Defining and Coding Negotiations

I define negotiation as direct or mediated communication between active belligerents with the ostensible aim of reaching a mutually acceptable agreement. As mentioned in chapter 1, this definition is largely based on that of Fred Iklé and his well-known book How Nations Negotiate.⁵

I further unpack and operationalize this general definition on three dimensions. First, communications must involve an exchange and/or assessment of offers that are related to the settlement of hostilities. I make no distinction regarding whether these communications produce tangible agreement or whether the parties are working in good faith. The inclusion of the word ostensible in the definition reflects this decision. Indeed, a key element of my argument is that one benefit to negotiations is exploiting them for insincere ends unrelated—if not antithetical—to ending conflict. According to this criterion, shuttle diplomacy by a third party, such as Secretary of State Alexander Haig’s travels between the United Kingdom and Argentina during the Falklands War of 1982, is a form of negotiation. Discussions hosted by a third party that involve the belligerent states, such as the Organization of African Unity’s emergency summit regarding the war over Angola in 1975, are also negotiations. On the other hand, mediation attempts that fail to get the consent or participation of the belligerents do not qualify.

Second, communication must take place between officially appointed representatives of at least one belligerent nation from each side who, as outlined by the previous point, make and/or assess bargaining offers. Such activity can be either publicly known or performed in secret. Formal exchanges that take place at a summit or conference, whether direct or mediated, clearly satisfy this requirement. The unsuccessful London Conference of 1864, which was organized to address the Second Schleswig-Holstein War between Prussia and Austria on one side and Denmark on the other, is one example. Moreover, informal negotiations may also qualify, conditional on explicit offers being exchanged by appropriate representatives. For example, during the Russo-Finnish War of 1939–40, the first round of communication took place via notes secretly exchanged between Alexandra Kollontai (the Soviet Union’s ambassador to Sweden) and Hella Wuolijoki (a feminist leftist playwright and outspoken Finnish Communist) at the Grand Hotel in Stockholm. A passing conversation that takes place in the hallways or corners of the United Nations, on the other hand, does not reach the necessary threshold.

Third, and related to the question of timing, short recesses between talks are still considered days with communication as long as parties expressly agreed to reconvene at a specific later date. The Korean War presents a useful illustration. Between July 27 and October 8, 1952, the United Nations Command and Communist delegations transitioned from meeting on a daily basis to convening on a weekly basis and ended each meeting by agreeing to continue their discussion over how to repatriate prisoners of war the following week. It would be misguided to consider each weekly meeting to be a new round of discussion. I consider this entire span of time to be an uninterrupted phase of communication.

Using these criteria, I create a binary negotiation variable that takes the value 1 on war-days that have negotiations and 0 on days that do not. Each sequence of uninterrupted 1s in a war can be considered a negotiation period. In sum, the ninety-two interstate conflicts feature 189 negotiation periods that occur over 6,332 war-days. This represents 17 percent of the total 36,834 war-days in the overall dataset.

Third-Party Initiatives

One of the challenges to assessing my theory using quantitative data is that we cannot adduce belligerents’ true intentions when they choose to negotiate. We may safely say that any talks that lead to the termination of conflict are likely sincere in nature. Approximately one-quarter of negotiations fall into this category. But this is a determination we can make only in retrospect. In what other manner can we plausibly ascertain whether talks are more likely to be sincere or insincere before we are able to observe their ultimate outcome?

My theory suggests that negotiations that are borne of significant third-party pressure are more likely to be insincere, as they take place in conditions that lessen the costs of negotiating and increase actors’ liberties to talk with no intention of settling. These acts of diplomacy should therefore have different consequences for war compared to those that are initiated by the belligerents themselves at high costs. If a period of negotiation occurs in large part due to a third party’s initiative, we can expect these talks to have a lower likelihood of producing a diplomatic settlement.

For each of the 189 negotiation periods in my data, I ascertain whether talks were the result of third-party pressures or initiatives. I call talks sought by the belligerents themselves internal negotiations, while I call talks primarily borne of outside diplomatic pressure external negotiations. The unrelenting mediation efforts during the Falklands War, discussed in the introduction, are a clear example of external negotiations. That said, it is important to note that external negotiations are not synonymous with mediation.⁶ President Theodore Roosevelt’s mediation efforts in the Russo-Japanese War of 1904–5—touched on in the introduction—serves as an instructive case. The peace conference held in Portsmouth between August and September 1905, which successfully ended the conflict, led to widespread praise of Roosevelt and culminated in his receiving the Nobel Peace Prize in 1906. But these talks transpired only because Japanese officials quietly asked Roosevelt to offer his services back in April 1905.⁷ Roosevelt never proactively attempted to press either side to negotiate. In fact, he never went to Portsmouth to attend talks.⁸ This conference is therefore an internal negotiation.

Although my data include this measure of whether negotiations were the product of third-party pressure, they do not contain information regarding which specific belligerent made the first actual request to initiate negotiations. Historical sources make it very difficult to ascertain the responsible actor in a consistent manner. Nevertheless, my theory is not dependent on identifying which specific belligerent initiated negotiations. I instead focus on whether talks took place at all and, in some cases, whether they were external or internal in nature.

Examples

Figure 2.1 summarizes negotiations in several conflicts that span the temporal scope of my negotiation data. Briefly describing the key diplomatic events of each war will help elucidate the overarching coding process.

The Austro-Sardinian War began on March 24, 1848, when Charles Albert, King of Sardinia, declared war against Austria. After some initial victories, Sardinia suffered a nearly constant string of defeats. The first and only negotiation period to take place spanned from March 26 to March 30, 1849, when Victor Emmanuel—who took power in Sardinia on March 23 after his father went into exile—requested and struck an armistice with Austria. This internal negotiation ended hostilities.

The Third Central American War started on May 27, 1906, between Guatemala on one side and El Salvador and Honduras on the other. The United States almost immediately urged all belligerent states to consider peace. Mexico later joined the United States in this diplomatic press. After receiving constant offers to mediate, the three warring parties accepted the joint United States–Mexico overture. Talks began on July 19 and brought the conflict to an end the next day. The negotiation period is coded as being external. Note that this classification is made based on the fact that the talks were prompted by third parties, regardless of the fact that the meeting did lead to conflict termination.

The Korean War is the subject of deeper scrutiny in chapter 6. Hostilities began on June 25, 1950, and continued violently for a year. Negotiations to establish an armistice in the Korean War began on July 10, 1951, between the United Nations Command (UNC) and the Communist delegation of North Korea and China. Talks were halted by the Communists on August 23, 1951, over accusations that the UNC permitted bombings in the neutral zone around the negotiation site. The second negotiation period started on October 25 of the same year and lasted until October 8, 1952—the day that the UNC suspended talks indefinitely due to seemingly irreconcilable differences about how to deal with prisoners of war (POWs). Between July 27 and October 8, the delegations transitioned from near-daily meetings to weekly meetings. Harking back to my third criterion, I count this whole span of time as part of the same negotiation period, as each meeting concluded with an agreement to speak again the following week. The third and final negotiation period began on April 26, 1953, and ended the conflict on July 27 after both sides overcame their impasse regarding POWs. Notably, I do not count a brief period of discussion in February 1953 when the delegations worked out a deal to exchange some sick and wounded POWs. Even though the talks did involve POWs, they did not involve an exchange or assessment of offers related to the original POW issue that precluded a negotiated settlement. While a variety of states encouraged both sides to settle their differences through diplomacy, the negotiations were not directly spearheaded, orchestrated, or led by an external party. All three periods of negotiation are therefore coded as being internal.

Last, the Ugandan-Tanzanian War presents a case where (external) negotiations took place but hostilities ended through a military victory. Fighting broke out on October 28, 1978, as Ugandan forces attempted to annex several hundred square miles of Tanzanian territory. The first negotiation period began on November 5 when a delegation led by Philip Obang (Sudanese ambassador to the Organization of African Unity, or OAU) and Peter Onu (assistant secretary general to the OAU) arrived in Kampala to speak with Ugandan president Idi Amin, who agreed to a withdrawal plan. The effort fell flat on November 11, when the Tanzanian government rejected the OAU’s proposal.⁹ A second negotiation period started on February 21, 1979, when the OAU convened an ad hoc Mediation Committee prior to its Council of Ministers conference. Representatives from Uganda and Tanzania both participated in this meeting but reached no agreement.¹⁰ The OAU declared this effort a failure on March 2, 1979. Both negotiation periods are clear cases of external diplomacy where outside actors took the initiative to force open channels between the two warring states. By April 11, Tanzanian forces took the city of Kampala and deposed Idi Amin, ending the war.

Trends and Patterns in Negotiations

Below, in table 2.1, I report top-line numbers on the frequency of negotiations across all war-days in my data. As mentioned in chapter 1, 17 percent of war-days feature negotiations. Approximately two-thirds of the time spent talking involves internal negotiations orchestrated by the warring parties, while the remaining one-third is a product of external negotiations pressed on belligerents by third parties.

Table 2.2 disaggregates the 189 negotiation periods by whether they are internal or external. To reiterate, whether talks are driven by outside pressures does not perfectly predict the settlement of hostilities. Even so, the table also shows a distinct pattern where negotiations sought directly by belligerents are more than twice as likely to terminate conflicts as those that are externally imposed on the belligerents. Diplomacy that takes place at the behest of the belligerents themselves thus tends to indicate a more sincere interest and ability to reach a negotiated settlement.

Recall that figure 0.1 in the introduction uses my new data to display the occurrence of all forms of negotiation across the universe of wars included in my study. Figure 2.2 uses kernel regression smoothers to further disaggregate and explore negotiations over the course of war. First, the black line reflects the average propensity for negotiations across time. We see that approximately 10 percent of wars feature negotiations almost immediately after hostilities begin. The rate of negotiations gradually increases over the course of war, reaching a local maximum around the time the conflict reaches what will become its halfway point. The sudden upswing in negotiations at the right-hand side of the figure indicates that a healthy share of wars come to an end based on negotiations that immediately preceded the cessation of hostilities.

Note: Proportions in parentheses.

Note: Row-wise proportions in parentheses.

The gray and dashed lines on Figure 2.2 reveal a striking breakdown in terms of which forms of negotiation contribute to the overall trend captured by the black line. Through the first half of the figure, negotiations that take place are mostly under duress and spearheaded by third parties. At some point in the middle of an average conflict, external talks are supplanted by internal ones. It is highly telling that the vast majority of negotiations associated with the settlement of hostilities are internal. This pattern is consistent with my theory’s expectation that when talks are initiated by the belligerents themselves, at least one side believes that reaching an agreement is preferable to the costs of continued conflict or the potential benefits of any side effects from insincerely motivated discussions.

It is worth emphasizing that my core argument speaks to the existence of latent third-party pressures. The distinction I have made between internal and external negotiations is one way of measuring the manifestation of these pressures. We should expect more external negotiations to take place in environments that feature higher levels of latent third-party pressure. I will return to and substantiate this point in chapter 3.

Defining and Justifying Battles

While the notion of a “battle” feels familiar, it is not necessarily straightforward to define. For my purposes, I synthesize extant definitions and consider a battle to be a clash at a specific time and location between organized state-level forces over a contested strategic objective.¹³ A contested strategic objective can be a hill, fort, city, or any other unit that the belligerents each hope to claim or defend, that can exchange hands, and that affects each side’s ability to continue to prosecute the war.¹⁴

No battle-based data can capture all war activity that fits these criteria. Work by Alex Weisiger—one of very few past endeavors to quantify intrawar hostilities—cites this fact to justify the use of casualties instead of battles to measure battlefield activity.¹⁵ There are nevertheless several reasons why casualties are not a direct or preferable substitute for battles. First, as already mentioned, battles are an undeniably important unit of analysis and a standard lens through which leaders often view military activities. Differences between reference materials concern depth of coverage and not fundamental debates about what constitutes a battle. Second, many battles also do not have reliable or accessible measures of troop strength on each side. For example, historians’ casualty estimates for the battle of Tuyutí during the War of the Triple Alliance (or Paraguayan War) range from seven thousand to sixteen thousand. Estimates for the battle of Manila in the final months of World War II span an enormous range from one hundred thousand to five hundred thousand.¹⁶ Third, contemporary theories of war often focus on the exchange of “forts,” or military objectives, over the course of fighting.¹⁷ Battles are a much closer approximation to the concept of a “fort” than casualties and thus allow for a more proper test of these ideas. Fourth, casualties often represent a concerted decision by leaders to gain or defend strategic objectives that they feel are vital to bargaining. States may be entirely willing and ready to suffer immense costs if they believe those potential losses to be worthwhile.¹⁸ Casualties are a less convincing proxy for exogenous information than the observed gains and losses of military objectives.

None of this suggests that casualty data are unimportant or not worth further research. My battle data simply provide an alternative and complementary path to understanding wartime activity in a manner that plausibly reflects the information in leaders’ minds as they navigate wars.

Coding Battles

I thus create a dataset of battles in interstate wars. The foundations of this dataset are built on an encyclopedia of battles created by Tony Jaques.¹⁹ I supplement and verify the initial list of battles from this volume using several additional reference guides from other military history scholars.

Each reference book contains brief narratives about individual battles or a longer description of an entire conflict that contains mentions of particular clashes. All of these narratives are qualitative and must therefore be converted into quantitative measures. Given the complexity of battles, I lean on the side of caution and create measures only on battle characteristics that can be consistently coded for every battle using the raw qualitative data. The process that I describe below produces a final dataset of 1,708 discrete battles that took place during interstate wars listed in the COW Inter-State War dataset.

Examples

Two battles may prove enlightening in demonstrating how the qualitative accounts of battles are converted into quantitative measures. First, I consider the battle of Plevna from the Second Russo-Turkish War of 1877–78. Russian forces led by Grand Duke Nicholas faced little resistance as they swept down toward the Danube River. Approximately 30,000 Ottoman forces took defensive positions in the town of Plevna to fend off the approaching Russians, who numbered between 90,000 and 110,000. Between September 10 and 12, 1877, the Russians engaged in a major assault to overtake Plevna but failed at the cost of approximately 20,000 casualties.²⁵

Briefly stated, Russian forces took the offensive against a Turkish stronghold but were repulsed, suffering a humiliating loss that received enormous international attention. In terms of the individual battle at Plevna, Russia was the attacker and the Ottoman Empire was the defender. These designations do not factor into the score given to the battle but help define the criteria used to determine which belligerent won the battle. At the level of the entire war, Russia was the initiator and the Ottoman Empire was the target. Since the battle resulted in a victory by the war target, the assault on Plevna receives a score of –1.

Second, consider the battle of Ouadi Doum during the Libyan-Chad War of 1986–87, also known as the War over the Aouzou Strip or the Toyota War. On March 19, 1987, Libyan forces stationed at a major air base at Ouadi Doum headed toward the city of Fada with the intention of reclaiming it from Chadian forces, who had gained control in a battle on January 2. The decision proved immensely costly for Libya. Chad’s army trounced the Libyans, diverting them back to Ouadi Doum and eventually taking the air base for themselves. On March 22, Chad declared victory and announced that its forces had killed upward of 1,200 Libyan soldiers and captured approximately 450. Libya’s loss of the Ouadi Doum air base marked the end of Libya’s hold on northern Chad.²⁶

Libya started the battle by sending its forces against Chad but faced a terrible defeat. Libya was thus the battle attacker, while Chad was the battle defender. More broadly, Chad was the initiator of the overall conflict, while Libya was the target. The battle at Ouadi Doum was therefore a victory for the war initiator and is assigned a score of +1.

Table 2.3 summarizes the distribution of battles based on actors and outcomes for all 1,708 events. The data indicate that actors who start individual battles tend to have a significant advantage, winning approximately two-thirds of all clashes.

Two caveats merit mention here. First, as table 2.3 intimates, my data treat all battles as being equivalent in magnitude. On one hand, this is a limitation of data; current information on all of these individual battles is not sophisticated enough to create a reliable system that weighs events differently according to forces deployed, casualties suffered, or the like. On the other hand, this property of my battle data parallels how many well-known formalizations of war treat “forts,” the object of individual battles, as identical units.²⁷ My analysis therefore reflects a relatively direct assessment of the learning dynamic featured in these models.

Second, wars vary substantially in the number of battles recorded. Perhaps most prominently, the two world wars account for approximately 47 percent of my data. Much of this variation likely captures real differences in the intensity and frequency of conflict, yet I cannot eliminate the possibility that uneven coverage also plays a role. For example, the Saudi-Yemeni War of 1934, which lasted two months, is reflected by only one battle. English-language documentation for the conflict is relatively limited and piecemeal.²⁸ The battle data for the Vietnam War contain only fifty entries over a dozen years of hostilities—an apparently low number that may reflect the difficulties of tracking and quantizing guerrilla warfare.

Regardless of whether every single conceivable battle is captured, the more relevant question is whether the current data present a faithful portrayal of the broader trends and changes on the battlefield. The figures presented below suggest that my data do satisfy this more elemental condition.

Note: Proportions in parentheses.

Measures of Battlefield Activity

I use these raw data regarding 1,708 battles to create daily-level measures of battlefield outcomes. For each day t in war w,

Several of my hypotheses scrutinize the effect of clear battlefield information on negotiation. There are multiple plausible paths we could take to measure the concept of “clear information” based on these daily sums. Previous research suggests that distinct battlefield trends supply the most overt information regarding the future trajectory of conflict and also create conditions for a self-enforcing agreement. For those reasons, some studies find that distinct battlefield trends are predictive of war termination and the resilience of postconflict peace.²⁹ I will therefore generate and rely on measures designed to track the outcomes of recent fighting.

My first measure, which I call momentum, calculates the sum of all battles that have ended in the previous d days of the conflict (or all days of the conflict—whichever is closer). I use d = 60 as my default window of time. This variable represents recent battlefield trends and reverts to 0 if no new battles end in the next d days. Formally,

Note, however, that my hypotheses are agnostic as to which particular belligerent wields an advantage or disadvantage while fighting. As such, much of my analysis of battlefield trends relies on a measure called recent imbalance, which is simply the absolute value of momentum.

Although we have reasons to focus on recent battle trends, some may justifiably believe that the entirety of hostilities influence decision-making and that events in the more distant past do not fade in importance.³⁰ I account for the accumulation of all battle outcomes using a measure called position, which is a running total of all daily sums for the entire war up to the given war-day:

I then take the absolute value of this measure to reflect the overall imbalance observed from the battlefield:

Regardless of the specific trends on the battlefield, we would also want to gauge the overall intensity of hostilities. To do so, I count the number of active battles taking place on each day of war.

Figure 2.3 illustrates these measures for World War I.³¹ The data align well with our qualitative understanding of the conflict. The Central powers initiated the war and are thus represented using positive values, while the Allied powers’ victories are captured with negative values. Early German offensives found success in invading Luxembourg, Belgium, and parts of France. By September 1914, however, British and French troops were able to blunt German momentum at the first battle of the Marne. This is reflected by the small shift from positive to negative values in both position and momentum in the opening months of the war. The somewhat middling values observed in the first several months of the conflict reflect the stalled nature of trench warfare, which led to a back-and-forth exchange of victories and losses on the western front.

The Central powers began accumulating significant victories in April 1915. Many of these occurred on the Gallipoli peninsula, where Allied forces arrived in late April hoping to seize control of the Turkish straits from the Ottomans. The offensive was a disaster for the Allies,³² leading to a withdrawal in late 1915. Much of early 1916 featured victories by the Ottomans against Allied forces in Mesopotamia and numerous failed Russian offensives against the Germans, both of which are reflected by significant upward movement in position and momentum.

The war entered a significant shift in the middle of 1916. The battle of the Somme in the summer, despite its enormous costs and somewhat inconclusive outcome, was a turning point in favor of the Allies.³³ Figure 2.3 exhibits a downward shift around this time. The German loss at the battle of Verdun later in the year severely blunted German initiative. The entry of the United States on the side of the Allies in April 1917 only worsened Germany’s situation.

Seeing the window of opportunity for victory close, the Germans launched several last-ditch offensives in early 1918. While these saw some partial success, these efforts did not make any significant breakthroughs. The second battle of the Marne in 1918, which would be Germany’s last major offensive, was countered by the Allies and was the start of an implacable Allied advance. The sudden and aggressive changes in battlefield trends starting in the middle of 1918 capture this shift and continue until the war ends in favor of the Allies on November 11, 1918.