A recent project based on GIS-technology was concerned with re-conceptualizing the idea of “borders” and developing new measures of the “nature” of borders. It generated two dimensions for contiguous land borders: ease of interaction and salience (and for a combined effect which I have called “vital” borders). A data set for each contiguous land border was generated, which would be relevant for analyses that broadly cover the period 1980-2000. As noted in a number of writings (e.g. Starr 2002), there exists a wide array of research questions based on the assumption that borders indicate proximity, salience, and ease of interaction which may be addressed by this data set. For example: which types of borders are most or least related to the spatial diffusion/ growth of ongoing violent conflicts? Which types of borders are most or least related to recurring violent conflicts between rivalry dyads? These questions will be addressed. The paper will also point out limitations in the GIS-generated data.

This research project has been supported by grants from the University of South Carolina (Research & Productive Scholarship Award #13570-E120), and the National Science Foundation (SBR-9731056). Questions regarding the availability of the data set should be directed to Harvey Starr. I would also like to thank Roger Chi-Feng Liu for his research assistance.

Paper prepared for the Annual Meeting of the International Studies Association, February 28-March 3, Chicago.

A recent project based on GIS-technology was concerned with re-conceptualizing the idea of “borders” and developing new measures of the “nature” of borders. It generated two dimensions for contiguous land borders: ease of interaction and salience (and for a combined effect which I have called “vital” borders). A data set for each contiguous land border was generated which would be relevant for analyses that broadly cover the period 1980-2000. As noted in a number of writings (e.g. Starr 2002), there exists a wide array of research questions based on the assumption that borders indicate proximity, salience, and ease of interaction which may be addressed by this data set. For example: which types of borders are most or least related to the spatial diffusion/ growth of ongoing violent conflicts? Which types of borders are most or least related to recurring violent conflicts between rivalry dyads?

In the Most and Starr analyses of the diffusion of conflict (e.g. 1980) they investigated whether states which were subjected to the "treatment" of having a Warring Border Nation (WBN) were more likely to become involved in conflict than those without such a treatment. In the analyses presented in Siverson and Starr (1991) the growth of ongoing conflict was investigated in the same way– looking at the “treatment” of having a Warring Border Nation. In addition, to help capture “willingness” (as well as the “opportunity” of the WBN), the treatment of having a Warring Alliance Partner (WAP) was added. As discussed in later work (Starr and Thomas 2002, 2005) one weakness of such analyses is that they rely on a simple "on-off" characterization of borders– does a border exist between State A and State B, yes or no– especially in regard to contiguous land borders. This on-off characterization is a hallmark of using contiguous borders in extant studies that use contiguity as an independent, intervening, or control variable. With the GIS-generated data we can ask whether or not we can do better. Can we go beyond the simple idea that contiguity provides the possibility for interaction between states, and look at proximity/distance/opportunity in another way. Is it the nature of the border rather than simply the existence of a border that affects conflict behavior?

One finding of Siverson and Starr in The Diffusion of War (1991) is that the relationship between joining an ongoing war and being subjected to WBNs is one of "loose necessity." Many states have treatments, but do not join wars. Thus, we are left with questions about the contiguous border that connects a state with a WBN. The GIS-generated data set permits us to investigate the nature of the borders that separate the state from its WBN. Do “more porous borders”– that is, those with a greater ease of interaction, more salient borders (those with a greater “importance” on certain measures), or “vital” borders (those with a greater combination of ease of interaction and salience) facilitate the growth of a conflict? As noted, these earlier studies were concerned with "treatments." We could simply propose the alternative hypothesis that any WBN treatment enhances the probability of war diffusion (or the recurrence of conflict), just as much research supports the position that, in general, any contiguous border promotes conflict. However, more fully specified hypotheses would propose that borders with greater ease of interaction, salience, or vitalness have a greater effect on diffusion or a greater effect on the continuation of an enduring rivalry. Is it simply "borderness" in some vague sense, or these more specific qualities that are involved? The purpose of this paper is to build on Starr (2002) as well as Starr and Thomas (2002, 2005), and take another cut at these questions.

The paper will first provide a brief overview of the GIS-generated border project, including how the measures of ease of interaction, salience, and vitalness were developed, as well as findings from studies using this data set. A new set of questions on conflict– as it spreads or recurs– will then be presented, and an initial cut at those questions with the GIS-data will be presented. Following from the prior and current results, a concluding section will present a partial assessment of what we have learned from this project, including the limitations of this data set.

As noted by a number of scholars, the location of states, their proximity to one another, and especially whether or not they share "borders," emerges time and again as key variables in studies of international conflict phenomena: from major power general war, to the diffusion of international conflict, to the analysis of peace between pairs of democracies.Clearly, a key dimension for many researchers is proximity. The early diffusion research of Starr and colleagues (noted above) moved to the study of borders after concluding that the diffusion of certain phenomena could only be studied by looking at units that were "relevant" to one another-- and that such relevance could be indicated by geographical proximity (see also the work of Lemke 1995, 1996). Proximity, in turn, could be operationalized through "borders." Borders were seen as important indicators of proximity because they had important relationships to both the opportunity and willingness of state actors as conceptualized by Most and Starr (1980) and Starr and Most (1976). One key aspect of borders is that they affect the interaction opportunities of states, constraining or expanding the possibilities of interaction that are available to them. States that share borders will tend to have a greater ease of interaction with one another, and thus will tend to have greater number of interactions. This idea developed from multidisciplinary sources, such as economist Kenneth Boulding's (1962) concept of the loss-of-strength gradient; or geographer G.K. Zipf's (1949) "law of least effort."

Such opportunity might be seen in terms of the number of other countries with which any single state has interaction opportunities. It might also be seen in the degree to which such opportunity exists between any particular pair of states, such as the length of a common border between two countries (e.g Wesley 1962). Secondly, borders also have an impact on the willingness of decision makers to choose certain policy options, in that they act as indicators of areas of great importance or salience. Because other states are close, having greater ease of interaction and the ability to bring military capabilities to bear, they are also key areas of external cues (or diffusion). Accordingly, activities in these areas are particularly worrisome, can create uncertainty, and thus deserve attention. The notion that changes in bordering areas create uncertainty because of their proximity was based on arguments developed by Midlarsky (e.g. 1970, 1975), and applied in Most and Starr (1980).

Starr and Most (1976:10) were also particularly concerned with the "roles that different types of borders appear to play" in war involvement. Different types of border might have differential impacts on both opportunity and willingness. Thus, borders were differentiated in terms of homeland borders and borders generated by colonial territories. This differentiation allowed the testing of whether all territory was seen as equally important, or whether homeland territory generated greater willingness than more distantly held colonial/imperial territories. Implicitly tested in such analyses was the notion that it was homeland territory per se, that was important: that the proximity of any homeland territory of one state to any homeland territory of another state was the important factor. Some of the Most and Starr diffusion analyses also indicated the strong impact of colonial territorial borders on the diffusion of war– that colonial territories were responsible for creating a greater number of opportunities for conflictual interaction. Starr and Most (1976) also distinguished between land-based contiguity and across-water proximity. Again, such a distinction implicitly dealt with possible variations in ease of interaction and salience.

As noted, this project was designed to establish a major reconceptualization and revision of how borders may be seen (theory) and measured (method). The use of GIS (Geographic Information Systems) permitted a much fuller and clearer specification of borders by enabling us to operationalize the specific qualities of borders in terms of opportunity and willingness: ease of interaction and salience, respectively. Using data available in the 16 data layers found in ARC/INFO's 1992 Digital Chart of the World, indexes of both ease of interaction and of salience have been constructed. Aggregating values generated from ARC/INFO, they can be used to characterize any border or border segment on the globe, and thus values can be attached to the ease of interaction and/or the importance of any particular border or border segment. These two dimensions can be used separately or combined. A border with high values on both could be considered a "Vital Border." This data set therefore permits investigators to go beyond simply observing the number of borders a state possesses, whether or not a border existed between two states, or the length of that border, to a fuller conception of the nature of that border.

For ease of interaction (or opportunity), three central factors for the movement of land-based military capability (or other goods) were selected-- out of the hundreds of variables found in ARC/INFO-- the existence of roads, railroads, and the steepness of terrain. An index was created which notes the presence or absence of roads and railroads, and represents the hypsography or slope of terrain. This creates a simple combined 1 to 4 index, with 4 representing the greatest ease of interaction, and 1 the most difficult areas to move across.

Salience (or willingness) is concerned with the importance or value of territory along or behind a border. Here we must be concerned with indicators which would discriminate the level of value or concern over territory. Drawing on upon the work of geographers, demographics are seen as important: the territory on which a state's population lives. This is operationalized by areas of population concentration. A capital city, the locus of governmental activity and the symbol of the state, is also be used to indicate the importance of territory. Other coverages provide the location of items that indicate the importance of an area: active civil and military airports are identified, as well as such items as: military camps, forts, oil wells and refineries, power plants of various kinds, water tanks, factories, industrial complexes, hospitals, telecommunications stations, etc. The wide variety of items taken from the GIS are used because the substantive importance of any single type of installation can vary considerably across states. By identifying the location of key aspects of a state's transportation, communication, energy production, industrial, agricultural, and security infrastructures, we have items that tap "importance" in a manner generally relevant to all states.

Hence, the salience of a border area is determined by places of population concentration, state capitals, airfields, and selected cultural features located within a 50,000 meter buffer of the region's borders. Capital cities are automatically coded with the highest value found in any of the units of analysis.

Each feature identified has been given a value based on the number of other features that fall within 4 kilometers of it. These can then be mapped based on the value, showing where clusters arise. Now any area in a buffer around a border can be characterized by a value from 1 to 4. A four value scale has been created, again with 4 indicating areas of the greatest salience, 1 indicating those areas with the least.

The core of the vital border concept is that any border or border segment may combine high or low values reflecting both opportunity and willingness. Again, scores of 4 indicate a high level of "vitalness" (with either a 4 or a 4 and 3 on both indexes), with 1 indicating the lowest level for the combined indexes.

The global dataset derived from the GIS analyses includes 151 states with land borders, which generate 301 separate contiguous land borders between states. For each of these borders, 17 variables have been developed, which can be transformed into a variety of nominal, ordinal, and interval measures. For any dyad border (see the examples in Table 1) we can present the length of that border in kilometers, and the area in square kilometers under the buffers created from that border. From these two variables we can present the percentage of each border that falls into categories 4 through 1. This can be done for ease of interaction, saliency, and vitalness. Knowing the length of the border (or arc), the area under the buffer along it, and the percentage of each category, permits the analyst to use interval data (as noted below) or broadly based categories such as high-salience or low-salience. Note also that Table 1 provides a weighted average for each border in terms of ease of interaction, salience, or vitalness, showing the average value across the whole border.

Table 2 provides descriptive data on the total set of global contiguous borders, using the weighted averages. Just looking at these data indicate the variety of analyses that can be pursued with this data set. For example, we see that the average salience is quite low, barely getting above 1.000 (with a maximum value of 1.369 on the 4.000 scale). This means that although we find many areas with a value of 4 along borders, they constitute only very small portions of the total border. The values for ease of interaction are much higher (thus, so are the values for vitalness). Note that the border with the highest salience score exists between Moldova and the Ukraine. In many ways this should not be surprising since less than two decades ago, this was only an internal border, or the equivalent of the border between Connecticut and Massachusetts. With a weighted average of 1.342, the German-Dutch border is the next highest. A cluster of relatively high salience borders are found among the original members of the EEC. And, because of a high density of road and rail facilities, EU dyads also have the highest weighted averages in terms of ease of interaction. The border with the highest weighted average of ease of interaction is between Belgium and France (3.296). The next three highest are: Belgium-Netherlands (3.291), Germany-Netherlands (3.287), and France-Luxembourg (3.284). That is-- the borders of the original core countries of the EU also have borders that look like the internal jurisdictional boundaries of states, in terms of both salience and ease of interaction.

It should also be noted that color maps can be generated for any border or border segment in the international system (see Starr 2002). The measurement procedures and the indexes created were specifically developed to generate four-category schemes This was purposely done in order to facilitate the translation of the color maps into black and white representations. While I have stressed that any section of any border can now be represented by a value from 1 to 4 for ease of interaction, salience, or vitalness– values that can be used in data analyses within the GIS, both with other GIS variables or any other data sets that are imported into the ARC/INFO GIS– . maps are also an important medium for the presentation of results. They are particularly important in demonstrating that these three measures can vary along any single border that a state might have with a contiguous neighbor (the values, as represented by different colors, vary). Thus, in some ways maps might be more useful than the summary measures in weighted border averages.

Starr and Thomas (2002: 219) state clearly how the nature of borders might be important to our study of conflict and cooperation:

One basic point raised in Most and Starr (1989) is that researchers need to be much clearer as to the broader concepts which are really under investigation, so that their models and the resulting research designs can be more logically and fully specified. Perhaps "borders" can be used in some research for reasons that are innate to "borderness"-- that they separate entities from one another. However... most uses of borders involve their representation of proximity-- that is, entities are close to one another, important to one another, and have an enhanced ability to interact with one another. But, does the existence of a border actually represent these notions? Borders that are difficult to traverse, either commercially or militarily may not fit this idea of proximity. Borders which are "buffered" by empty and meaningless spaces may not fit this idea of proximity. Conversely, legal borders in the contemporary world may be meaningless in terms of full permeability and high levels of transactions-- as in the European Union. The concept of a vital border-- with its two subcomponents-- specifies more completely and precisely how a border might represent "proximity" and allows us to investigate the meaning of a border, in both its traditional or transnational senses.

Starr and Thomas (2002) look at the analysis between proximity and the analysis of crisis, revisiting hypotheses investigated in Brecher and Wilkenfeld (1997). As with war/militarized conflict more generally, Brecher and Wilkenfeld’s analysis of geographical factors incorporates adversarial proximity as a leading determinant of crisis behavior. They test four hypotheses related to crisis behavior: (1) the greater the proximity of the crisis adversaries, the more likely it is that the crisis will be triggered by violence; (2) the greater the proximity of the crisis adversaries, the more likely it is that violence will be employed in crisis management; (3) the greater the proximity of the crisis adversaries, the more likely it is that the crisis will terminate in agreement; (4) the greater the proximity of the crisis adversaries, the more likely it is that the crisis will be part of a protracted conflict. The GIS- data set permitted a finer grained analysis of proximity. Brecher and Wilkenfeld found that “Contiguous” countries were more conflict prone than their “Near Neighbors” or “Distant” countries. With that basic effect demonstrated, Starr and Thomas investigated those contiguous states looking more deeply at the nature of the contiguous borders– and found that at some point greater ease of interaction and salience picked up not the conflict effects hypothesized, but interaction/interdependence/integration effects– in a manner proposed by Deutschian models of integration (2002: 229):

A dynamic that reflects the Deutschian model of integration– with its attendant focus on transactions– definitely appears to be affecting countries with borders at the upper ends of the measurement scales for ease of interaction, salience, and vitalness. This is an important finding not only because an interdependence/integration dynamic exists, but also because the presence of this dynamic statistically suppresses the conflict dynamic normally found in hypotheses relating to contiguity.

That is, greater and greater levels of interaction opportunities do not produce greater and greater conflict effects beyond the simple existence of contiguity; (Starr 2002 found similar results, with length of borders having a positive relationship with conflict).

Continuing several of these strands, Starr and Thomas (2005) again look at contiguity and conflict, and hypothesize that high levels of ease of interaction and salience– for countries that are contiguous, and therefore already reflect the basic opportunity for interaction that facilitates conflict– may better reflect interdependence/integration effects. Two different views on the relationship between contiguity and conflict from the literature were presented. The first is the “standard” view, based on territory both as an interaction opportunity and territory as being intrinsically valuable, that the easier a border is to cross and the more salient the border, then the higher the probability of militarized dispute. The second view, deriving from Deutschian integration theory based on high levels of transactions, proposed that the easier a border is to cross and the more salient the border, then the lower the probability of militarized dispute. Each view, however, represents a linear (positive or negative) relationship between the nature of borders and conflict. Starr and Thomas (2005) proposed a curvilinear relationship, with the low occurrence of conflict at both the lowest and highest levels of ease of interaction (opportunity) and salience (willingness). Starr and Thomas proposed– and found– that conflict is most likely where the expected utility of conflict is greatest– in the middle– where states have both the opportunity and willingness to engage in conflict. (See Appendix Table 1, which is a modified version on Table 1 in Starr and Thomas 2005, for a representation of the hypotheses.) Interestingly, in these analyses, the length of a border did not show a significant effect on conflict (with MID’s as the dependent variable).

In finding support for our hypothesis we have made two significant contributions to our understanding of the political geography of conflict and cooperation. We have examined and demonstrated the utility of moving beyond a simple “on-off” dichotomous view of contiguous land borders to examine the terrain and human activity along shared border areas... Secondly, we have demonstrated that the relationship between borders/contiguity and conflict behavior is non-linear. Neither contribution could have been made using previously collected data on contiguity which simply noted whether or not two states shared a land border. These results could only be possible with the type of data provided by the GIS project on the nature of borders.

Thus, we have seen that dropping to levels of greater specificity below a simple continuity (yes-no), can be of great utility in addressing some questions, while for others the use of simple contiguity suffices as an explanatory variable.

One area which utilized borders in their role as indicators of proximity and as opportunities for interaction involved the question of the positive spatial diffusion of such phenomena as violent conflict (e.g. Most and Starr 1980). As noted above, the concept of Warring Border Nation tapped the existence of a violent conflict as a “treatment” in the study of diffusion (conceived of a the growth of an ongoing war in Siverson and Starr 1991). One initial area of application for the creation of the GIS-generated data set was to revisit findings on diffusion with a more specified set of measurements of the nature of borders. A pilot study of one set of enduring rivals– Israel and its neighbors– indicated that the actual movement of military capabilities during militarized conflicts coincided with the border areas with the greatest ease of interaction, thus providing some confidence in the face validity of the measures (see Starr 2000). Indeed, the diffusion of war was to be the main focus of the present paper, addressing the question of whether or not “more porous” borders (those with higher ease of interaction) facilitated the diffusion/growth of interstate wars. However, the time frame in which the GIS data set could be used has precluded the broad time period needed to revisit that question. Using Correlates of War data for interstate wars, only the first Gulf War was found to have grown beyond the original combatants in the relevant time frame. In the U.S.-led coalition, only Syria and Saudi Arabia shared contiguous borders with the initial combatants (Iraq and Kuwait). This situation (as with Gulf War II) did not involve these “participants” in significant (if any) combat, which was pursued by major powers not neighbors. To anticipate comments in the conclusion, we must remember that in the identification and analysis of “relevant dyads” spatial proximity was only one factor. The other central factor included interactions that involved major powers! This factors clearly stands out in the cases of the two Gulf Wars.

The question of whether or not “more porous” borders (those with higher ease of interaction) facilitated the diffusion/growth of war could also be investigated looking at intrastate wars.

Additionally, we could hypothesize that salience, as measured by important elements of economic, military, social and demographic infrastructure, would also be associated with the diffusion or growth of conflict– with such territory serving (at least partially) as the stakes of the conflict. Using Correlates of War data (up to 2000), 10 intrastate wars were identified where additional state actors joined the initial government/internal challenger conflict. In four of those conflicts the states that joined the conflict had no contiguous borders with the state engaging in internal violent conflict. The COW data show the remaining six conflicts having 10 joiners, of which eight had contiguous borders with the country in conflict (see Appendix Table 2).

In order to test the null hypothesis that borders with a greater difficulty in interaction have either no effect on the likelihood that states will go to war with neighbors or make states more likely to go to war, we have to compare the nature of borders where conflict has occurred with those where conflict has not occurred. The same is true for the effect of the perceived importance (salience) of a border area, as well as the combined measure of Vitalness. One strategy would be to compare the nature of conflict borders with all other borders in the system. However, this strategy is flawed since it fails to account for differences in government and for differences in the propensity of individual states to enter into wars. Also, having run these analyses, we see that the summary statistics for all 301 contiguous borders show higher ease and salience than the conflict borders under consideration. Much of this is generated by the borders of west and northwest Europe, where the European Community was born and developed. So, instead of comparing borders of conflict joiners to all the world’s contiguous borders, a more conservative strategy was used that can take these differences into account. I tested for statistically significant differences between conflict dyad borders (the shared contiguous homeland border between a state involved in an intra-state conflict and a joining state) and the remaining borders of the state involved in the intra-state conflict. The summary statistics are shown in Table 3.

Using t-tests no statistically significant differences in means were found, comparing the weighted averages generated by the eight dyads– the borders between the eight countries that joined the intra-state conflicts and the states involved in the conflict– and the means for the 54 borders that the intra-state conflict hosts had with all of their neighbors. Wilcoxon Signed Rank Tests on the medians for the two groups were also used. But again, no statistically significant differences were found.

Very simply, a crude first cut at whether greater ease of interaction or greater salience are associated with the diffusion of intra-state conflict, provides no support for these hypotheses (as operationalized by the two measures generated by the GIS project). Perhaps more importantly, we see that within the limited time period surveyed of 20-30 years, violent conflict in the international system– whether inter-state or intra-state– tended not to diffuse or grow. The small number of case that were available for analysis was a striking finding (even taking into account differences between the COW lists and other lists of conflicts).

As with the proposition that more porous border might facilitate conflict diffusion, there are areas in the conflict literature that would support the view that more porous borders are associated with enduring rivalries (see Starr 2005). Whereas questions of diffusion look at spatial factors and the consequences of conflict for its recurrence across space, the study of rivalry looks at the consequences of conflict in terms of its recurrence across time for a given pair of states. Using the same methods noted above, the most recent list of rivalries produced by Diehl and Goertz was reviewed for any rivalry which lasted past 1980.

The mean and median figures for the weighted averages of ease of interaction and salience of the contiguous borders between the two countries in the rivalry were compared to the all the contiguous borders for the rivalry countries. 37 rivalry dyads were identified, which generated 114 contiguous neighbors. The summary statistics are presented in Table 4.

Again, using both t-tests and Wilcoxon Signed Rank tests, we find no support for the hypothesis that greater ease of interaction characterizes the borders between rivals. No statistically significant differences between means were found. Indeed, the mean and median ease of interaction between rivals were lower than the mean and median for all countries bordering the rivals– that is, rival borders were harder to cross; (note that the median for ease of interaction was statistically significantly lower than the median for all 301 contiguous borders). However, while the absolute differences are small, the median for the salience of the 37 rival borders is statistically significantly greater than that of the median for all rival borders (p=0.002). These borders do seem to be of greater salience. It is not clear exactly why. This statistical analysis needs to be augmented with analysis of the cases involved (as was done in Starr 2000). For example, we could hypothesize that the higher salience scores derive from a more developed military infrastructure along the rival borders, or because of urban areas along the border which were settled specifically in order to make border territorial claims more explicit or to demonstrate commitment to control over the border area (as was done by the Jewish West Bank settlers).

I have directed much of my research program toward demonstrating the importance of taking spatial factors into our accounts of international politics and the study of conflict (and cooperation). In turn, much of this work has focused on the geo-political arrangement and position of states as reflected by their borders, particularly contiguous homeland borders. The GIS-generated data set was developed in order to investigate questions involving proximity and borders which might require knowing more about the specific “nature” of specific borders (or any parts thereof). Analyses presented in past work as well as the current paper have indicated both the value of this data set as well as limitations in its content, coverage, and impact.

The spatial effects of contiguity and the nature of borders have been revealed in the study of crises, MIDs, and other forms of conflict. The general interaction opportunity argument captures the overall effects that borders tend to have in increasing the probability of conflictual interactions. The two papers by Starr and Thomas have highlighted a new effect– given contiguity, and the conflict promoting effects of contiguous borders– as ease of interaction and salience become greater then the more positive effects hypothesized in Deutschian social communication/transaction- based integration theory can be seen. In studies of conflict, our analyses have not supported the proposition that ever greater levels of ease of interaction (i.e. “more porous” borders) generate greater conflict (in some linear manner). Clearly, for many of our questions, the on-off variable of presence/absence of contiguous borders is enough.

The latter point has been seen in the present paper: first-cut analyses (or in the words of Most and Starr 1989, the use of “stylized facts”) of diffusion or rival borders are not affected by greater levels of ease of interaction. The finding that rival borders have higher salience is suggestive that the measures of salience and importance need to be investigated more closely, and that case methods are needed to supplement the use of this data set.

While past studies have had large enough numbers of cases for analysis, all applications of the GIS-generated data set require articulation of its limits– in the time period covered and the inability to study time series with variation in the independent variables (ease of interaction and salience) across time. In addition, the numerical data using the weighted averages for specific borders (or “arcs” in GIS terminology), may also be problematic. The weighted averages pick up the percentages of any border area which are “high” or “low”– i.e. 1 through 4– on ease of interaction or salience for the whole border. But the maps that can be generated from the GIS (see Starr 2002) more clearly indicate the number, the length, and the specific locations of these different areas. Our theories are not good enough to let us specify how much of a border needs to in the category indicating the greatest (or least) ease of interaction to affect some dependent variable. As shown in Starr (2000), some questions require map inspection and case study, while others can be addressed with the weighted average data.

A second major issue about the data set, noted in footnote 4, is that it incorporates only those variables which can be seen and labeled on maps (mostly generated from satellite imaging). To fully discuss questions such as diffusion or rivalry we need to map the spatial distribution of such factors as ethnic groups, subterranean mineral resources (or even water tables), and other issues such as the historical importance or symbolism of certain border areas. One of the strengths of GIS analytic systems, however, is that a GIS is designed to import other data sets to be incorporated into its spatial analyses. As the use of GIS becomes more widespread, such data sets need to be developed and put into forms compatible for importing into a GIS system.

These issues point to future directions for the application of this data set. From the beginning I have noted that these measures could be used to help indicate cooperative as well as conflictual interactions. A number of questions about areas undergoing increasing interdependence and integration can be addressed by these data– including studies of the effects of the growth of the European Union, or the impact of changes in legal boundaries between states (such as the Schengen Agreements which have become part of the EU legal framework through a protocol attached to the Treaty of Amsterdam).

As in Starr (2002), the present results indicate that we should accept the null hypotheses that more difficult to cross borders and less important border areas do not lower the probability of conflict– here diffusion or rivalry. Goertz and Diehl (1992), Holsti (1991), Huth (1996), for example, focus on territory per se as a cause of war; as both the issue over which war breaks out, and as a factor which increases the stakes of a war. As noted, such analyses provide us with a very important alternative hypothesis: it is territory– any territory– which creates an opportunity for conflict. The GIS-based data set has permitted analyses to test the range of applicability of this alternative.

Brecher, M., and J. Wilkenfeld (1997) A Study of Crisis. Ann Arbor: University of Michigan Press.

Goertz, G., and Diehl, P.F. (1992) Territorial Changes and International Conflict. London: Routledge.

Holsti, K.J. (1991) Peace and War: Armed Conflicts and International Order 1648-1989. Cambridge: Cambridge University Press.

Huth, P.K. (1996) Standing Your Ground: Territorial Disputes and International Conflict. Ann Arbor: University of Michigan Press.

Klein, J.P., G. Goertz, and P. F. Diehl (2006) The New Rivalry Dataset: Procedures and Patterns. Journal of Peace Research 43:331-348.

Lemke, D. (1995) The Tyranny of Distance: Redefining Relevant Dyads. International Interactions 17:113-126.

Lemke, D. (1996) "Why So Much Peace?: Interstate War in Africa." Paper presented at the Annual Meeting of the American Political Science Association, San Francisco, August.

Midlarsky, M. I.(1970) Mathematical Models of Instability and a Theory of Diffusion. International Studies Quarterly 14:60-84.

Most, B. A.,and H. Starr (1980) Diffusion, Reinforcement, Geopolitics, and the Spread of War. American Political Science Review 74:932-946.

Most, B. A.,and H. Starr (1989) Inquiry, Logic and International Politics. Columbia, SC: University of South Carolina Press.

Siverson R. M., and H. Starr (1991) The Diffusion of War. Ann Arbor: University of Michigan Press.

Starr, H. (2000) Using Geographic Information Systems to Revisit Enduring Rivalries: The Case of Israel. Geopolitics 5:37-56.2

Starr, H. (2001) Reconceptualizing International Borders Through the Application of GIS (Geographic Information Systems): A New Dataset on the ‘Nature’ of Borders. International Studies Perspectives 2: 441-43.

Starr, H. (2002) Opportunity, Willingness and Geographic Information Systems: Reconceptualizing Borders in International Relations. Political Geography 21:243-261.

Starr, H. (2005) Territory, Proximity, and Spatiality: The Geography of International Conflict. International Studies Review 7:387-406

Starr, H., and B.A. Most (1976) The Substance and Study of Borders in International Relations Research. International Studies Quarterly 20:581-620.

Starr, H. and G. D. Thomas (2002) The ‘Nature’ of Contiguous Borders: Ease of Interaction, Salience, and the Analysis of Crisis. International Interactions 28:213-235.

Starr, H. And G. D. Thomas (2005) The Nature of Borders and International Conflict: Revisiting Hypotheses on Territory. International Studies Quarterly 49:123-139.

Wesley, J. P. (1962) Frequency of Wars and Geographical Opportunity. Journal of Conflict Resolution 6:387-389.

Zipf, G.K.(1949) Human Behavior and the Principle of Least Effort. Cambridge, MA: Addison-Wesley.

Summary Statistics, Neighbors of Intrastate Conflicts
	Ease of Interaction	Salience	Vitalness
Median	2.88	.011	1.943
Mean (s.d.)	2.58124 .4815834)	1.03937 (.0626112)	1.79617 (.2781512)

Summary Statistics, Rivals (37 dyads)
	Ease of Interaction	Salience	Vitalness
Median	2.641	1.014	1.837
Mean (s.d.)	2.50159 (.4725045)	1.0333 (.0562775)	1.74981 (.2659244)

Summary Statistics, Rivals (37 dyads)
	Ease of Interaction	Salience	Vitalness
Median	2.641	1.014	1.837
Mean (s.d.)	2.50159 (.4725045)	1.0333 (.0562775)	1.74981 (.2659244)

Summary Statistics, Neighbors of Rivals
	Ease of Interaction	Salience	Vitalness
Median	2.703	1.0075	1.8455
Mean (s.d.)	2.522439 (.5141293)	1.025746 (.043323)	1.761684 (.2877284)

APPENDIX TABLE 1: Hypotheses on the Probability of Conflict Based on the Nature of Borders^*

	Border Ease of Interaction, Salience, and “Vitalness”
	Low	Medium	High
“Standard”	CONFLICT	CONFLICT	CONFLICT
Deutsch	CONFLICT	CONFLICT	CONFLICT
Starr and Thomas	CONFLICT	CONFLICT	CONFLICT
^*The increased likelihood of conflict is visually depicted by an increased font size.