Industry-Level Supply-Side Market Concentration and the Price of Military Conflict

Abstract

In economics, supply-side market concentration profoundly impacts firm behavior. This dimension of economic interaction can be used to predict the conflict initiation of countries in the context of international relations. The following investigation uses industry-level trade data to define four new market concentration variables, which are incorporated into the traditional model of military conflict. The article finds that high dyadic market concentration significantly decreases the likelihood that a state initiated military conflict in the period 1962–2001, and argues that market concentration is an important factor in the trade–conflict relationship.

Keywords

conflict interdependence market concentration supply-side trade

I. Introduction

The effect of trade on military conflict has been fiercely debated, though current work predominantly suggests that interdependence has a pacifying effect on interstate relationships (Russett and Oneal, 2001; Oneal et al., 2003; Hegre et al., 2010; Lu and Thies, 2010). Nonetheless, support for commercial liberalism is not unanimous, and some scholars (Barbieri, 1996; Beck, 2003; Keshk et al., 2004) contend that trade has an ambiguous or even inflammatory effect. The existence of such prominent dissent suggests our current understanding of the relationship between trade and conflict requires further revision.

One particular omission in the current literature is the influence of industry-level, supply-side market concentration on the opportunity cost of conflict. Though trade leads to welfare gains in both competitive and non-competitive markets (Mankiw, 2008), markets with concentrated production, and thus inelastic supply curves, distribute gains from trade differently than competitive markets do. In markets with inelastic supply, producers receive a greater portion of the trade surplus, though perhaps at a loss in efficiency to society as a whole. Thus, the opportunity cost of conflict is higher for major exporters in concentrated markets, and the domestic firms, individuals, and government bodies that share these producer surpluses will create political pressure to avoid conflict.

The addition of supply-side market concentration into the current model of commercial liberalism follows from a sequence of theoretical developments, beginning with the foundational idea that trade increases the cost of conflict (Angell, 1913), and the subsequent empirical expected utility model of military conflict (Polachek, 1980; Gasiorowski and Polachek, 1982; Gasiorowski, 1986; Polachek et al., 1999). Since the expected costs of conflict due to trade depend on the difficulty of replacing a lost trade flow (Hirschman, [1945] 1980; Keohane and Nye, 1989; Wagner, 1988), other authors have isolated factors that decrease states’ ability to substitute trade, including the elasticity of demand and supply (Hirschman, [1945] 1980; Reuveny, 2003), the specificity of the trade flow (Yarbrough and Yarbrough, 1992; Williamson, 1996), and the nature of the traded commodity (Dorussen, 2006).

However, a persistent theoretical complication in many of these expected utility arguments is the difficulty of distinguishing the differing impact of trade on importers and exporters. Scholars have found that gains from trade are greater when the elasticity of demand is smaller (Polachek and McDonald, 1992), using bilateral trade elasticity in their empirical argument, but there is little work on the impact of inelastic supply due to producer market concentration. Thus, one of the central contributions of this article is to consider the effect of elasticity on exporters in isolation, using a disaggregated commodity-specific model based on previous approaches to using disaggregated trade data (Dorussen, 2006).

There are also two types of channels through which supply-side market concentration can deter conflict: monadic and dyadic. Here, a monadic effect is independent of partner-specific factors and implies that, ceteris paribus, a state will initiate fewer conflicts merely for having large shares in concentrated markets for widely-traded resources. This effect follows directly from the economic implication that markets with inelastic supply generate a relatively large amount of producer surplus. Conversely, the dyadic effect of supply-side market concentration on military conflict builds on the monadic effect by incorporating the value of dyadic trade. Because trade conditions between two states can alter the expected effect of monadic supply-side market power, this dyadic channel is a necessary addition. A hypothetical situation illustrates this point: if State A (the potential initiator) is a major exporter in several commodities whose suppliers are concentrated, it should be reluctant to attack other states and forgo its large producer surplus. However, if State B (the potential target) imports none of the commodities of which State A is a major producer, State A receives no trade benefits from State B and thus risks no trade gains in attacking. Thus, dyad-specific conditions can alter or diminish the impact of market power on conflict.

To summarize, including supply-side market concentration in the standard model of military conflict yields several testable hypotheses about the relationship between trade and peace:

Hypothesis 1: Conflict is less likely between states that trade more with one another.

Hypothesis 2: States with large market shares in concentrated markets for widely-traded resources will initiate fewer military conflicts.

Hypothesis 3: A state with a large market share in a concentrated market for widely-traded resources that exports extensively to a partner state will be less likely to initiate a military conflict against that partner state.

II. Data and Measurement

My investigation is closely modeled after the baseline specification of Oneal and Tir (2006); the major point of divergence is the addition of four market concentration variables composed of commodity-level trade data. I use the directed-dyad year as the unit of measurement in order to distinguish between initiators and targets of military disputes. As in Oneal and Tir (2006), this model includes control variables from both the realist and liberal camps. Realist variables address factors that affect the opportunity and willingness to engage in military conflict including contiguity, distance, great powers, capabilities ratios, and alliances. On the other hand, traditionally liberal variables are measures of democracy and interdependence.

The variable DIRCONT, which codes for direct contiguity, equals 1 when two states share a land border and 0 when this is not the case. It also accounts for indirect contiguity through colonies or dependencies. Close geographic proximity is critical in increasing the likelihood of conflict, as it improves the ability to project military force and increases the number of issues that may prompt states to fight.

Another measure of geographic proximity, LNDISTANCE, is a continuous variable that measures the natural log of the shortest distance between the two states’ capitals, or in the case of a large state, the closest port city.

MAJPOWER takes a value of 1 if either member in the dyad is a major power and takes a value of 0 if this is not the case. The Correlates of War project identified major powers in this time period by referring to a consensus among historians (Singer and Small, 1995). This variable controls for states that have the ability to project power all over the world, thus increasing their ability and motivation to engage in military conflict.

LNCAPRAT is derived from an index coded by the COW project (Singer and Small, 1995) to measure the balance of power in a dyad based on urban population, total population, energy consumption, iron and steel production, military manpower, and military expenditures; the index weights all of these factors equally. The variable is the natural log of the ratio of the stronger state’s index to the weaker state’s index (Singer et al., 1972). A second capabilities variable takes the square of this ratio to account for previous research that suggests dispute initiation most often occurs close to parity (Bennett and Stam, 2004; Hegre, 2004; Oneal, 2006).

The variable ALLIES equals 1 if the countries in a dyad are both members of a mutual defense treaty, neutrality pact, or entente. If not, it takes a value of 0.

The democracy data come from Polity IV and use the traditional 21-point scale (Jaggers and Gurr, 1995), which ranges in value from −10 (very autocratic) to 10 (very democratic). The variable DEMOC_ST is the democracy score of the initiator state, and DEMOC_TGT is the democracy score of the target state.

This regression also includes a measure of trade interdependence, which is coded as the variable DEPEND. To account for the fact that trade must be economically important to be politically important, I follow Oneal and Tir (2006) in their specification by dividing the sum of a state’s exports and imports with its partner by its GDP. Their data, as well as mine, are drawn from Gleditsch (2002), who used information from the International Monetary Fund (various years) and outside sources for non-IMF members. For the post-1962 period, this dataset is quite complete.

\begin{matrix} {Dependence}_{i t} = (\frac{{Exports}_{i j t} + {Imports}_{i j t}}{{GDP}_{i t}}) \\ i - state j - target; t - year \end{matrix}

Additionally, for pairs of minor, noncontiguous powers the variable SYSTSIZE equals the negative natural logarithm of the number of states in the international system. For all other dyads, the international system size variable takes a value of 0.

Finally, this analysis follows Oneal and Tir (2006) in taking the recommendation of Beck et al. (1998) that regressions control for duration dependence. This regression contains a peace years variable that codes for the number of years elapsed since the last military dispute as well as a natural cubic spline with three interior knots. These controls account for some time-variant endogeneity in the explanatory variables.

Measures of Market Concentration: Monadic Supply-Side Concentration

The new market concentration variables rely on commodity-level trade data from the United Nations Comtrade database (United Nations, various years). To maximize the available date range, I use the first revision of the SITC classification, which begins in 1962 and continues to the present.

The SITC classification allows for several levels of disaggregation, represented by a code that may be up to five digits long. Each successive digit signifies an additional level of specificity. The level of aggregation most appropriate for this analysis involved two-digit commodity codes, as there is a tradeoff between capturing substitute goods in one category and separating truly independent markets. At the highest levels of disaggregation, commodities are narrowly defined, and two separate commodities could plausibly be substitutes for one another. For example, commodity 65121 “Yarn carded sheeps lambs wool, not for retail” and commodity 65122 “Yarn combed sheeps lambs wool, not for retail” are nearly identical. Such over-disaggregation would overstate the effect of market concentration on conflict by diminishing the impact of any given market. On the other hand, the one-digit level of aggregation defines commodities too broadly and would lump unrelated goods into one category. Analysis at the two-digit level of disaggregation strikes the proper balance between under- and over-disaggregation.

The market concentration index is calculated using only market share in widely-traded commodities. This restriction ensures that the markets in question will have a bearing on every state in the international community: market concentration in any of these commodities will to some extent affect all international players. To put this restriction into practice, the new market concentration variable is based only on the top 25 commodities traded in each year. The commodities are ranked by the total dollar value exchanged in international markets each year, so the top 25 commodities may differ between years. The decision to use precisely 25 commodities is not arbitrary. At the second level of disaggregation, the UN Comtrade database has 62 distinct commodities. Using the top 25 commodities strikes a balance between targeting widely-traded commodities and maintaining reasonable external validity.¹ For illustration, the top 25 commodities for three years (1962, 1980, and 2001) are presented in Table 1.

Table 1.

Commodities Ranked by Total Value Exchanged SITC Revision 1; 1962, 1980, and 2001

Rank	1962	1980	2001
1	Machinery, other than electric	Petroleum and petroleum products	Machinery, other than electric
2	Transport equipment	Machinery, other than electric	Electrical machinery, apparatus, and appliances
3	Petroleum and petroleum products	Transport equipment	Transport equipment
4	Textile yarn, fabrics	Electrical machinery, apparatus, and appliances	Petroleum and petroleum products
5	Iron and steel	Iron and steel	Miscellaneous manufactured articles
6	Electrical machinery, apparatus and appliances	Chemical elements and compounds	Special transactions
7	Cereals and cereal preparations	Textile yarn, fabrics	Scientific instruments
8	Non-ferrous metals	Miscellaneous manufactured articles	Clothing
9	Fruit and vegetables	Non-ferrous metals	Chemical elements and compounds
10	Miscellaneous manufactured articles	Cereals and cereal preparations	Medicinal and pharmaceutical products
11	Textile fibers and waste	Non-metallic mineral manufactures	Textile yarn, fabrics
12	Coffee, tea, cocoa, spices	Manufactures of metal	Iron and steel
13	Metal ores and metal scrap	Scientific instruments	Non-metallic mineral manufactures
14	Manufactures of metal	Clothing	Plastic materials
15	Chemical elements and compounds	Special transactions	Manufactures of metal
16	Paper, paperboard	Plastic materials	Non-ferrous metals
17	Scientific instruments	Paper, paperboard	Paper, paperboard
18	Non-metallic mineral manufactures	Metal ores and metal scrap	Gas, natural and manufactured
19	Crude rubber	Gas, natural and manufactured	Fruit and vegetables
20	Meat and meat preparations	Fruit and vegetables	Chemical materials
21	Wood, lumber, and cork	Meat and meat preparations	Furniture
22	Clothing	Coffee, tea, cocoa, spices	Cereals and cereal preparations
23	Chemical materials	Wood, lumber, and cork	Perfume, toiletries, cleaners
24	Coal, coke, and briquettes	Textile fibers and waste	Fish and fish preparations
25	Dairy products and eggs	Chemical materials	Meat and meat preparations

Since this model is based on supply-side market concentration, the data are collected by exporter rather than importer. The variable specification employs the basic rationale behind the Herfindahl-Hirschman Index (HHI), which is used to measure the size of firms relative to an industry (Hirschman, 1964). For each year, the dollar value of exports for each supplier for each widely-traded commodity is divided by the total dollar value of exports in that commodity. That ratio is then squared. Since squaring the ratio of state commodity exports to total commodity exports causes each marginal increase in market share to quadratically increase market concentration, this measure favorably weights large players. This emphasis corresponds to the theoretical foundation of the market concentration hypothesis, since markets with monopolies and oligopolies should have especially high costs of substitution and producer surpluses.

\begin{matrix} {HHI}_{i j t} = {(\frac{{Exports}_{i j t}}{{TotalExports}_{j t}})}^{2} \\ i - state; j - commodity; t - year \end{matrix}

The commodity and state specific HHI is then summed across all commodities for each state, resulting in a value that represents the overall market share of a specific state across all of its exports. After the summation there exists one HHI value for each state-year.

\begin{matrix} \sum {HHI}_{i j t} = \sum_{j} {(\frac{{Exports}_{i j t}}{{TotalExports}_{j t}})}^{2} \\ i - state; j - commodity; t - year \end{matrix}

To incorporate the new market variable into the traditional regressions used in Oneal and Tir (2006), I merge the HHI market concentration data with the directed-dyad conflict data so that each observation has two new right-hand variables: HHI_INITIATOR for the initiator state and HHI_TARGET for the target state.

Measures of Market Concentration: Dyadic Dependence

Beyond the monadic HHI index, I also introduce an interaction term that includes dyad-specific trade. This dyadic market concentration variable captures the global supply-side market share of each state as well as the impact of relevant trade between each state and its specific partner.

The specification of the dyadic HHI variable is as follows. The first part is the monadic measure of supply-side market power discussed in the previous section, which captures the state’s global market power in widely-traded resources. The second portion of this interaction term is the total value of dyad-specific trade in the top 25 widely-traded commodities each year divided by the real GDP of the state’s trading partner. Therefore, this variable will have a different value for each partner within each dyad-year, one for the initiator and one for the target.

\begin{matrix} {DyadicHHI}_{i p t} = \sum_{j} {HHI}_{i t} {(\frac{{TotalTrade}_{i p t}}{{RealGDP}_{p t}})}^{2} \\ i - state; p - partner; t - year \end{matrix}

Theoretical reasons led to this particular specification. It is clear that including some measure of dyadic trade is necessary, but I specifically choose to include trade in the top 25 commodities per year in order to isolate trade that is directly affected by this model’s measure of supply-side market concentration. Since the monadic HHI measure only accounts for market share in the top 25 widely-traded commodities in each year, it can only predict changes in conflict opportunity cost due to trade in those markets. Therefore, it makes sense to include the value of trade in only the annual top 25 most-traded commodities. In the denominator, the dyadic HHI variable contains the real GDP of the partner state to account for the inverse relationship between the deterrent effect and the partner’s economic size.

Measures of Military Conflict

Data for the dependent variable come from the Military Interstate Dispute (MID) dataset (Jones et al., 1996) as coded in the Correlates of War (COW) project and corrected by Maoz for the pre-1992 period (Maoz, 1999). MID takes a value of 1 if a militarized interstate dispute occurs in that dyad-year and a value of 0 otherwise. Within the 672,492 observations in this dataset, there are 1,094 documented military interstate disputes. While some of the conflict literature uses fatal MIDs rather than all MIDs as the outcome variable of analysis, using all MIDs is consistent with my theory’s implication that the opportunity cost of trade with high market share should inhibit even threats or low-level use of force.

III. Results

This article provides one central finding regarding the pacifying relationship between supply-side market concentration and military conflict. In dyads with high levels of exchange in widely-traded commodities, increasing the supply-side market share of a state decreases the likelihood that it will initiate military conflict. I use three primary models for my analysis: the traditional baseline model of international conflict, a model that includes a measure of monadic initiator market concentration, and a model that includes both monadic and dyadic measures of initiator market concentration. Coefficients from the three models in this article are presented in Table 2, starting with the baseline model on the left and moving right with incrementally more market power variables.

Table 2.

The Effect of Supply-Side Market Concentration on Military Conflicts All MIDs; 1962–2001

	Model 1: Baseline Model	Model 2: Monadic Market Power Model	Model 3: Monadic and Dyadic Market Power Model
Initiator Democracy Score	−0.0344** (.0074)	−0.0330** (.0075)	−0.0304** (.0074)
Target Democracy Score	0.0040 (.0077)	0.0037 (.0075)	0.0037 (.0074)
Economic Dependence	−1.619 (2.225)	−1.470 (2.169)	−0.8348 (1.857)
Ln(Capability Ratio)	0.4176** (.1598)	0.4438** (.1633	0.4728** (.1668)
Ln(Capability Ratio)²	−0.0208** (.0076)	−0.0218** (.0077)	−0.0233** (.0079)
Direct Contiguity	1.955** (.3046)	1.959** (.3012)	2.026** (.2987)
Ln(Distance)	−0.3132** (.0709)	−0.3053** (.0705)	−0.3147** (.0689)
Major Power	1.228** (.2485)	1.271** (.2516)	1.388** (.2634)
Initiator HHI Index	–	−0.0838** (.0352)	−0.0690** (.0355)
Dyadic Initiator HHI Index	–	–	−1.3x10^−6** (5x10^−7)
Pseudo R²	0.2185	0.2196	0.2211
χ² (df)	2248.61** (14)	2275.68** (16)	2342.78** (18)
N	672,492	672,492	672,492

These regressions are modeled on the baseline regression in Oneal and Tir (2006). The unit of analysis is the directed dyad-year and the results are for logistic regressions. Standard deviations are noted in parentheses. Statistical significance is noted with asterisks: * indicates p < .05 and ** indicates p < .01. For variables that take a different sign than predicted, † indicates p < .05 and †† indicates p < .01. All tests are one-tailed tests. The coefficients of all three peace-years splines, system size, and alliance are significant at the p < .001 level; they have been omitted here for brevity.

Model 1, my baseline model, is a replication of Oneal and Tir’s (2006) baseline model and employs precisely the same explanatory variables. Model 2 builds on this foundation by including the monadic measures of market concentration into the logistic regression. Model 3 further expands Model 1 by including both the monadic and the dyadic measures of market concentration.

All three models in Table 2 generally concur with the findings of Oneal and Tir (2006). In each regression, an increase in the democracy score of the initiator, squared capability ratio, and natural log of distance between the states all significantly decrease the incidence of military conflict. Furthermore, an increasing capability ratio, the existence of direct contiguity, and the participation of a major power all significantly increase the chance of military conflict in a dyad. Finally, while democracies are less likely to initiate conflict, the effect of the target state’s democracy score is consistently insignificant. The high level of agreement between these models and the findings of Oneal and Tir (2006) provides confidence in the coefficients of the new variables.

However, there is one major difference between these results and those of Oneal and Tir (2006): economic dependence does not have a significant effect on the incidence of all MIDs. Economic dependence fares even worse with the introduction of the market concentration variables. These statistical results strongly indicate that this particular measure of economic dependence does little to explain the incidence of all MIDs, a surprising result. While it is possible that multicollinearity or the restrictions on my dataset caused this unusual outcome, I show in the online appendix that the deviation in the impact of economic dependence likely arises from my usage of all MIDs rather than fatal MIDs as the dependent variable of analysis.

On the whole, the three models in Table 2 represent a good corroboration of existing conflict literature. Moreover, Models 2 and 3 offer some statistically compelling insights into the impact of supply-side market concentration on military conflict.

First, increasing monadic supply-side market concentration decreases a state’s likelihood of initiating a militarized conflict, an outcome that supports Hypothesis 2. In Model 2, the coefficient on the initiator’s HHI index is significant (p < .01) and negative, indicating an inverse relationship between supply-side market concentration and the number of military conflicts a state initiates. This effect is also robust to the addition of dyadic HHI market concentration variables, as the coefficient on the initiator’s HHI index remains negative and significant (p < .01) in Model 3.²

The second implication of my analysis is that increasing a state’s dyadic supply-side market concentration deters it from initiating military conflict. The coefficient on the initiator’s dyadic HHI index in Model 3 of Table 2 is negative and significant at the p < .01 level. This conclusion is robust to extensive statistical tests, some of which are presented within this article and some of which are discussed in the online appendix. Ultimately, the negative and highly significant coefficient on initiator dyadic market power strongly supports Hypothesis 3 and provides persuasive evidence that supply-side market power is an important force in the incidence of military conflict.

Between the three models the R-squared values increase very little. However, the R-squared of a regression is a poor tool for determining whether to include one or several variables in a model (Wooldridge, 2009). Instead, the most important factor in establishing the theoretical importance of a variable is the value and significance of its coefficient. By this measure, these conclusions are theoretically and statistically sound.

IV. Robustness Checks

I also run a number of robustness checks, the first of which tests whether my four new market concentration variables contribute predictive power to the current model of international conflict. Specifically, I test the joint null hypothesis that the coefficients on all four market concentration variables are equal to zero, and resoundingly reject it (p < 0.0008). This result clearly indicates the importance of including supply-side market concentration in the current structural model of interstate conflict.

One issue with adding supply-side market concentration to the current model of conflict is that its multiple measures of trade may result in multicollinearity. To check for this problem, I present a basic correlation matrix in Table 3 of all of the trade indices: economic dependence, monadic initiator HHI, monadic target HHI, dyadic initiator HHI, and dyadic target HHI. This check loosely suggests that there are no multicollinearity problems, as the highest level of correlation between any of the trade variables is 0.1495.

Table 3.

Robustness Check Tests for Multicollinearity Correlation Matrix

	Econ. Dependence	Initiator HHI Index	Target HHI Index	Dyad. Initiator HHI	Dyadic Target HHI
Econ. Dependence	1.00	–	–	–	–
Initiator HHI Index	0.002	1.00	–	–	–
Target HHI Index	0.040	0.1495	1.00	–	–
Dyad. Initiator HHI Index	0.086	0.056	0.045	1.00	–
Dyadic Target HHI Index	0.136	0.057	0.057	0.491	1.00

Another robustness check replicates the three initial regressions without the economic dependence variable; the results are reported in Table 4. Across each of the three models, the value and significance of the coefficients remain consistent with the original results in Table 2, providing further evidence that multicollinearity between the trade variables was not an issue.

Table 4.

Robustness Check Specification without Economic Dependence All MIDs, 1962–2001

	Model 1: Baseline Model	Model 2: Monadic Market Power Model	Model 3: Monadic and Dyadic Market Power Model
Initiator Democracy Score	−0.0349** (.0074)	−0.0334** (.0075)	−0.0306** (.0074)
Target Democracy Score	0.0032 (.0077)	0.0029 (.0075)	0.0032 (.0074)
Economic Dependence	–	–	–
Ln(Capability Ratio)	0.4317** (.1615)	0.4568** (.1646)	0.4808** (.1678)
Ln(Capability Ratio)²	−0.0213** (.0076)	−0.0223** (.0077)	−0.0236** (.0079)
Direct Contiguity	1.956** (.3049)	1.960** (.3015)	2.027** (.2985)
Ln(Distance)	−0.3090** (.0710)	−0.3015** (.0706)	−0.3130** (.0689)
Major Power	1.230** (.2483)	1.273** (.2514)	1.391** (.2627)
Initiator HHI Index	–	−0.0846** (.0351)	−0.0825** (.0217)
Dyadic Initiator HHI Index	–	–	−1.3x10^−6** (5x10^−7)
Pseudo R²	0.2184	0.2195	0.2205
χ² (df)	2235.33** (13)	2263.75** (15)	2332.13** (17)
N	672,492	672,492	672,492

I also investigate the performance of dyadic HHI variables without monadic HHI variables. Table 5 presents two versions of this test. Model 1 contains all the baseline variables as well as the two dyadic HHI indexes, while Model 2 drops economic dependence from the regression. The coefficients from both of these regressions strongly support the initial findings in Table 2, retaining the same signs and significance as their earlier counterparts. Importantly, for both models in Table 5, the coefficient on the initiator’s dyadic HHI index is positive and significant, and the coefficient on the target’s dyadic HHI index is negative and significant, just as in Table 2. This result suggests that the effect of dyadic market concentration on military conflict is robust to the exclusion of monadic concentration and economic dependence.

Table 5.

Robustness Check Specification without Monadic HHI Indices All MIDs, 1962–2001

	Model 1: Dyadic Market Power Model	Model 2: Dyadic Market Power Model, No Economic Dependence
Initiator Democracy Score	−0.0313** (.0074)	−0.0315** (.0074)
Target Democracy Score	0.0039 (.0076)	0.0035 (.0076)
Economic Dependence	−0.8196 (1.854)	–
Ln(Capability Ratio)	0.4578** (.1654)	0.4658** (.1663)
Ln(Capability Ratio)²	−0.0228** (.0079)	−0.0230** (.0079)
Direct Contiguity	2.041** (.3017)	2.042** (.3015)
Ln(Distance)	−0.3215** (.0691)	−0.3198** (.0692)
Major Power	1.379** (.2649)	1.382** (.2641)
Dyadic Initiator HHI Index	−1.5x10^−6** (6x10^−7)	−1.5x10^−6** (5x10^−7)
Pseudo R²	0.2204	0.2203
χ² (df)	2327.52** (16)	2315.63** (15)
N	672,492	672,492

In the online appendix, I also re-specify all the market concentration variables using the top 10, 20, and 30 commodities from each year rather than the top 25. My central result that the dyadic market concentration of a state decreases the likelihood that it will start a conflict is highly robust to all of these tests. The appendix also includes a brief investigation into the coefficient on economic dependence. I find that my result likely differs from the result of Oneal and Tir (2006) because my models use all MIDs rather than fatal MIDs as the dependent variable.

V. Conclusion

Understanding the impact of economic interdependence on military conflict is of great strategic, scientific, and humanitarian value. Many scholars have argued that interdependence is generally pacifying, but debate continues both within and without the liberal school, demonstrating the need for further work.

My investigation provides a critical addition to the theory of liberal peace by introducing supply-side market concentration as a variable explaining military conflict. Supply-side market concentration influences the cost of conflict by increasing the producer surpluses for powerful suppliers. Very powerful suppliers enjoy higher surpluses from their trade, which translate into more foregone value if military conflict erupts. This additional cost consequently deters conflict initiation.

The analysis of four new variables based on commodity-level trade data for the period 1963–2001 tests whether market power in widely-traded commodities actually creates incentives for peace. Two of the new variables measure monadic market concentration, which varies at the state-year level, and the other two measure dyadic market concentration, which varies with the dyad-year. Within each level of analysis there is one variable for the initiator and one variable for the target. The initiator indexes are included as controls, and the target indexes are the core of this article’s argument.

My central finding is that increasing a state’s dyadic market concentration significantly reduces the likelihood of initiating a military conflict. This outcome holds across a number of robustness checks, including dropping economic dependence, omitting the monadic HHI variables, and changing the number of commodities considered in specifying the market concentration variables.

In sum, supply-side market concentration is an important addition to the structural model of peace through trade. A state with high dyadic supply-side market concentration will be less likely to initiate military conflicts, and this effect may be explained using a basic opportunity cost model. Robust and significant statistical results demonstrate the importance of pursuing further research on the structural argument of peace through trade, and especially the value of using economic principles to inform models of interstate behavior.

Footnotes

*

Special thanks to Bruce Russett, Michael Tomz, John Oneal, Jaroslav Tir, Jessica Weeks, Nathaniel Levine, and the entire MIDs team at Stanford University. An online database with all data, do-files, and an appendix with additional discussion and robustness checks can be found at .

1

Nonetheless, the appendix includes robustness checks using varying numbers of the most widely-traded commodities in each year. The core results are robust to using each year’s top 10, 20, and 30 commodities.

2

Later robustness checks suggest this particular result is weaker than the conclusion about dyadic HHI.

References

Angell

Norman

. 1913. The Great Illusion. New York: Putnam.

Barbieri

Katherine

. 1996. Economic interdependence: A path to peace or a source of interstate conflict? Journal of Peace Research 33(1): 29–49.

Beck

Nathaniel

. 2003. Modeling dynamics in the study of conflict: A comment on Oneal and Russett. In Globalization and Armed Conflict, eds Schneider

Gerald

Barbieri

Katherine

Gleditsch

Nils Petter

, pp. 165–177. Lanham, MD: Rowman & Littlefield.

Beck

Nathaniel

Katz

Jonathan N.

Tucker

Richard

. 1998. Beyond ordinary logic: Taking time seriously in binary time-series–cross-section models. American Journal of Political Science 42: 1260–1288.

Bennett

D. Scott

Stam

Allan

III . 2004. The Behavioral Origins of War. Ann Arbor, MI: University of Michigan Press.

Brookes

James

. 1988. South Africa apparently shifting loyalty to support Mozambique. New York Times. 3 May: A1.

Dorussen

Han

. 2006. Heterogeneous trade interests and conflict: What you trade matters. Journal of Conflict Resolution 50(1): 87–107.

Gasiorowski

Mark

. 1986. Economic interdependence and international conflict: Some cross-national evidence. International Studies Quarterly 30(1): 23–38.

Gasiorowski

Mark

Polachek

Solomon W.

1982. Conflict and interdependence: East–West trade and linkage in the era of détente. Journal of Conflict Resolution 26(4): 709–729.

10.

Gleditsch

Kristian Skrede

. 2002. Expanded Trade and GDP Data, version 1. Journal of Conflict Resolution 46(5): 712–724.

11.

Hegre

Håvard

. 2004. Size asymmetry, trade, and militarized conflict. Journal of Conflict Resolution 48(3): 403–429.

12.

Hegre

Håvard

Oneal

John R.

Russett

Bruce M.

2010. Trade does promote peace: New simultaneous estimates of the reciprocal effects of trade and conflict. Journal of Peace Research 47(6): 763–774.

13.

Hirschman

Albert O.

[1945] 1980. National Power and the Structure of Foreign Trade. Reprint, Berkeley, CA: University of California Press.

14.

Hirschman

Albert O.

1964. The paternity of an index. American Economic Review 54(5): 761–762.

15.

International Monetary Fund. Various years. IMF World Economic Outlook Database. Washington DC: International Monetary Fund.

16.

Keohane

Robert

Nye

Joseph

. 1989. Power and Interdependence. 2nd edn. Glenview, IL: Scott Foresman.

17.

Jaggers

Keith

Gurr

Ted Robert

. 1995. Transitions to democracy: Tracking democracy’s third wave with the Polity III data. Journal of Peace Research 32(4): 469–482.

18.

Jones

Daniel

Bremer

Stuart

Singer

J. David

. 1996. Militarized interstate disputes, 1816–1992: Rationale, coding rules, and empirical patterns. Conflict Management and Peace Science 15(2): 163–213.

19.

Keshk

Omar M.

Pollins

Brian M.

Rafael

Reuveny

. 2004. Trade still follows the flag: The primacy of politics in a simultaneous model of interdependence and armed conflict. Journal of Politics 66(4): 1155–1179.

20.

Lingyu

Thies

Cameron G.

2010. Trade interdependence and the issues at stake in the onset of militarized conflict. Conflict Management and Peace Science 27(4): 347–368.

21.

Mankiw

Gregory

. 2008. Principles of Microeconomics. Mason, OH: Cengage Learning.

22.

Maoz

Zeev

. 1999. Dyadic Militarized Interstate Disputes (DYMID1.1) DatasetFVersion 1.1. Codebook, Department of Political Science, Tel Aviv University.

23.

Oneal

John R.

2006. Confirming the liberal peace with analyses of directed dyads 1885–2001. In Approaches, Levels, and Methods of Analysis in International Politics: Crossing Boundaries, ed. Starr

Harvey

, pp. 73–94. New York: Palgrave Macmillan.

24.

Oneal

John R.

Tir

Jaroslav

. 2006. Does the diversionary use of force threaten the democratic peace? Assessing the effect of economic growth on interstate conflict, 1921-2001. International Studies Quarterly 50(4): 755–779.

25.

Oneal

John R.

Russett

Bruce

Berbaum

Michael L.

2003. Causes of peace: Democracy, interdependence, and international organizations, 1885–1992. International Studies Quarterly 47(3): 371–393.

26.

Polachek

Solomon W.

1980. Conflict and trade. Journal of Conflict Resolution 24(1): 57–78.

27.

Polachek

Solomon W.

McDonald

Judith A.

1992. Strategic trade and the incentive for cooperation. In Disarmament, Economic Conversion, and Management of Peace, eds Chatterji

Manas

Forcey

Linda R.

, pp. 273–284. Westport, CT: Praeger.

28.

Polachek

Solomon W.

Chang

Yuan-Ching

Robst

John

. 1999. Liberalism and interdependence: Extending the trade–conflict model. Journal of Peace Research 36(4): 405–422.

29.

Reuveny

Rafael

. 2003. The trade and conflict debate: Exploring the frontier. In Globalization and Armed Conflict, eds Schneider

Gerald

Barbieri

Katherine

Gleditsch

Nils Petter

, pp. 189–204. Lanham, MD: Rowman & Littlefield.

30.

Russett

Bruce

Oneal

John R.

2001. Triangulating Peace: Democracy, Interdependence, and International Organization. New York: Norton.

31.

Singer

J. David

Small

Melvin

. 1995. National Military Capabilities Data. Ann Arbor, MI: University of Michigan, Correlates of War Project.

32.

Singer

J. David

Bremer

Stuart

Stuckey

John

. 1972. Capability distribution, uncertainty, and major power war, 1820–1965. In Peace, War, and Numbers, ed. Russett

Bruce M.

, pp. 19–48. Beverly Hills, CA: Sage.

33.

United Nations. Various years. United Nations Commodity Trade Database. New York: United Nations.

34.

Wagner

R. Harrison

. 1988. Economic interdependence, bargaining power, and political influence. International Organization 42(3): 461–483.

35.

Williamson

Oliver

. 1996. The Mechanisms of Governance. New York: Oxford University Press.

36.

Wooldridge

Jeffery

. 2009. Introductory Econometrics: A Modern Approach. Mason, OH: South-Western College Publishing.

37.

Yarbrough

Beth V.

Yarbrough

Robert M.

1992. Cooperation and Governance in International Trade: The Strategic Organizational Approach. Princeton, NJ: Princeton University Press.