Mapping Interstate Territorial Conflict

Population of Cases

The data set is based on the population of interstate territorial disputes identified by Huth and Allee (2002). The start date of 1947 employed here reflects a goal of capturing all territorial disputes that extended past or began after the World War II settlement in Europe. This means that virtually all disputes post-1945 are included, with the exception of several short-lived European disputes that were settled by the 1947 peace agreements. Readers interested in a full description of the sample criterion should consult Huth and Allee (2002), particularly Appendix A (pp. 298-304), which lays out the coding rules. Briefly, the data set includes disputes between the governments of independent states that arise from competing claims to territory. These can involve disagreements over the location of the international boundary between the states, either wholly or in part, disagreements over the ownership of offshore islands, or cases in which a state does not recognize the sovereignty of another and claims the entire territory for itself (Huth and Allee 2002, 300). Three kinds of disputes can be identified depending on whether the territory in question is homeland or colonial/dependent territory for one or both states: colonial–colonial (e.g., the dispute between the Netherlands and France over the border between Suriname and French Guyana), homeland–colonial (e.g., the dispute between Saudi Arabia and Britain over the former’s border with Oman), and homeland–homeland (e.g., the same dispute as before but after Oman became independent in 1971).

The data set primarily includes disputes over land areas, and the map reflects this focus. Disputes over maritime boundaries appear only to the extent that they implicate offshore islands. Thus, the overlapping maritime claims in the South China Sea appear as disputes over various islands and atolls in the Spratly Island chain. Similarly, the dispute between Greece and Turkey over their continental shelves in the Aegean Sea does not enter the data set until 1996, when the dispute implicated the islands of Imia/Kardak. The same is true for most disputes over the division of rivers, which typically reflect disputes over the ownership of islands within those rivers (e.g., Lété Island between Benin and Niger). An exception is the dispute over the Shatt al-Arab waterway between Iran and Iraq, which revolved around the division of the river with respect to navigation rights. In this case, the difference between the Iraqi claim to the whole river and the Iranian claim to the thalweg was mapped.

Given the focus on interstate disputes, separatist conflicts involving nonstate groups are excluded unless an outside state specifically claimed the separatist region. This rule creates some difficult cases because states sometimes advance territorial claims under the guise of demanding self-determination for a separatist group. Support for such groups does not, however, constitute a territorial claim under the coding criteria. Thus, for example, the conflicts between Armenia and Azerbaijan over Nagorno–Karabakh, between Serbia and Croatia over Eastern Slavonia, and between Russia and Georgia over South Ossetia and Abkhazia are not coded as interstate disputes.

For each case, the starting point was the description of the dispute provided by Huth and Allee (2002). This description led to additional research to determine, as accurately as possible, the location and bounds of the disputed region. ² In a few cases, states had multiple small disagreements at a number of places along the border, but only some areas generated a dispute in the sense of conflicting claims. In those cases, only the latter were mapped, though the existence of additional points of disagreement is noted. For example, the primary dispute between Saudi Arabia and Iraq dealt with the area in the former, diamond-shaped neutral zone near the border with Kuwait; however, they also negotiated over the rest of the border in order to clear up some ambiguity and awkwardness generated by colonial-era lines. In this case, only the former region is mapped. A small number of cases that involve demands for basing rights were excluded.

A number of disputes were split into subdisputes, either because they involved different pieces of territory or because the scope of the dispute changed over time with partial settlements or new seizures. For example, Syria’s demands against Israel are broken into three subdisputes covering different time periods: (1) 1949 to 1967, over the territory in several demilitarized zones created by the 1949 armistice agreements; (2) 1967 to 1973, over the parts of the Golan Heights seized by Israel in the Six-day War; and (3) 1974 to the present, still over the Golan Heights but excluding several pockets of land that Israel returned in the 1974 separation of forces agreement. The dispute between Argentina and Chile is broken into three separate subdisputes with different end dates: the Beagle Channel Islands (1984), the Laguna del Desierto (1995), and the Palena region (1965).

The data identify the challenger in each dispute, defined as the state that sought to alter the status quo in its favor (Huth and Allee 2002, 302). In many cases, the identity of the challenger is quite straightforward, when one state in the dyad makes a claim for territory that is controlled by the other. There are also a number of reciprocal disputes in which both states are coded as challengers, and hence two disputes are recorded. These can arise in two different ways that are treated differently here. One possibility is that both states claim a piece of territory through which there is no status quo border, either de facto or de jure. In these cases, both states are coded as the challenger over the same disputed region. For example, there was no border defined border between Saudi Arabia and the British protectorate of Aden (later South Yemen) in the “Empty Quarter,” so both states are coded as challengers over the entire region that falls between their claim lines. A second possibility is that both states claim territory beyond the de facto or de jure status quo. For example, while the entire region of Kashmir is disputed between India and Pakistan, after the conclusion of their first war in 1949, each controlled a portion of the region and claimed the rest. Thus, the map divides Kashmir into an Indian-controlled region that is the subject of Pakistan’s challenge and a Pakistani-controlled region that is the subject of India’s challenge.

Mapping Disputes

To create digital maps, each case was researched to determine, as accurately as possible, the extent of the territory in dispute. This process involved consulting a variety of sources, both primary and secondary, to determine the exact region(s) that were in dispute and/or the claim lines articulated by the challenging state(s). For each dispute, a polygon was drawn corresponding to the bounds of the disputed region. In general, each polygon is bounded by the claim lines advanced by the states and any boundaries (including shorelines) that were not contested. For example, Venezuela claims all of Guyana’s territory west of the Essequibo River. The disputed region is, therefore, bounded on the east by that river, on the southwest by Guyana’s boundary with Brazil, on the northwest by the existing boundary between the two states (which Guyana claims is the rightful boundary) and on the north by Guyana’s shoreline (see Figure 1a). The base map used to identify existing boundaries and shorelines is the Large Scale International Boundary Lines and World Vector Shorelines from the US Department of State’s Office of the Geographer, hereinafter referred to as the DoS map. These are highly detailed maps from 2012 to 2013 that reflect the US government’s policy on the location of international boundaries (though, as noted below, the United States recognizes some areas as being in dispute). ³

OPEN IN VIEWER

Figure 1.

Examples of disputed regions with precision codes.

The central challenge in this project was to obtain the precise boundaries of the states’ claims in order to identify the disputed region. Although accurate information was available for most cases, there were some disputes for which detailed information about claims was lacking or in which the claims were not articulated with great precision (at least not publicly). To deal with this, each dispute was given a precision code from one to five, corresponding to different categories of uncertainty. In addition, I also created a variable flagging cases in which the map of the disputed region is incomplete. Walking through the different levels of precision, along with examples of each, will provide a good sense of how the disputed regions were drawn. Figure 1 depicts some of the examples discussed below.

The most precisely mapped disputes are those in which the claims align with some feature for which a high-quality digital map already exists. In some cases, the disputed region is bounded by features on the DoS map. This occurs when the disputed area corresponds to an existing country (e.g., South Korea), a former country that has since broken up (e.g., Sudan), or entire islands or island chains (e.g., the Paracels). In addition, the DoS map identifies some areas that are (or were) in dispute, such as Western Sahara, Hong Kong, the disputed region between Croatia and Slovenia inland of the Bay of Piran, and the “no man’s lands” between Israel and the Jordanian-controlled West Bank. A second class of highly precise claims are those that are bounded by rivers, such as the Venezuelan and Surinamese claims on Guyana (bounded by the Essequibo and New rivers, respectively), as depicted in Figure 1a. For such claims, I relied on the maps of river and drainage systems produced by the Digital Charts of the World (DCW) or, in some cases, Natural Earth, a public domain data set. ⁴ A third class of precisely mapped disputes are those that correspond to internal administrative units, for which maps are available from the database of Global Administrative Areas (GADM). ⁵ For example, the Kashmir dispute between India and Pakistan encompasses the Indian state of Jammu and Kashmir and the Pakistani districts of Azad Kashmir and the northern areas (plus the Siachen glacier region, which is marked on the DoS map). Internal administrative divisions can also be used to locate formerly disputed areas that were incorporated into an existing state without a change of boundaries, such as the former East Germany or the Portuguese and French colonial enclaves in India.

The next level down in terms of precision are cases for which a map of the disputed region could be found on a paper or digital image. This was generally the case when claim lines corresponded to some historical boundary that no longer exists, such as Estonia’s and Latvia’s claims for their pre-World War II boundary with Russia (see Figure 1b); when the disputants advanced different maps, such as in the dispute between the China and Mongolia; and/or when the dispute corresponded to features for which no digital map was available, such as the difference between the left bank and median line of the Shatt al-Arab waterway between Iran and Iraq. In addition to books and articles on the individual disputes, the International Court of Justice (ICJ) produced invaluable sketch maps for the disputes that came to it for resolution. When a paper map or digital image were used, producing a digital map requires geo-referencing the image, which involves assigning coordinates to a number of reference points and then transforming the image to obtain the closest possible match with the underlying geography. Once the image is geo-referenced, the relevant claim lines are then traced by hand. Both of these steps can introduce some error. Furthermore, the quality of the maps that could be obtained varied considerably. For example, the Library of Congress had high-quality maps of the neutral zones between Iraq and Saudi Arabia and Kuwait and Saudi Arabia which could be geo-referenced and traced with little error. In other cases, the available maps were relatively crude sketch maps, such as the one produced by a United Nations envoy of the discrepancy between the Iranian and Iraqi views of their land border. While most disputes in this group were given a precision score of 2, some received 2.5, or even 3 to indicate a lower quality map. ⁶

The next two categories of precision are for cases in which a map or highly detailed description of the conflicting claims could not be obtained. In some cases, this was due to a lack of precise information, particularly if the area of the dispute was relatively small. There may also have been a lack of clarity in the claim itself—that is, the disputants never clearly articulated precisely what they wanted, at least not publicly. In the best such cases, descriptions of the dispute permitted a reasonably well-constrained map. For example, Ethiopian demands for a corridor to the Red Sea port of Zeila or Bolivian demands for a corridor to the port at Arica can be mapped approximately, even if the exact width of the desired corridors was never clearly articulated. The Philippine claim on Malaysia (Figure 1c), which corresponds to land granted to Britain by the Sultan of Sulu, cannot be precisely mapped because while the extent of the area was well defined along the coast, it is unclear how the line runs through the interior. Similarly, Somalia’s claim to Somali-inhabited areas of Ethiopia was not precisely defined geographically, but it corresponds approximately to the present-day Somali district of Ethiopia. These cases received a precision score of 3.

In other cases, the drawing of the claim was poorly constrained by the available information. At best, an approximate location could be determined from place names or general descriptions. Particularly if combined with some indication of the size of the disputed area, a reasonably sized polygon could be drawn. In most such cases, the areas in dispute are quite small, so a small circle of appropriate size indicates the area. For example, very little is publicly known about the disagreement between China and Vietnam over their land border. Reports indicate that the core dispute was over 164 areas totaling 227 square kilometers, meaning that the average size of each area was quite small, and detailed maps of the contested areas are unavailable. However, reporting suggested that the two most salient disputes centered on the Ban Gioc/Detian Falls and the Nam Quan Gate. Hence, these areas are indicated on the map with small circles. I similarly lack clear information on the preunification border conflict between North and South Yemen, except for the general location, as depicted in Figure 1d. These cases receive a precision score of 4, and in the (likely) event that some disputed areas were not mapped (such as the remaining 162 areas between China and Vietnam), this incompleteness was coded in a separate variable.

Finally, there are several cases in which detailed information on the conflicting claims was lacking, but we know how the states changed the status quo border in order to resolve the dispute. For example, the Soviet Union and Iran had a series of disputes along their border which were resolved in a 1957 treaty. From the information provided along with this treaty, we can identify where the new border was different from the prior, contested one. The transferred regions give some indication of where the disputes presumably were, but they do not necessarily correspond to the full extent of the conflicting claims, as some of these were dropped or compromised. This case, along with China-Nepal (shown in Figure 1e), Oman-South Yemen/Yemen, and portions of Saudi Arabia-Jordan, received a precision code of 5.

Heterogeneity across Disputes: Unpacking the Effect on International Trade

A first use for these data is to more accurately measure and characterize heterogeneity across disputed regions. There is a good deal of variation in the extent to which territorial disputes generate conflict. Some disputes are more intractable than others and some are more likely to become militarized than others. To the extent that this variation is explained by differing characteristics of the territories under dispute, it is valuable to have precise indicators of what exactly is there. Existing studies depend on assessments of whether the territory is valuable for ethnic, military, or economic reasons (e.g., Huth 1996; Huth and Allee 2002) or scales measuring a variety of tangible and intangible aspects of claim salience (Hensel and Mitchell 2005). While not denying the value of such indicators, spatial methods permit even more precise measurements of certain features. In addition to the area and length calculations done above, disputed regions can be spatially joined to maps of other relevant features, such as population, the location of ethnic/linguistic groups, oil or mineral deposits, road and rail networks, ports, water sources, agricultural land, and so on.

Although length and area are crude measures that capture only some of the relevant heterogeneity across disputes, they suffice to illustrate one potential contribution of this approach. To do so, I revisit an important paper by Simmons (2005) demonstrating that territorial disputes reduce international trade between adversaries, even after controlling for militarized actions. Simmons (2005, 828-29) proposed two mechanisms that might explain this effect. The first is jurisdictional uncertainty, which arises when contested sovereignty creates uncertainty over which state’s rules and laws govern a given transaction. Such uncertainty could raise the costs of moving goods through territory that is contested, or it could dissuade economic actors from investing in contested areas, due to the lack of clear property rights. The second source of uncertainty Simmons (2005) identifies is policy uncertainty, which arises from the risk that states will respond to the territorial dispute by taking trade-disrupting actions, such as engaging in militarized conflict, border closures or fortification, or economic reprisals. Not only do such actions directly affect trade when undertaken, but the risk that they might happen casts a shadow over cross-border economic activity as long as the dispute is ongoing.

Although these two mechanisms are not mutually exclusive, we can use the data created here to make some empirical distinctions and say something about their relative weight. Jurisdictional uncertainty arises from the sheer amount of the states’ territory that is contested and, relatedly, how much of the border is subject to dispute. All other things equal, as more of the total dyadic territory is in dispute, the greater are the uncertainties due to unclear property rights. Similarly, the more of the border, by length, is in dispute, the harder it is for goods to cross the border without passing through a contested area. ¹³ By contrast, we might conjecture that policy uncertainty is related to the level of territorial threat felt by the target state. All other things equal, as more of a state’s land area is claimed, the greater are the stakes in dispute, and the greater are the risks of costly actions such as military or economic sanctions. As mentioned above, territorial threat is proxied by the share of the target’s area claimed by the challenger, summed across both states in the event that both are targeted.

Table 2 presents estimates from several regressions that replicate Simmons’ (2005) results with the addition of three variables measuring the area and length of disputed regions. The sample in that study covered all dyads that were either contiguous by land or were separated by no more than 400 miles of water in the period 1950 to 1995. In all models, the dependent variable is the logged sum of dyadic imports in millions of US dollars in the given year. The control variables include the lagged dependent variable, the logged sum of the states’ gross domestic product (GDP), the logged sum of the states’ total population, the logged distance between capitals, an indicator for the presence of a militarized interstate dispute (MID) in the dyad, a measure of the states’ voting affinity in the United Nations General Assembly, a measure joint democracy derived from the sum of the state’s Polity scores, and a measure of overall trade openness created by logging the product of the states’ total trade over GDP. A variable for the calendar year and region and country fixed effects are also included. Additional details on the sources for these variables can be found in Simmons (2005, 844-45) and the replication materials. ¹⁴

OPEN IN VIEWER

Table 2.

Estimated Effects of Dispute Size on International Trade.

	Original Simmons (2005) specifications				Corrected country fixed effects			Dyadic fixed effects
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
Lagged trade	0.71*** (0.02)	0.71*** (0.02)	0.72*** (0.02)	0.71*** (0.02)	0.76*** (0.01)	0.76*** (0.02)	0.76*** (0.01)	0.65*** (0.02)	0.69*** (0.02)	0.64*** (0.02)
Territorial dispute	−0.36** (0.15)	−0.28* (0.16)	0.07 (0.19)	−0.24 (0.16)	−0.07 (0.12)	0.05 (0.14)	−0.04 (0.11)	−0.02 (0.25)	0.26 (0.22)	0.09 (0.26)
Share of dyadic area		−3.17 (2.50)			−2.11 (1.52)			−1.10 (6.72)
Share of border length			−0.41 (0.56)			−0.12 (0.31)			−0.65 (0.68)
Share of target’s area				−1.17** (0.53)			−0.77** (0.33)			−1.67* (0.95)
Total GDP	0.45*** (0.09)	0.45*** (0.09)	0.41*** (0.12)	0.46*** (0.09)	0.39*** (0.08)	0.36*** (0.11)	0.40*** (0.08)	0.55*** (0.10)	0.47*** (0.13)	0.55*** (0.10)
Population	−0.13 (0.12)	−0.12 (0.13)	0.30 (0.19)	−0.10 (0.13)	−0.09 (0.08)	0.35*** (0.12)	−0.08 (0.08)	0.02 (0.23)	−0.16 (0.30)	0.01 (0.22)
Distance	−0.15 (0.11)	−0.15 (0.11)	0.24** (0.10)	−0.14 (0.11)	−0.34*** (0.08)	−0.30*** (0.09)	−0.34*** (0.09)	0.01 (0.08)	−0.89*** (0.10)	0.01 (0.08)
MID	−0.64*** (0.14)	−0.63*** (0.14)	−0.75*** (0.16)	−0.63*** (0.13)	−0.64*** (0.14)	−0.77*** (0.16)	−0.64*** (0.13)	−0.57*** (0.15)	−0.74*** (0.17)	−0.57*** (0.15)
Openness	0.28*** (0.04)	0.28*** (0.04)	0.28*** (0.05)	0.28*** (0.04)	0.25*** (0.04)	0.25*** (0.05)	0.25*** (0.04)	0.33*** (0.05)	0.28*** (0.06)	0.33*** (0.05)
UN voting	0.24** (0.12)	0.24** (0.12)	0.10 (0.20)	0.24** (0.12)	0.26** (0.11)	0.15 (0.17)	0.25** (0.11)	0.13 (0.15)	−0.01 (0.22)	0.13 (0.14)
Joint democracy	0.01 (0.00)	0.01 (0.00)	0.01** (0.00)	0.01 (0.00)	0.01* (0.00)	0.01** (0.00)	0.01* (0.00)	0.01* (0.00)	0.01* (0.00)	0.01* (0.00)
Observations	14,327	14,327	8,181	14,327	14,327	8,181	14,327	14,327	8,181	14,327
R 2	0.856	0.856	0.859	0.856	0.852	0.854	0.852	0.863	0.862	0.863

Note: Region fixed effects, constant, and year trend variables included but not reported. Robust standard errors, with clustering by dyad, reported in parentheses. GDP = gross domestic product; MID = militarized interstate dispute.

***p < 0.01.

**p < 0.05.

*p < 0.1.

Column (1) replicates Simmons’ (2005) original specification with two changes. First, cases that only entailed demands for military bases were not counted as territorial disputes. Second, the Turkey-Cyprus dyad is dropped due to concerns I raise elsewhere (Schultz 2015) that trade reports in that dyad are misleading. Despite these changes, the estimates closely match those reported by Simmons (2005, 835). Columns (2) to (4) then add three measures of the extent of the territorial dispute in dyads that had disputes: the percentage of total dyadic area in dispute, the percentage of the land border in dispute, and the percentage of the target(s) area in dispute. In models that measure the dispute in terms of length (columns 3, 6, and 9), the sample is smaller because it is restricted to dyads that have a land border. All of the measures of dispute size are negatively correlated with trade but only the last—the total share of each state claimed by the other—is statistically significant at conventional levels. This suggests that what matters is not how much of overall dyadic territory is disputed nor how much of the border is contested but how much of the target state’s territory is under threat. The remaining columns report some additional tests. Columns (5) to (7) calculate the country fixed effects in a more traditional way than Simmons (2005) does in the original paper. ¹⁵ Columns (8) to (10) introduce dyadic fixed effects. The results are as before: of the three measures of dispute size, only the one that captures the extent of the threat to the target in the dyad has a significant effect on trade.

Two findings emerge from this exercise. First, the overall effect that Simmons (2005) identified using a dichotomous measure masks significant heterogeneity across disputes. Simulations based on the estimates in column (7) show that the trade dampening effect of territorial conflicts is not statistically significant unless the claim exceeds 14 percent of the target’s land area. Since the majority of disputes do not meet this threshold, the overall effect is driven by a minority of relatively high stakes cases. Second, the results suggests that the size of the “barrier” through which goods must pass does not affect levels of trade. Instead, the reduction in trade associated with territorial disputes varies with the amount of the targeted state’s territory that is claimed by the challenger. To the extent that this variable proxies the stakes of the dispute to the target, it plausibly captures the risk that the dispute will lead to trade-reducing policies, such as militarized conflict, economic sanctions, border closures, and so on. This exercise thus suggests that the policy uncertainty that comes from territorial threat is the main culprit in explaining why states with territorial disputes engage is less trade.

Disaggregating the Dyad: Does Oil Fuel Disputes?

A second possible use of these data is to examine the determinants of territorial conflicts. Prior work has examined the origins of territorial disputes at the dyadic level (e.g., Huth 1996; Englebert, Tarango, and Carter 2002): that is, is there a dispute between state A and state B? But, as we have seen, there is considerable variation in the extent of territorial claims that states make against one another. States rarely claim all or even most of their neighbors’ territory; indeed, the relatively small size of most claims suggests that states consider some pieces of territory worth claiming and others not. What explains this variation? To the extent that the relevant determinants vary along the length of a given dyadic border, analyses performed at the dyadic level are not well suited to answering this question (Goemans and Schultz 2015).

We can illustrate the benefits of disaggregation—and the potential pitfalls of aggregation—by considering the relationship between oil deposits and territorial claims. The idea that oil fuels “resources wars” is a common theme in popular discourse, and Caselli, Morelli, and Rohner (2015) show a relationship between interstate militarized conflict and oil deposits near the border—an effect they attribute to states’ desire to obtain oil-rich territory. Although this argument make some intuitive sense, Meierding (2014; 2016) argues that “oil wars” are in fact quite unusual and that that there are often good reasons for states to cooperate, rather than fight, over the exploitation of transborder resources. Moreover, an inspection of the Caselli, Morelli, and Rohner (2015) data shows that many of the most conflict-prone dyads with oil near the border were not, in fact, fighting about oil: for example, India–Pakistan, Israel–Egypt, Israel–Syria, Russia–Japan, and Armenia–Azerbaijan. Indeed, in some of these cases, the disputed territory does not overlap with the oil deposits. This suggests that dyadic analysis is susceptible to false positives: cases in which there is both a territorial dispute and oil near the border, but the former does not contain the latter.

Disaggregating to a finer spatial resolution can ameliorate this problem. To demonstrate this, I explore the effect of oil deposits on territorial claims. The sample consists of all directed dyads that were territorially contiguous (or separated by no more than 500 kilometers of water) at some point in the period covered by the data. For each directed dyad, I carve the potential target’s territory into fifty-kilometers grid cells and then restrict the sample to cells that were no more than 500 kilometers away from the border of the potential challenger. ¹⁶ The base map used for this purpose comes from the CShapes data set (Weidmann, Kuse, and Gleditsch 2010). Although somewhat less precise than the DoS maps used for generating the dispute data, the CShapes data track changes in all state borders over time. Thus, we can identify cells that at any point fell within 500 kilometers of the dyadic border. One drawback of this map is that it lacks a number of very small island features, including some in the Persian Gulf, as well as the Spratly, Paracel, and Senkaku islands that are the subject of disputes often attributed to offshore deposits. Thus, the tests here look at the effect of oil on disputes over land or relatively large islands.

For each cell, we can determine whether any territory in that cell was implicated in a dispute and whether the cell contained oil. For the latter, the cells were spatially joined to Petrodata, a geocoded data set of oil and gas fields by Lujala, Rød, and Thieme (2007). A cell could provide access to oil in one of three ways. First, a cell could contain an onshore deposit. Specially, a cell was coded as containing oil if at least 5 percent of the cell’s area overlapped with an onshore deposit; the threshold excludes cells that only glance an oil deposit, the bounds of which are imprecise and likely inflated by the Petrodata methodology. Second, islands and coastal territory could give access to deposits located off shore. To code this, each offshore deposit was cut into small pieces, and each piece was connected to the nearest cell on the nearest coastline. We can thereby identify coastal cells that would likely impart rights to those deposits. Finally, since territorial claims extend outward from the border, a cell might be valuable not because it contains oil, but because it is located between the border and cells that do contain oil. Thus, for every cell identified in the first two steps, I drew the shortest line from that cell to the border. All cells that intersect such a line were coded as being on an “oil path.”

In principle, it would be useful to take into account when the oil was discovered and how that relates to the onset of the dispute. Time-varying analysis is challenging, however, for two reasons. First, discovery dates are missing for a significant number of deposits in Petrodata. Second, the presence of oil may be suspected before it is actually discovered. Still, oil that was discovered after the last dispute in a given cell was resolved is unlikely to have been the cause of the dispute and should be excluded. As a first cut, then, I perform a cross-sectional analysis that explores whether a cell was ever subject of a dispute and whether oil was ever discovered in that cell, unless the discovery came after the last dispute in the cell was resolved. Even with this restriction, this method can mistakenly attribute a dispute to oil that was not discovered until well after the dispute began, thus inducing a bias in favor of finding a positive correlation.

At the directed dyadic level, the probability of a claim by the challenger was 7.5 percent if the target had no oil within 500 kilometers of the border and 12.4 percent if there was some oil (either on or offshore) on the target side, a difference that is statistically significant below the 5 percent level. Thus, the presence of oil in the target is associated with a 66 percent increase in the risk of a dispute at this level of analysis. Changing the spatial resolution, however, flips the sign of this relationship. The probability that a grid cell with oil (including being on an oil path) was the subject of a claim was 1.6 percent; the corresponding probability for a grid cell without oil was 2.1 percent. Thus, grid cells that provide access to oil are about 25 percent less likely to be the subject of claim than those that do not. ¹⁷ Part of the explanation for this sign reversal is that the directed dyadic data include a large number of false positives. Of the 92 directed dyads with oil and a dispute, in only 42 does the claim include at least one cell with access to an on- or offshore deposit. Moreover, there are other cases where oil is incidentally included in much larger claims that we have no reason to believe were driven by this resource: for example, the North Vietnamese claim for South Vietnam. Restricting attention to the 42 directed dyads in which a claim overlapped with an onshore deposit or coastal access to an offshore deposit, we find that 17.0 percent of cells with oil or along an oil path were part of the dispute, while 19.8 percent of cells without oil were implicated. Thus, even in disputes that encompassed some oil, the majority of the cells containing oil were off the table, and such cells were, if anything, less likely to be part of the claim than those without oil.

Table 3 presents estimates from two multivariate models that explore this relationship further. Model (1) presents a logit model in which the existence of a claim over a given cell is a function of whether the cell provides access to oil in each of three ways coded here, along with controls for the distance from the cell to the border and whether the states shared a land border. Model (2) presents the same specification but with directed-dyad fixed effects, implemented using a conditional logit. As is always the case with a fixed effects estimator and a binary dependent variable, directed dyads in which either all or none of the grid cells were disputed are dropped form the sample. Thus, the estimates in column (2) are driven by variation within challenger–target pairs in which some, but not all, of the target’s territory was claimed. In both models, standard errors were clustered at the dyadic level. The results show that the grid cells located on top of onshore deposits are associated with a lower probability of a dispute, while grid cells that give access to offshore oil or that sit on a path to oil are neither more nor less likely to be implicated in a claim than those that do not.

OPEN IN VIEWER

Table 3.

The Effect of Oil on the Probability of a Claim on Grid Cell.

	(1)	(2)
Onshore oil	−0.59* (0.33)	−0.63** (0.25)
Offshore oil	0.12 (0.37)	−0.43 (0.43)
Oil path	−0.15 (0.19)	−0.30 (0.33)
Distance from border	−0.01*** (0.00)	−0.01*** (0.00)
Land border	1.21 (0.84)
Constant	−3.56*** (0.86)
Fixed effects	None	Directed dyad
Observations	121,552	20,723
Dyads	548	87
χ2 statistic	52.22***	37.62***

Note: Robust standard errors, with clustering by dyad, reported in parentheses.

***p < 0.01.

**p < 0.05.

*p < 0.1.

These results do not imply that no territorial claims have ever been motivated by oil, and the omission of a few offshore island and maritime disputes from the sample means the results speak primarily to land-based claims. This is clearly a first cut at these data, and more can be done to unpack whether and when oil deposits increase the risk of conflict. It should also be noted that these results do not necessarily imply that oil plays no role even in disputes where most of the available deposits are off the table. It may be, for example, that international norms and political considerations prevent states from confining their claims to oil-rich regions. States rarely claim territory on the basis of “wants”; instead, they must be able to justify a right to the territory (Zartman 1969, 85). This constraint might cause states to make oil-based claims inefficiently—that is, in a way that includes territory that does not contain oil and excludes some territory that does (see, e.g., Murphy 1990, 538-39). At a minimum, then, these data allow us to estimate the size of this inefficiency; at most, they call into question the belief that oil underlies a significant amount of territorial conflict in world.

The Local Effects of Conflict: Lights on the El Salvador–Honduras Border

Finally, the data can be used to explore the implications of contested sovereignty for political and economic outcomes in affected areas. Gibler (2012) argues that territorial threats create important state-level effects, including greater political centralization and larger standing armies. But it is also possible that territorial conflicts have local effects in and around the areas subject to contestation. To what extent do interstate claims impact the ability of the state to consolidate its control in a given area, the delivery of public goods, and the welfare of people who live there? To answer such questions, analysts can exploit the increased availability of subnational, geo-located data on political and economic outcomes such as the Armed Conflict Location Event Data (Raleigh et al. 2010), the Demographic and Health Surveys (see, e.g, Hegre, Østby, and Raleigh 2009), and satellite imagery of visible light at night time, which has been used not only as a direct indicator of electrification (Min et al. 2013) but also as a proxy for local level development ( Michalopoulos and Papaioannou 2013) and conflict-induced displacement and recovery (Agnew et al. 2008).

To illustrate this application, I examine the legacy of the conflict between El Salvador and Honduras on economic development in the border region using the night-lights data. The territorial dispute in this dyad dates back to the nineteenth century, but it remained dormant for much of the twentieth century until the 1969 Football War (Ireland 1941, 144-59; Calvert 2004, 110-13; Anderson 1981). That war, which arose due to tensions over Salvadoran migration into Honduras, left El Salvador in control of some disputed areas. A 1980 peace treaty delimited 225 kilometers of border over which there was no controversy and identified six land pockets that were still disputed (along with some islands in the Gulf of Fonseca). After several failed attempts to resolve those disputes, the case was brought to the ICJ for resolution. The court’s ruling, handed down in 1992, gave about two-thirds of the disputed territory to Honduras. As a result of the ruling, some 12,000 Salvadoran citizens were transferred to Honduras, and 3,000 Hondurans ended up in El Salvador (Ayala 2011).

The night-lights data start in 1992, which means that we cannot track development in the areas before and after the resolution of the dispute. But we can examine the legacy of the conflict by comparing development in the contested areas since 1992 with that in nearby areas that were not part of the dispute. One plausible hypothesis, consistent with Simmons (2005), is that the dispute retarded development in the affected regions; in this case, these pockets would start at a lower level than surrounding areas but grow faster and potentially catch up in the two decades following resolution. An alternative hypothesis is that the disputed areas remain at a disadvantage even two decades after the dispute was resolved because the people who were transferred were cut off from their conationals. Ayala (2011) reports that Salvadorans on the Honduran side generally refused to adopt Honduran citizenship, and the government has been slow to integrate them politically and economically.

Overlaying the satellite images on top of the map of the dispute shows unambiguous support for the latter hypothesis. ¹⁸ Figure 4 contains three images: panel (a) shows the distribution of night-lights in 1992, panel (b) shows the distribution in 2012, and panel (c) depicts the difference between two. The images have been rendered as negatives, so darker coloring corresponds to greater illumination or a greater positive change in illumination. ¹⁹ The six land areas bounded by the states’ overlapping claims are surrounded by a thick grey border and indicated in panel (a). Not surprisingly, the entire border area is relatively undeveloped compared to other parts of both countries. Even so, the regions that had been contested are striking for their lack of illumination in both 1992 and 2012. Moreover, to the extent that there was development near the border in those twenty years (i.e., the dark areas in panel [c]) it largely bypassed the formerly disputed areas.

OPEN IN VIEWER

Figure 4.

Night-lights on the El Salvador–Honduras border.

Quantitative analysis corroborates the impression conveyed by these images. The data report an illumination value ranging from 0 to 63 on a thirty arc-second grid (which corresponds approximately to one-kilometer grid cells). Among grid cells within a disputed region, the average change in illumination from 1992 to 2012 was 0.10 with a minimum of zero and a maximum of six. Among grid cells within five kilometers of the border but not part of the dispute, the average change was 0.80, with a minimum of zero and a maximum of 16. This difference is statistically significant below the 1 percent level. Thus, there is very clear evidence that this territorial dispute continues to negatively impact the welfare of people living in the affected regions.