered territorial states in post-Westphalia Europe to identify the relevant set of historical boundaries. The data describing international boundaries for all states in post-1650 Europe are measured in five year panels and are taken from a number of sources. We use Geographic Information Systems (GIS) to measure the density and variability of international boundaries, as the location of boundaries is inherently spatial. Two of the main sources from which the GIS boundaries are constructed are the Centennia Historical Atlas (Reed Reference Reed2008) and the Euratlas Digital Atlas (Nussli Reference Nussli2010). The Nussli data is measured in one-hundred year panels whereas the Reed atlas utilizes a much higher frequency approach, taking observations in tenths of years. The boundaries are used as defined by both datasets aligning them at every hundred-year mark based upon the coding scheme defined above. The Nussli data matches the Reed data nearly perfectly at 100-year intervals. Because the Reed data is not geo-referenced, GIS compatible shape files are built by manually constructing the boundaries from reprojected images provided by the atlas and then referencing each observation using the European Alpers Equal-Area projection system.

The Nussli data have been used in several prominent publications and are quite accurate (Stasavage Reference Stasavage2011a; Reference Stasavage2011b; Blaydes and Chaney Reference Blaydes and Chaney2012). Nevertheless, even after combining the data from these digital sources there are still a number of imperfections, for example, units coded as independent states are absent from the reconstructed shape files. These tend to be small independent principalities, ecclesiastical units, and city-states that were not picked up by the historical geographers who created the digital reproductions from which the maps are constructed. In order to rectify these flaws and prevent the potential problems that could plague any statistical analysis based on the uncorrected data, a number of historical and contemporary primary source maps are used to create high frequency boundary changes for these missing units. From this combination of secondary and primary cartographic sources boundaries for all political units that are not present on the initial maps from the Nussli and Reed data are projected that meet the coding criteria. Using known pieces of physical geography, known political boundaries, and the location of cities and towns to properly reference these maps, shape files are created that, with a high degree of accuracy, reflect the geographic scale of each unit. For each of these units the history of their boundary changes is tracked, i.e., expansions and contractions, and the shape files are adjusted accordingly.

Our units of observation are arbitrarily located 225 × 225 kilometer grid squares (approximately 50,000 square kilometers).Footnote ¹² There are several important inferential and measurement advantages to the use of grid squares as a unit of analysis, which we clarify with comparison to the commonly used dyadic unit of analysis. First, the shape and location of arbitrary grid squares are not plausibly endogenous to the variety of political, economic, and military factors that determine the shape and location of a dyad’s boundary. In fact, the same reasons that make Europe the most theoretically interesting region for this study, e.g., its long history of experience with the territorial state, also elevate our concerns about endogeneity if we use post-1815 dyadic borders as our unit. It is important to note possible concerns with the grid-square approach, such as their size. If we choose grids that are too large, a grid might contain historical boundaries that are far from contemporary borders, making the association with claims in the same grid spurious. However, our grid squares are not especially large relative to the size of European states, as a 50,000 square kilometer area is similar in size to current Bosnia and Herzegovina or Slovakia.Footnote ¹³ This was a factor in our choice of grid-square size, with the other factors being measurement issues and the fact that making grid-square units too small can artificially decrease standard errors.

Measurement of our key variables within grid squares ensures that we do not rely too much on the precise spatial location of our historical data relative to contemporary boundaries. While our data is quite accurate, political boundaries were somewhat less precise in the 17th and 18th centuries relative to the 19th and especially 20th centuries (Biggs Reference Biggs1999; Branch Reference Branch2014). Placing historical precedents in grid squares largely sidesteps these potential pitfalls.Footnote ¹⁴ In contrast, the use of post-1815 dyadic boundaries to define our unit makes the precise spatial accuracy of all variables relative to these boundaries essential, while the grid-square units allow us to demonstrate that any results are not sensitive to perturbing the location and size of the units.

To more concretely demonstrate that the grid squares are superior in our application, we also construct all of our data using the post-1815 dyad as the unit of analysis. We are careful to construct the dyadic test in a way that shows our grid-square analysis does not suffer from several potential drawbacks. First, given that most territorial claims made by a state are proximate to its current borders, we want to ensure that our results with grid squares do not reflect historical boundary precedents that are actually far from borders, e.g., 200 kilometers, and associate these with unrelated territorial claims. To do this we create dyadic data with 50 kilometer buffer zones around each post-1815 state’s border and limit our measurement of historical boundary precedents to these areas close to current borders.Footnote ¹⁵ We replicate the analysis of our main results, i.e., Table 1 below, using the post-1815 dyad as the unit of analysis and find very similar results, which are reported in our Online Appendix. Estimation of models using the post-1815 dyad as the unit of analysis also makes possible systematic evaluation of our argument that selection on unobservables is more of a threat to inference in the dyadic analysis than the grid-square analysis. Specifically, we subject each of these approaches to the statistical test developed by Oster (Reference Oster2013), which allows researchers to evaluate the degree to which unobservable factors are likely to bias estimates of their key variable. The results, reported in the Online Appendix, suggest that the grid-square estimates are very robust to potential violations of the exogeneity assumption, while the dyadic analysis are not robust at any specification of the test.Footnote ¹⁶

TABLE 1.

The Effect of Historical Borders on Territorial Claims

Notes: ***p < 0.01, **p < 0.05, *p < 0.1.

Robust standard errors in parentheses.

Using these data and the grid square as our unit we construct our primary measure of the presence of competing historical border precedents. We start by measuring the density of historical borders in each grid square in five-year panels. We use this information to create two measures. First, we create Historical Border Density, which captures the mean density of historical borders from 1650 to 1789 on a given grid-square unit. Second, we create our primary measure of competing historical precedents, Historical Border Variability, which captures temporal variation in the density of borders over the 140 years preceding the French Revolution. Thus, while the density measure identifies areas that have higher levels of average border density, the variability measure identifies areas that experience more variation in border density from 1650 to 1789.

We construct measures of border density and variability for each grid square as follows. Every five years between 1650 and 1789, for each grid-square i, we observe the length of borders within its bounds. Because there is variation in the amount of land within each grid square (some include coast) we then divide the length of borders by the area of landmass contained within each unit, yielding a measure of border density for unit i in period t, which we call Historical Border Density _it . We take the mean of the measure across all t to produce our averaged measure Historical Border Density _i for each unit i.

Given our measure of average Historical Border Density _i , we are able to construct a measure of the variability of border density across time. A measure of border variability allows us to assess whether areas near post-1815 international borders that experienced more border changes across time, and thus have multiple historical precedents, are especially prone to territorial claims. Border variability for each territorial unit i is measured as follows:

$$\begin{eqnarray*} &&\text{\emph {Historical Border Variability}}_i\nonumber\\ &&\quad = \sum _{t=1}^T\sqrt{(Density_{it} - \overline{Density}_i)^2}. \end{eqnarray*}$$

Thus, Historical Border Variability _i effectively captures how variable borders are across time for each grid square.Footnote ¹⁷ As noted above, variability is our most theoretically important measure, as it identifies whether there are multiple historical precedents. However, Historical Border Density _i is an important control variable in all of our models, as it identifies the density of state boundaries in a given grid square. Thus, inclusion of Historical Border Density _i in our regression models ensures that our key measure of Historical Border Variability _i does not just proxy for the density of states and borders in a given grid square. To account for the fact that Historical Border Density _i and Historical Border Variability _i are right skewed, we take the natural logarithm of both measures.

Given the central theoretical importance of competing historical precedents, we also develop an alternative measure that identifies each particular boundary change. To produce a measure at the grid-square level, we identify all j boundaries in a given grid square i in each time period t. We create Border Changes_it by giving a value of 1 to each boundary j that changes in time period t, and a value of 0 to each boundary j that remains in place, then taking the sum across j. To create a measure of the number of historical precedents in each grid square i, we sum across t, which yields Border Changes_i = ∑ ^T _{t = 1} Border Changes_it . Thus, Border Changes _i measures the number of historical precedents in each grid square.

Given that Border Changes _i specifically identifies each boundary change, we also use it to construct measures of how long historical precedents were in place as international boundaries and how old each historical precedent is. To measure the duration of precedents as boundaries in each grid square we first calculate the amount of time each border j was in place in each grid square i. We then take the average of this measure across all borders in each grid square to produce Average Border Duration _i . Given that Border Changes_i measures the volume of precedents in an area and Average Border Duration _i tells us how long the average border remained in place on a given unit before becoming a precedent, estimation of the interaction between Border Changes_i and Average Border Duration _i identifies whether leaders systematically make claims over better established precedents in areas where boundaries are historically quite variable.

Our measurement of how old precedents are in each grid square is analogous to the construction of Average Border Duration _i . Specifically, for each precedent j in grid square i we calculate how many years passed between the time the border changed (and became a precedent) and 1815. Given this information for each of j precedents in grid square i, we take the average age of all precedents in grid square i to obtain Average Time Since Last Change _i . To assess whether leaders systematically make claims over more recent precedents in areas with more historical boundary variability, we estimate an interaction between Border Changes_i and Average Time Since Last Change _i .Footnote ¹⁸

THE NUMBER AND TIMING OF TERRITORIAL CLAIMS

We begin with the effect of historical boundary variability on the total number of territorial claims made on a given piece of territory. The geographic distribution of territorial claims are mapped in the left-hand panel of Figure 1, while Historical Border Variability _i is mapped in the right-hand panel. The boundaries of Europe in 1650 are also mapped in each panel. The map of Historical Border Variability _i in the right-hand panel of Figure 1 demonstrates a lot of spatial variation in pre-1790 historical boundaries. It is also clear that boundaries vary in many areas relative to their starting point in 1650. From visual inspection, there is a clear spatial relationship between where historical boundaries are most variable and where more territorial claims emerge.Footnote ²³

FIGURE 1.

The Number of Territorial Claims across Territory (Europe (1650))

To be more systematic, we estimate several regression models to assess the effect of competing historical precedents on the number and timing of territorial claims. Table 1 shows the results of ordinary least squares and negative binomial models of the number of territorial claims. We first report simple bivariate model specifications that only include Historical Border Density _i or Historical Border Variability _i . Thus, models 1 and 2 only include Historical Border Density _i , while models 3 and 4 only include Historical Border Variability _i . Next, we include both measures in models 5 and 6. Finally, we repeat the pattern of model specification for these two variables from models 1–6 in multivariate specifications that also include all of our other measures of territorial value, i.e., models 7–12. This specification strategy helps us to demonstrate three key things about the effect of Historical Border Variability _i . First, we can establish that Historical Border Variability _i is not a proxy for the number of borders or states by showing that its effect is not influenced by inclusion of Historical Border Density _i . Second, the inclusion of our measures of territorial value in models 7–12 helps us determine whether the effect of Historical Border Variability _i in models 3–6 are a product of it being a proxy for these included measures of territorial value. Third, comparison of the coefficients in the bivariate models with those of the full models in columns 11 and 12 helps as conclude that the effect of Historical Border Variability _i is not driven by systematic connection with omitted measures of territorial value. As noted by Altonji, Elder and Taber (Reference Altonji, Elder and Taber2005), comparison of the size of coefficient estimates on a variable before and after including other important covariates provides a good heuristic for whether selection on omitted variables drives an estimate.Footnote ²⁴ Oster (Reference Oster2013) builds on this idea with a formal test of whether an estimated coefficient is robust to selection on unobservable variables, which we also implement. We take the same approach in our analysis of claim timing in Table 2.

TABLE 2.

The Effect of Historical Borders on the Timing of the First Territorial Claim

Notes: ***p < 0.01, **p < 0.05, *p < 0.10.

Robust standard errors in parentheses.

Across specifications that include our measures of economic and military value we find that the relationship between the existence of multiple historical precedents and the number of territorial claims is both statistically and substantively significant. The magnitude of the effect of Historical Border Variability _i on the number of claims is large, as a one standard deviation increase in this variable from its minimum increases the predicted number of claims from 0 to more than 3. Furthermore, when we include both density and variability in the same specification (columns 5–6 and 11–12 in Table 1) or include both measures along with the measures of territorial value, the estimated effect of border variability remains very similar to what it is in bivariate specifications. The robustness of the effect of Historical Border Variability _i to the measure of mean density in models 5–6 and 11–12 nicely demonstrates that it does not reflect borders being more variable in some areas simply because there are more borders in these grid squares.Footnote ²⁵ The robustness of the coefficient on Historical Border Variability _i to inclusion of all other measures of territorial value demonstrates that it is not a proxy for these measures and that it is not very plausible that it is a proxy for omitted measures of territorial value (Altonji, Elder, and Taber Reference Altonji, Elder and Taber2005). Moreover, the formal test developed by Oster (Reference Oster2013) reveals that our finding is robust to the most difficult possible specification of this test.Footnote ²⁶

The models in Table 2 represent two complementary approaches to the analysis of how historical boundary variability affects the timing of territorial claims. First, in the odd numbered columns we estimate the simplest possible model, regressing the number of months from November 1815, when the Treaty of Paris was signed, until the first territorial claim was made in a unit with OLS. However, the OLS model does not correct for grid squares that never experience a claim, i.e., right-censored units. Thus, we also estimate Cox proportional hazard models that correct for right censoring to ensure this does not bias our results. The OLS and Cox proportional hazard estimates are summarized in Table 2. Note that a negative coefficient in the OLS models indicates a variable is associated with claims being made quickly in a grid square, while the opposite is true for the Cox models where the coefficient indicates the influence of a variable on the hazard rate for claims.

The results across all 12 model specifications in Table 2 are substantively consistent with those in Table 1.Footnote ²⁷ The magnitude of the estimates for mean density and variability are similar across the models if we only include one of these two measures, e.g., model 2 or 4, but the effect of variability is robust across all specifications while the effect of mean density washes out when we also include variability in the specification, e.g., models 11 and 12. This again nicely demonstrates that the presence of multiple historical precedents drives the emergence of territorial claims. Moreover, the coefficient on Historical Border Variability _i again easily survives the formal robustness check developed by Oster (Reference Oster2013).

Several of the measures of territorial value also attain statistical significance in some specifications, although no variable is as consistently associated with claims as Historical Border Variability _i . The length of rivers is positively associated with the number and timing of territorial claims, which is consistent with the notion that rivers are of economic and military value (Strachan Reference Strachan1983). Urban population is also significant in the OLS models of the number of claims, although it fails to attain significance in the negative binomial models. Urban population is also associated with quicker claims being made on a unit, although the coefficient is not always significant. Terrain ruggedness has significant effect across all specifications except the negative binomial model of the number of claims that includes all other regressors, i.e., model 12 in Table 1. We find that agricultural productivity is a significant predictor of claim timing in Table 2, but find that it has no effect on the number of claims in an area.

These results in conjunction with those of Table 1 suggest that economic and military factors do affect the number of claims, as many historians and much of the existing literature suggest. The fact that we condition on variables that measure economic and military significance and that these variables are also important makes clear that historical border variability is not just serving as a proxy for military or economic value. The latitude and longitude indicators that control for any additional features that make some territory more attractive to leaders perform inconsistently across the specifications. Moreover, the results do not change from those reported here if we estimate the effects of both latitude and longitude on claim emergence nonparametrically, which allows a lot of flexibility in estimation of the effect of being in different parts of Europe. This gives us further confidence that our results do not suffer from an omitted spatial variable that measures territorial value.Footnote ²⁸

We also estimate interactions between each of our measures of territorial value and Historical Border Variability _i to assess the idea that historical precedents provide opportunity to make a claim, while value indicators such as Iron Production provide incentive. However, none of these interactions reach statistical significance in the models of the number of claims and we only find a modest interactive effect between Historical Border Variability _i and river density when we examine claim timing. This is the case regardless of whether we estimate a separate model for each interaction or if we interact each measure with Historical Border Variability _i in a single model. Regardless of our approach Historical Border Variability _i remains statistically significant with a substantively similar effect on the number of claims.Footnote ²⁹

As an additional assessment of the idea that perpetual conflict over especially valuable territory drives the connection between precedents and claims, we construct a test of whether militarized conflict propensity drives the estimated relationship to Historical Border Variability _i . We test this idea using the causal mediation framework developed by Imai et al. (Reference Imai, Keele, Tingley and Yamamoto2011), which allows us to statistically identify whether the effect of historical borders on the number and timing of territorial claims is mediated by an intervening post-treatment variable, in this instance post-1815 militarized conflict. We operationalize the observed propensity for conflict as the the count of militarized interstate disputes taking place within each of our grid squares after 1815.Footnote ³⁰ If the effect of Historical Border Variability _i operates through MID propensity, then we will observe a large mediating effect of MIDs and a diminution of the direct effect of historical variability. However, the estimated direct effect of Historical Border Variability _i on the number and timing of claims is essentially identical to what we find in Tables 1 and 2 and we find no evidence of a mediating effect through MID propensity. These results are fully described in the Online Appendix.

Historical Border Variability and Claims Drawn Along Precedent

The results presented thus far show that areas with multiple historical precedents experience a significantly higher volume of territorial claims that are made more quickly. However, we do not explicitly distinguish among claims that are drawn along historical precedents and claims that do not follow precedent in Tables 1 or 2. Thus, it is conceivable that the territorial claims made near multiple historical precedents are not actually drawn along these historical lines. To demonstrate that our findings are the result of leaders drawing claims along historical precedents, we separate claims that follow precedent and claims that do not follow precedent and replicate the main models in Tables 1 and 2. Out of the 133 distinct territorial claims in the data, 85, or around 65%, follow historical precedents.

The results when we distinguish between precedent-based claims and claims without precedent clearly demonstrate that our key findings are driven by claims drawn along historical precedent. Table 3 shows that historical border variability, our key measure of multiple historical boundaries, is strongly related to the volume of claims that follow precedent but not to claims that that do not follow precedent. Specifically, the effect of historical border variability is similar in the models of claims over historical precedents in columns 1–6 of Table 3 to what we find in the models of all territorial claims in Table 1.

TABLE 3.

The Effect of Historical Borders on Territorial Claims By Precedent

Notes: ***p < 0.01, **p < 0.05, *p < 0.1.

Robust standard errors in parentheses.

In contrast to the results for claims drawn along precedent, we find that historical border variability is not systematically connected to the much rarer set of claims without historical precedent. Historical border variability loses statistical significance and importantly, also loses substantive significance when the focus is on claims without precedent. Interestingly, our control for the density of borders in an area consistently predicts the number of claims that do not follow precedent, i.e., columns 11 and 12. This suggests that claims without precedent tend to occur where there are more borders to dispute, but that the presence of multiple competing historical borders is not associated with these claims. The results when we focus on the timing of claims that follow precedent versus the timing of claims that do not follow precedent are similar and are reported in the Online Appendix.Footnote ³¹

INCENTIVE AND VARIATION IN THE HISTORIES OF PRECEDENTS

Thus far, our findings are consistent with the idea that persistent coordination around precedents drives subsequent claims. To recap, we find that the effect of Historical Border Variability _i is not affected by the inclusion of our measures of territorial value. Moreover, we find little to no evidence of interactive effects between measures of economic or military value and Historical Border Variability _i . These results weigh against the idea that perpetual conflict over the same valuable territories drives the connection between precedents and claims. In this section we explore variation in the origins and histories of precedents to provide critical tests of whether persistent coordination effects account for leaders’ incentive to make territorial claims over historical borders. Specifically, we test the claim that historical precedents are increasingly attractive when they were international borders for longer periods of time and when they were more recently removed from the map.Footnote ³² We also test the claim that persistent coordination around precedents will be stronger in areas where boundaries were more variable, as we analyze the number of historical precedents in a grid square conditional on how well established the average historical boundary was and how old the average precedent is. Our expectation is that the effect of how well established precedents are and how old they are on claims is larger in areas where boundaries have been more variable. The models with the interactions are the theoretically interesting specifications, so we focus our discussion on those.Footnote ³³

The estimated interactive effects between Border Changes _i and Average Border Duration _i are strongly supportive of the idea that better-established precedents are more attractive, particularly in areas where borders have changed more frequently. The interaction between Border Changes _i and Average Border Duration _i is positive and significant in all specifications. Thus, the number of precedents increases the number of claims conditional on some precedents in a grid square being well established.Footnote ³⁴ The negative and statistically significant interaction between Border Changes _i and Average Time Since Last Change _i in columns 7 and 8 are also consistent with the idea that persistent coordination effects drive claims. This indicates that the number of precedents increases the number of claims conditional on these precedents, on average, being relatively recent.

The substantive importance of both interaction effects are made clear by examining Figures 2 and 3. Figure 2 shows how the predicted number of claims from column 10 of Table 4 is influenced by increasing both the number of precedents, or Border Changes _i , and the length of time these precedents were international borders, or Average Border Duration _i . The four graphs in Figure 2 all allow Average Border Duration _i to vary on the x axis and vary Border Changes _i across the four panels. The panel on the far left depicts a grid square with one historical precedent, the second panel a grid square with five precedents, the third a grid square with 10 precedents, and the final panel on the far right a grid-square with 15 precedents.Footnote ³⁵ The predicted number of claims is depicted on the y axis.

FIGURE 2.

The Duration of Precedents as Border Institutions and Subsequent Territorial Claims

FIGURE 3.

The Age of Border Precedents and Subsequent Territorial Claims

TABLE 4.

Average Border Duration, Average Precedent Age and the Emergence of Claims

Notes: ***p < 0.01, **p < 0.05, *p < 0.1.

Robust standard errors in parentheses.

The graphs show that the relationship between these two variables is truly conditional in the sense that both have greater effects on territorial claims as they mutually increase. Thus, while the effect of Average Border Duration _i is positive across all panels with multiple precedents, the slope of its effect increases sharply as Border Changes _i increases. Moreover, the effect of Average Border Duration _i is not statistically significant in the panel that depicts a grid square with one border change, while it is significant for most of its range when there were at least five precedents. Figure 2 provides striking evidence supportive of persistent coordination driving claims, as the proportion of precedents that are claimed increases sharply in Average Border Duration _i in areas where there are multiple precedents. For example, when there are five precedents with an average duration of about 50 years, three of them are claimed, on average.

Figure 3, which is based on column 8, shows a similarly important interactive effect between Border Changes _i and Average Time Since Last Change _i . The predicted number of claims is decreasing in the average age of precedent when there are more than one of them, an effect whose slope steepens as we examine grid squares with more precedents. Again, we find no significant effect for either variable when there is only one precedent in a grid square, i.e., the left-most panel. The steepening of the slope of the effect as we examine areas with more historical precedents shows that precedents more quickly wane in attractiveness as potential claims when they are in areas where boundaries historically changed more frequently. This is strikingly consistent with our expectation that leaders focus their consolidation efforts strategically on areas that are at greater risk to experience claims.

The results in Figure 2 also imply that there are limits to how much variability historical borders can exhibit while remaining attractive to leaders. These results suggest that Historical Border Variability _i , our primary variable, should start to have lesser effects on claims when variability in historical borders is extremely high, as borders would have changed too often for precedents to become well established. We explore this idea with a generalized additive model (GAM) version of the negative binomial model in column 12 of Table 1; the only difference here is that the effect of Historical Border Variability _i is estimated with penalized regression splines, a flexible nonparametric approach that imposes few restrictions on the shape of the estimated effect between Historical Border Variability _i and territorial claims. Figure 4 shows the estimated effect of historical border variability when we model its effect on claims with splines penalized for smoothness.Footnote ³⁶ The plot strikingly shows how the predicted number of territorial claims increases rapidly as Historical Border Variability _i increases across most of its range, and subsequently declines when past border variability reaches very high levels. It is worth emphasizing that the portion of the plot in Figure 4 that shows a decreasing effect for Historical Border Variability _i represents only about 12% of the data. Thus, for almost 90% of the data, increases in Historical Border Variability _i lead to an increase in claims, while the effect of past border variability is only decreasing when it reaches very high levels.

FIGURE 4.

Non-parametric Estimate of the Relationship between Historical Border Variability and the Number of Claims

Finally, recall that if perpetual conflict over the most valuable territories accounts for the clustering of precedents and subsequent claims, this implies that border changes consistently occurred in the same areas over time. However, the correlations between Historical Border Variability _i measured during the periods 1650–1699, 1700–1749, and 1750–1790 are quite low. The correlation between Historical Border Variability _{1650 − 1699} and Historical Border Variability _{1700 − 1749} is about 0.55, while the correlation between Historical Border Variability _{1700 − 1749} and Historical Border Variability _{1750 − 1789} is only 0.28. Thus, it is simply not the case that the same areas experienced border changes across the 140 years in which we measure historical precedents.

The results shown in Table 4 and Figures 2–4 provide evidence that persistent coordination effects drive subsequent territorial claims rather than persistent conflict over especially valuable territory. The finding that leaders make more territorial claims in areas with historical precedents that were better established as international borders and more recently became precedents is consistent with the importance of persistent coordination effects, but not of perpetual conflict over valuable territory. Moreover, there is really no way to explain the nonmonotonic effect of variability in historical borders on claims depicted in Figure 4 with the perpetual conflict explanation. Rather, the idea that border changes reflect perpetual conflict over the most valuable land suggests that the effect of Historical Border Variability _i should have its largest positive effect at its highest values, which is clearly not the case.

COMPETING HISTORICAL BOUNDARIES AND THE POLISH STATE

The results above provide consistent support for our arguments across a large number of statistical tests. Here we detail an important case, that of post-WWI Poland, to demonstrate that our interpretation of the statistics is born out when we examine a case in more detail. Poland is a good case to examine because its political boundaries have varied a lot and several of the relevant historical precedents were in place for long periods of time. Furthermore, Poland’s post-1919 borders were subject to five territorial disputes, which provides an opportunity to see if these claims are tied to the existence of competing historical precedents.Footnote ³⁷ We briefly outline the history of Poland’s boundaries and demonstrate that 20th century Polish leaders, and their opponents across the border, used historical precedents to make their cases. In the interest of space, we focus exclusively on the dispute between Soviet Russia and Poland.

At its largest, the Poland-Lithuanian Commonwealth encompassed more than a million square kilometers—nearly double the size of France. In 1795, before it was wiped from the map for the subsequent 123 years, the area contained within Polish boundaries was just over 200,000 square kilometers—or roughly the size of England. By our calculations, between 1635 and 1795, there were 12 major revisions of the territorial bounds of the Polish state. These revisions provided political actors in Poland and her competitor states conflicting precedents from which to assert rights of territorial control well into the 20th century. Furthermore, several of the precedents that historically bounded a large multiethnic Polish state were in place for long periods, which provided relatively attractive precedents to leaders staking territorial claims in the early 20th century. Depending upon which historical precedent Polish political actors chose to focus on, there existed between 386 and 203 years of union before the First Partition of Poland in 1772, a history Polish leaders could rely upon to justify claims to territory. Similarly, Russian political actors could lay claim based upon the 147 years of control which followed the First Partition of Poland. Moreover, as Wolf (Reference Wolf2005) demonstrates, these partition borders continued to coordinate local economic behavior well into the 1930s.

The events surrounding the reformation of the Polish state in 1918 highlight how the existence of precedents that were international borders for long periods led to territorial dispute between the Second Polish Republic and the Soviet Union after 1919. Following their surrender at Versailles the German armies retired from the Eastern front and the Soviets and Poles both laid claim to the former German sectors running roughly from Vilnius in present-day Latvia through the corridor running from Kiev to Lviv in Ukraine. The situation on the ground was one that is favorably described as ambiguous. Whereas Polish historiography tends to describe a Soviet invasion of Poland, Russian historians describe Polish aggression into established Soviet borders. A prominent historian notes that “the borders ‘belonged’ to nobody in 1919” (Davies Reference Davies1972, 26).

This was because once the Germans left there was no lasting, recognized, border that both parties could use to divide territory; the Soviets and Poles alike had historical precedent with which to claim territory and attempt to assert control. The Treaties of Andrusovo (1672) and “Eternal Peace” (1686) first codified the division of Ukraine between Tsarist Russia and the Polish-Lithuanian Commonwealth, awarding and then affirming control of “Right Bank” Ukraine, save Lviv, to the Poles and the rest of Ukraine to the Russians (Allen Reference Allen2014, 158; Ragsdale and Ponomarev Reference Ragsdale and Ponomarev1993, 108–110; Stevens Reference Stevens2013, 168). However, over the next century the three partitions of Poland between 1772 and 1793 gradually ceded all of this territory to Russia, making a rump Polish state that was to eventually entirely incorporated into the Tsar’s domains (Lukowski Reference Lukowski1998).

The independent state claimed by early 20th century Poles entailed recreating prepartition Poland-Lithuania. For example, Roman Dmowski, chief Polish delegate at Versailles, sought to establish a Poland along the pre-1772 partition borders. Marshal Pilsudski, head of the newly formed Polish army and future head of state similarly sought to create a multinational Polish federation along these same boundaries (Davies Reference Davies1972, 29–30; Dziewanowski Reference Dziewanowski1969, 377; Iivonen Reference Iivonen1990, 174). The Soviets, on the other hand, viewed themselves as the inheritors of the Tsarist realm and, while recognizing the inevitability of an independent Polish state, desired to base its territorial bounds on the final partition of 1795 (Davies Reference Davies1972, 295; Lieven Reference Lieven2002, 30).

A notable fact in this case is that national identity played little role in shaping these territorial claims—the geography claimed by both sides was neither Russian nor Polish but instead consisted of ethnic Lithuanians, Belorussians, Ukrainians, and Jews. Similarly, the area claimed by both the Soviets and Poles provided little in the way of natural or defensible frontiers. Not easily garrisoned, it was particularly difficult to maintain large armies in defense of this extensive border. Its railways and towns, both few and far between, made it difficult to transport men and goods across substantial distance. And lacking geographic barriers to impede the movement of armies there were few “natural” borders upon which either side could credibly make claims. Rather, both sides made claims based on competing and relatively well-established boundary precedents, which played a key role in the territorial dispute that culminated in the Polish-Soviet conflict of 1920.

SUPPLEMENTARY MATERIALS

To view supplementary material for this article, please visit http://dx.doi.org/10.1017/S0003055416000381