Lethality and deterrence in affairs of honor: The case of the Antebellum U.S. South

Dueling was common among Antebellum Southern gentlemen

Duels were commonly employed by Antebellum Southern U.S. gentlemen. In addition to the 56 senators mentioned in the introduction, at least 36 governors, ⁴ 57 U.S. congressman, and seven cabinet secretaries participated in duels, with almost all coming from the South or the pre-1804 North. Records of duels involving prominent politicians are particularly likely to persist across the years, so Southern gentlemen who never served in high office are surely underrepresented in any census of duelers. In our data set of two prominent newspapers (New York Times and Richmond Daily Dispatch) during the period of 1861–1865, prominent citizens participate in over half of all accounts of dueling.

Dueling grounds outside of cities such as Washington DC (Bladensburg), New Orleans (The Dueling Oaks), and Vicksburg TN (Dueling Island) acquired national reputations (Stevens, 1940). For example, a nineteenth century newspaper account claims “between 1834 and 1844 scarcely a day passed without duels being fought at the Oaks” (Times-Democrat, 13 March 1892). ⁵

In contrast, almost no duels took place north of New York City (Stevens, 1940), and very few took place in the rest of the North after the Burr-Hamilton duel, which spurred a crusade against dueling throughout most of the Northern states, leading to the practice’s virtual elimination north of Washington DC. A Massachusetts anti-dueling law and a related push by John Adams to ban dueling in the Continental Army are representative of contemporary attitudes towards dueling in New England.

While dueling was imported to the U.S. by European settlers, it persisted in the American South until shortly after the Civil War. Figure 1 depicts the number of senators we have recorded as participating in a duel, by the decade in which the duel took place. All but one such duel took place prior to 1860. Using the metric of senators participating in duels, the decline of the institution appears quite sudden indeed, with at least five duels taking place in each decade between 1800 and 1860, but none in the decade of the Civil War. The best explanation for this decline is institutional change. Dueling acted as an alternate legal system in the South where the formal legal system was unsophisticated and better suited for an agrarian society than an industrialized one. In the aftermath of the Civil War and Reconstruction, however, the South began to industrialize. Ranney (2002) writes that “Southern states had to decide whether to shape their legal systems to follow suit or to preserve their rural, agricultural pre-war character.” Their legal systems did evolve, rendering the duel archaic and ineffective. Posner (1996) describes dueling before the war as preventing “disputes from exploding into feuds by formalizing and channeling the means of enforcement.” After the war, however, the “superior efficiency of police and courts” rendered dueling inefficient as an alternate legal system. This is consistent with the evidence discussed in the introduction on the institutional shocks that Reconstruction brought to the South: the end of slavery, the death of 13.1% of white Confederate males between ages 10 and 44, ⁶ military and carpetbagger state governments, a federal government newly empowered to protect civil rights throughout the country, and a greatly expanded federal judiciary.OPEN IN VIEWER

Duels arose out of public contests

From 1816-1818, two Transylvania University medical school professors, Dr. Drake and Dr. Dudley, battled for influence over how the department should be run and methods of instruction. As the dispute became more acrimonious, Dr. Dudley charged that Drake “had attempted to destroy the medical school at Transylvania University.” The vitriol increased “with occasional outbreaks in pamphlet,” until August 1818 when a duel erupted (Coleman, 1953).

Henry Clay and Humphrey Marshall, while both serving in the Kentucky General Assembly in 1807, differed as to the propriety of an embargo on British-made products during an undeclared naval war with Great Britain. Arguments for and against the policy soon shifted to personal insults (Clay was a “demagogue” and “liar”), and a duel followed. Clay’s 1826 duel with John Randolph grew out of a dispute regarding a potential U.S. mission to Latin America (Holland, 2003).

Sitting congressmen George Washington Campbell and Barent Gardenier dueled in 1808 following a public dispute over the British embargo. Numerous duels arose from the Yazoo land deal, in which Georgia politicians attempted to sell seized Creek Indian lands at low prices to a company in which many of them held stock. Virginia Senator Armistead Mason’s fatal duel with John McCarty grew out of an amendment Mason introduced to a Senate bill that would allow Quakers and other conscientious objectors to avoid being drafted into military service by providing a substitute. ⁷ Holland (2003) describes physicians resorting to New Orleans’ Dueling Oaks to settle “differences in opinions on diagnosis and treatment,” for example Drs. Chopin and Foster dueling to settle a dispute about how to best treat a stabbing victim. Newspaper editors, controlling an outlet for political discussion, were frequently subject to challenge (Holland, 2003). Topics which spurred duels included slavery, whether or not to send aid to the Donner party, and tax policy (Holland, 2003).

According to Stevens (1940), “Men shot each other for gambling debts, for a dispute over billiards, an uncomplimentary word in an editorial, a jest at a table, a refusal to take a glass of whiskey, or, most of all, for disagreements in politics.” We take “disagreements in politics” to include any public conflict where gentlemen compete for the esteem of a third party. Of the dueling grounds outside of Washington DC, Holland (2003) said that “Most of Bladensburg’s visitors were more political […] Unlike the impulsive Old World quarrels over card games and jostling, political duels tended to be long-festering [disputes] that would suddenly flare to the fighting point over a nothing.” An 1859 Harper’s Weekly article stated that “There are parts of the United States where a politician must necessarily be prepared to fight duels… in many states of the Union, a politician who will not fight […] cannot command the popular suffrage.” ⁸ In the non-random sample of duels listed in Schwartz et al. (1984), of the 23 duels whose underlying cause is easily inferable, 15 clearly arise from public conflicts such as those described above. In our data set of newspaper articles, just over one third of duels with known reasons arose out of political disputes. The other reasons were military disputes (roughly 15%), personal insults (roughly 11%), and financial matters (roughly 6%).

Of course, some duels were fought over quite trivial matters that had nothing to do with any public conflict. The day following a drunken dispute “as to which understood some of the dead languages the best,” future U.S. Senator John Rowan shot and killed his friend Dr. James Chambers in a duel. Our reading is that the minority of duels arising from purely personal disagreements likely co-opted the institution of dueling to legitimize violence that would otherwise be thought unseemly.

Dueling was relatively safe

American duelers used purpose-built dueling pistols, designed for elegance and not accuracy, to settle disputes. These pistols were flintlock, short-barreled, smooth bore, and unsighted (as opposed to percussion cap, long-barreled, rifled, and sighted). The flintlock weapons misfired often, wasting shots and exhausting the dueler’s turn. Holland (2003) states that wearing glasses while firing was frowned upon.

One 1836 writer estimated that one in six duelers were injured, and one in 14 killed (Schwartz et al., 1984). Data collected by Kelly (1995) imply a 5.6% fatality rate in eighteenth century Irish duels when dueling pistols were used. ⁹ Another estimate puts the conditional probability of a naval officer dying on the field of honor at 20% (Stevens, 1940, pg. 71), while Byron (2008) compiles a selected sample of duels written up in 31 newspapers, finding a 23.9% fatality rate. McAleer (1990) finds that German dueling using pistols, had the highest mortality rate in Europe, around 20%. In our database of dueling senators, 41 received fire in a duel, three of whom died. As it is more likely that we failed to find politicians who dueled and lived than those who died, the mortality rate among senators was probably below 3 41 . In our newspaper data set, duels are deadlier. Just over one-quarter of duels were reported to have ended in the death of at least one participant. It is likely that this discrepancy arises due to our selected sample: newspapers are much more likely to report on duels with violent outcomes.

Deadlier weapons were available. In particular, swords were commonly used in European duels prior to the nineteenth century. The data in Kelly (1995) imply a fatality rate from eighteenth century Irish duels with swords of 39.4%. ¹⁰ Holland (2003) states that “perhaps 10 thousand” French gentlemen perished in sword duels between 1591 and 1610. However, Kiernan (1988) notes that even before the advent of dueling pistols, weapon choices kept mortality artificially low. Indeed, a 1777 Irish dueling code specified that were swords used, the duel should continue until “one is well bloodied, disabled, or disarmed” or until one party was both bleeding and willing to apologize (Holland, 2003, at 44). In contrast, pistol duels had an obvious and frequently non-lethal stopping point: after each side had fired one shot.

Percussion cap pistols were developed around 1830, while rifling was invented hundreds of years earlier. Holland (2003) describes “alarming results” when rifles, shotguns, or Bowie knives were employed in lieu of dueling pistols. In the preface to an 1878 edition of a treatise on Southern dueling first published in 1848, Barnwell Rhett “deplores the fearful number of street fights, which, he says, are attended with fifty times the mortality and without the moral effect and social amelioration of the code of honor” (Schwartz, 1940 at 135). Chernow (2017) describes an 1875 federal investigation which found 2141 murders of black Louisianans in the preceding 10 years. Domestic terrorists in other Southern states produced similarly grim results, often employing firearms designed for accuracy and not elegance.

The unpredictable behavior of dueling pistols rendered skill relatively unimportant and subjected both participants to similar risks regardless of experience. In contrast, skill was an important determinant of outcomes in sword duels. Holland (2003) says of such duels “the owner’s skill mattered more than the blade’s tempering, and fencing lessons were an essential part of a gentlemen’s education…. Fencing lessons could save your life.”

Finally, Leeson (2012) finds a higher mortality from a different nontraditional institution designed to resolve disputes. Ordeals in Nagyvarad, Hungary conducted between 1208 and 1235 condemned the defendant in 37.5%, or 78/208 cases, while English ordeals between 1194 and 1219 condemned only 11%, or 2/19 defendants.

Dueling was afforded widespread legitimacy in the South. Successful duels publicly ended conflicts

Written records exist for hundreds, if not thousands, of Southern duels; see Byron (2008) which catalogs accounts of 734 duels scraped from contemporary newspaper records. Such newspaper accounts were often especially concerned with the particulars of the affair (i.e., at how many paces was it fought, how many shots were fired), the nature of the underlying public conflict which led to the duel, and whether or not the affair was conducted honorably with a mutually satisfactory arrangement reached. In our newspaper data set, among duels that are known to have been conducted in the United States, about three-quarters are fought in the South. The following excerpt from the Greenville Mountaineer (1/24/1845), found on the front page below a story about a local expedition to search for sunken treasure off the coast of Margarita, is typical:

Affair of Honor. — A hostile meeting was had between Mr. Thomas Butler Kind and Mr. Charles Spalding, on Monday, the 6th inst. at Amelia Island. Weapons, pistols — distance, ten paces. Two shots were passed without effect, when, on the intervention of friends, the affair was adjusted and the parties exchanged friendly salutations.

Some announcements were terser, merely stating that a duel had taken place and the conflict ended honorably, such as the following announcement from a duel’s seconds appearing in the Columbia Telescope (9/20/1834):

The affair of Honor pending between Mr. William L. Allston and Mr. William M. Armstrong, having been referred to us, was Honorably adjusted.

Reporting standards varied across newspapers; for example, an 27 October 1830 article in the Arkansas Gazette, upon reporting “an honorable adjustment of the dispute, to the mutual satisfaction of both parties,” left it to “Madame Rumor, ‘with her hundred tongues,’ to communicate the names of the parties.” Examples of similar reports abound, in which Southern newspapers presented, without irony, the basic facts of a duel, what was known about the underlying dispute, and whether or not the affair was conducted honorably. ¹¹ The widespread uncritical reporting of particular duels suggests two things: one, dueling was accepted as common practice by Southerners. Two, duels were not private affairs, but were conducted at least partly for public consumption, and Southern gentlemen were likely to be well aware of how the parties acquitted themselves in any affair of honor. In addition to the press’ reporting of affairs of honor, Southern duels were often attended by scores of people. For example, the Camden Confederate (11/6/1863) reports “there were about seventy-five spectators at the scene” of a “fair stand up fight (with) both parties evincing great coolness” resulting in the death of a Mr. Copeland of Maryland.

Dueling appears to have been preferred to the legal system for the settling of disputes among Southern gentlemen, who disdained civil trials in which they would be judged by a jury they felt were socially inferior. Williams (1980) describes the duel as “clear evidence of the disinclination of Southerners to use the courts in connection with personal matters.” While a court may have been able to give pecuniary remuneration for an insult, they could not remedy the damage to a gentleman’s honor. As General Oglethorpe put it, a meeting on the field of honor was “essentially self-defense… a man has a right to defend his honor” (Stevens, 1940, pg 14). Schwartz et al. (1984) posit that contemporary courts’ reluctance to accept “truth of the matter asserted” as a viable defense may have rendered a jury award ineffective at restoring honor.

Southern gentlemen returning from the field of honor often enjoyed legal immunity; even in the rare instances in which a dueler was tried by a jury, the outcome was often acquittal. There is only one record of an execution resulting from a duel, in Illinois, and even this was more for dishonorable conduct than murder. ¹² A contemporary account claims “sometimes two or three hundred people hurried from the city to witness these human baitings,” suggesting that the probability of future legal trouble was quite low (New Orleans Times-Democrat, 13 March 1892). Similarly, participation in a duel seemed to (if anything) help, rather than hinder, a politician’s path to high office (see subsection ‘Duels arose out of public contests’, supra).

Greenberg (1990) describes dueling in the South as an “elite response to insult,” especially accusations of lying. While the duel appears to have remedied such an insult in the South, he writes that Northerners, including Benjamin Franklin, simply didn’t understand why participation would have such an effect. Northern newspapers were far more likely to report duels with derision. This treatment of affairs of honor in the press reflects Northern attitudes towards dueling, but it also greatly reduced the effectiveness of an affair of honor at achieving any end whatsoever—as duels were fought for public consumption, the fact that they would not be generally accepted by the press by itself ensured Northerners would pursue other methods of resolving conflicts, such as jury trials.

Duels were looked upon with scorn and derision of increasing intensity the farther north one traveled. The sarcasm in a 2/15/1872 New York Times article, describing three duels that occurred in New Orleans, is typical:

The Hamilton-Burr duel, involving a founding father and the sitting vice president, naturally attracted national attention, with views of the affair predictably fracturing upon regional lines. James Robertson of Tennessee, generally an opponent of dueling, wrote that “I suppose that if dueling could be justifiable, it must have been in his case” (Brands, 2006). Indeed, Brands writes that when passing through Nashville, Burr “was feted as a celebrity and a minor hero. No one in Nashville held his killing of Hamilton against him. Honor was honor, and, besides, to most Tennesseans, the fewer Federalists the better.” The duel, however, destroyed Burr’s reputation in the North. Brands writes that Burr was “politically ruined” and that his fellow Republicans considered him “an embarrassment and a liability.”

Perhaps because of the legitimacy of the institution, duels appeared to rehabilitate the reputations of both participants. As Parker (2001) notes, “the restoration of honor did not depend on the outcome of the confrontation. Ideally, a well-fought duel reconciled the two adversaries, reestablished mutual respect, and ‘cleansed’ the stain caused by the original insult.” Allen and Reed (2006) provide evidence that “the subsequent renewal of friendship among adversaries” was common, with duels refocusing the public’s attention away from the libelous behavior that led to the duel. Online Appendix B contains various accounts of duels that ended amicably, with any libel ostensibly withdrawn.

A game theoretic model of dueling

We model a duel as one possible outcome in the final stage of a multi-stage game. In the first stage, two randomly matched agents interact in a contest for public esteem (such as an election, letter-writing campaign, or gossip campaign). We assume the contest has one winner and one loser, and each agent would like to win the contest. ¹³ The agents have different types that render them more or less likely to win. An agent’s type—Agent i ’s type is denoted θ i in the formal model described in Online Appendix A—encompasses his positions on issues, his charisma, his past reputation, or any other factor that might affect the probability with which he wins the contest. Without loss of generality, we assume that a higher type is more likely to win the contest; such a type may have a position more in line with that of the median member of the public or may have a broader social network. To consider the concrete examples mentioned earlier, perhaps the contest is for influence in choosing methods of instruction at a university, or in choosing trade policy towards foreign countries. In these examples, type might represent each agent’s position on the policy question. Victory in the contest could be adoption of an agent’s preferred policy, or simply having relatively greater influence in crafting the policy. Each agent’s type is immutable and is publicly observable. ¹⁴

An agent’s probability of winning the contest is increasing in his type and decreasing in his opponent’s type. While higher types are more likely to win contests, type is not determinative; agents with less favorable types can and do win contests, due to (unmodelled) factors such as luck.

Agents may attempt to improve their chance of winning the contest in various ways. For example, they might work to clearly state their own views, or to explain the merits of these views to members of the group they aim to influence (i.e., they might use constructive communications). Further, they might seek to discredit their opponent, such as by undermining his credibility, attacking his motives, distorting his views, or suggesting he is generally untrustworthy and undeserving of public esteem (i.e., destructive methods). We abstract from the former with the understanding that both agents would use such means to the extent practicable. We summarize the latter, destructive communications in one variable, which we call “libel.” Libel—the amount of Agent ’s libel is denoted l i in the formal model presented in Online Appendix A—distorts the public’s perception of its subject’s positions. For example, while an agent with a high type may be more likely to win a contest ex ante, libel directed at that agent would cause the public to perceive that agent as though he were of a lower type, thus lowering his likelihood of winning the contest. Thus, we model the probability of an agent winning the contest as increasing in his “perceived type”, defined to be a decreasing function the amount of libel produced by his opponent.

Producing libel is costly. Costs may be pecuniary (e.g., printing libelous pamphlets) or non-pecuniary (e.g., the time cost of preparing editorial material regarding a rival’s moral fitness). We model these costs as convex, so that additional libel becomes costlier as more has been thrown (or, equivalently, that libel has diminishing effectiveness). In the formal model presented in Online Appendix A, Agent ’s cost of libel is given by a function c ( l i ) = c * l i 2 , with the parameter c being proportional to the marginal cost of libel. We assume that agents in a contest with one another simultaneously choose libel levels (although simultaneity is not essential to any result).

Once libel levels are realized, either agent may choose to issue a challenge to the other. A challenged agent may choose to accept the challenge—in which case a duel takes place—or to refuse. Consistent with the historical record, we assume that refusing a challenge lowered an agent’s perceived type, e.g., by tarnishing the agent’s reputation. In the formal model described in Online Appendix A, Agent ’s perceived type is θ i − θ i γ l j 1 + l j , where γ ∈ ( 0 , 1 ) is the effectiveness of Agent j ’s libel in affected Agent i ’s perceived type. Indeed, declining a challenge gave the aggrieved party license to ‘post’ his antagonist, publicly declaring him a poltroon (Holland, 2003). Thus, we assume that the cost of refusing a challenge was prohibitively high for most challenged parties. ¹⁵

If a duel takes place, it has the effect of diminishing the effect of both agents’ libel. Based on the material presented in the ‘Overview of dueling’ section, duels appeared to defuse arguments and to shift the nature of underlying contests away from libel and back to the merits of the participants’ positions. As described earlier, a successful outcome of a duel was for the combatants to exchange shots, and then to “exchange … friendly salutations,” and contemporary reports describe such resolutions without reference to the underlying contest. This is both because the friendly salutations that follow presumably constitute a withdrawal of previous libel, and because a duel was widely acknowledged as a way for a dueler to “defend his honor.” ¹⁶ Thus, we model a duel as removing a percentage of both agents’ libel, so that the contest is decided as though both agents issued less libel. In the formal model presented in Online Appendix A, the perceived type of Agent i after a duel takes place is given by θ i − θ i ( 1 − α ) γ l j 1 + l j , where α ∈ ( 0,1 ) is the effectiveness of a duel at reducing the effect of libel, So that a higher value of α eliminates more of the negative effects of libel on an agent’s perceived type.

Of course, dueling is costly to both agents to the extent it carries a risk of death or injury, each of which lowers an agent’s utility—in the formal model presented in Online Appendix A, A is the disutility of death, and d is the probability of dying in a duel. We assume that agents engage in standard cost/benefit analysis in deciding whether or not to challenge their rival to a duel. Agents weigh the probability of death or injury against the benefit to his reputation from participating in an affair of honor and removing some of the libel thrown at him. In particular, while a duel’s reduction of libel applies to both agents, an agent who received relatively more libel may have more to gain from a duel. It follows that the more libel leveled against an agent, the greater the benefit to his dueling, all else equal. While either agent could refuse the duel, the cost of turning down the challenge was high enough that such rare incidents were noted in newspapers in the Antebellum South.

Two parameters inform an agent’s choice of whether or not to issue a challenge. First, the “effectiveness” of a duel in reducing libel (specifically, α , the fraction of libel that is eliminated by a duel) determines the potential benefit to a duel. Evidence presented in the previous section suggests that this fraction was non-zero, and may have been large, in the Antebellum South. In contrast, this fraction was likely at or below zero in the North, at least following the Burr-Hamilton duel.

Second, the probability of dying in a duel (the parameter d ) affects the cost of dueling. We assume for simplicity that all duels carry the same probability of death; this is consistent with the widespread use of inaccurate dueling pistols (see subsection ‘Dueling was relatively safe’), and widely adopted conventions that specified how gentlemen should behave during interviews. We further assume away (again, for simplicity) the probability of sustaining non-fatal injuries in a duel. Thus, a duel results in the death of zero, one, or both agents. We further assume that both agents have the same disutility of death (the parameter A ).

Figure 2 depicts representative strategies from an extensive form representation of the underlying game; while we model libel as a continuous variable, for the purposes of the figure we have simplified the agents’ choice to one of high or low libel. Working backwards, if a challenge has been issued, the challenged agent is assumed to suffer a reputational cost from declining that outweighs any costs associated with participating in a duel, such Agent two prefers outcome III to outcome IV. Knowing this, an agent is more likely to prefer to issue a challenge if libel directed at him is high, and libel the agent directed at his opponent is low; the optimality of issuing a challenge also depends on the deadliness of a duel and its effectiveness in reducing the harmful effects of libel. Depending on these parameters, and within the example in the figure, Agent 1 may prefer either outcome II (from not issuing a challenge) or outcome III (from issuing a challenge). Finally, in the game’s initial stage, both agents choose libel to equate the marginal benefit of libel (a greater probability of winning) with the marginal cost, bearing in mind that too much libel may prompt a duel. Depending on the cost of the duel, the effect of a duel on libel, and the agents’ relative type, in the example depicted in the figure Agent 2 may prefer outcome I or whichever of outcome II or II would realize if Agent one chooses a high level of libel.OPEN IN VIEWER

As a general matter, three types of equilibria are possible. In an unconstrained equilibrium, the level of libel chosen by agents in the first stage is insufficient to trigger a challenge to a duel by either agent in the second stage. In this type of equilibrium, the game’s second stage does not alter agents’ decision making in the first stage. Second, in a dueling equilibrium, a duel takes place in the second stage. In such an equilibrium, one agent is libeled so much in the first stage that he finds it optimal to issue a challenge in the second stage. Of course, agents will anticipate a challenge in such equilibria, which diminishes the expected return to libel in the first stage. Finally, in a deterrence equilibrium, at least one agent chooses to libel his rival less than he otherwise would in order to avoid a second stage duel. No duel takes place in these equilibria, but the threat of a duel results in less libel in the first stage.

In summary, the primitives of our game theoretic model are: the cost of libel ( c ), the effectiveness of libel in influencing the contest ( γ ), the probability of dying in a duel ( d ), the effectiveness of a duel in reducing effective libel ( α ), and the disutility of death ( A ). Our results map these primitives into various outcomes. We rank outcomes in terms of social welfare and derive our main results in the following section. Two results are of particular importance. First, if dueling is effective at reducing libel, the legalization of dueling may be welfare enhancing. Second, the probability of dying in a duel affects social welfare non-monotonically, and welfare is maximized by an intermediate value, neither too high nor too low.

Results

Online Appendix A leverages functional form assumptions to formally solve the model via backward induction. We provide an intuitive description of major findings here.

First, at most one agent would prefer to issue a challenge to a duel in the game’s second stage. This is because the contest is zero sum: one agent will win and the other will lose. ¹⁷ A duel may change the probabilities of each agent winning, but it cannot do so in a way that would make both agents more likely to win the contest. Of course, it is possible that it does not increase the probability of either agent winning by enough to offset the possibility of dying in a duel, in which case no agent will choose to issue a challenge in the second stage.

Second, we use the standard Nash equilibrium concept to solve for optimal libel levels in the first stage of the game. This means that each agent chooses the level at which he libels his rival to equate the marginal benefit and marginal cost of libel, accounting for the possibility that a duel may occur in the second stage, which both lowers the return to libel (as a duel partially eliminates the effect of libel) and makes death a possibility. In an unconstrained equilibrium, each agent simply equates the marginal benefit and marginal costs of libel. In a dueling equilibrium, each agent does the same, but accounts for the diminished return to libel. Finally, in a deterrence equilibrium, the agent who is deterred plays the maximum libel level that will not induce a challenge from his rival. ¹⁸

Third, which type of equilibrium occurs depends on the model’s underlying parameters, such as the probability of dying in a duel, the effectiveness of a duel in reducing libel, and the difference in type between the two agents. In particular, duels occur when the probability of dying in a duel is low, the effectiveness of a duel in reducing libel is neither too high nor too low, and the difference in type between the two agents is large. The intuition is as follows. If duels are too deadly, no agent will want to partake, regardless of effectiveness in reducing libel. If duels are ineffective at reducing libel (e.g., because more traditional institutions provide greater value in resolving disputes), there is no point in risking death to duel. On the other hand, if duels are very effective at reducing libel, they provide an effective deterrent to excess libel, and neither agent will choose to libel his opponent enough to induce a duel in the game’s second stage. Finally, the difference in type between the two agents matters because an agent with a type that is ex ante less likely to win the contest has a greater return to libeling his opponent, and thus, all else equal, duels are more likely to occur in contests involving such agents.

Fourth, deterrence equilibria occur for intermediate levels of the probability of death from a duel. If a duel is somewhat likely, but not overwhelmingly likely, to result in the death of one or both agents, it becomes a more credible threat, and in the game’s first stage each agent will believe that his rival may well challenge him to a duel if he were to excessively libel the rival. This threat evaporates if duels are overly deadly, as an agent will know that even egregious libel is unlikely to result in a challenge, as his rival would not find it optimal to incur such a high risk of death to restore his honor. The balance between deterrence and acceptance is not unique to duels. Leeson and Coyne (2012) argue that a superstitious belief that sassywood ordeals are reasonably safe for the innocent but deadly for the guilty – supported by manipulation of the ordeals by sassywood specialists – underlies broad Liberian public support for the institution. Leeson (2012) argues that priests manipulated the outcomes of ordeals to ensure they harmed participants often enough to be credible, but not so often that innocents refused to participate.

Fifth, we posit that total social welfare is decreasing in both the number of duels (because duels result in at least occasional deaths) and the amount of libel (because libel distorts political and other processes, resulting in suboptimal outcomes). If so, it follows (from the previous paragraph) that an intermediate level of deadliness of dueling weapons generates the greatest social surplus, as such weapons can deter bad behavior (libel) without excessive deaths. Indeed, in the context of our model, the deadliness of dueling weapons can even be chosen so as to result in no duels in equilibrium, but to still deter excessive libel. In the absence of strong institutions allowing aggrieved parties to seek redress for damages caused by libelous speech (as in the modern era), this level of deterrence is arguably superior to a regime that outlaws dueling, as both regimes produce zero dueling deaths, but allowing for the possibility of a duel deters libel.

This result is the most important point we make in the paper, so it is worth unpacking further. Figure 3 plots, for a numerical example, the equilibrium number of deaths per public contest (as opposed to per duel), and the units of libel chosen, for various levels of the probability of dying in a duel, ranging from 0 (duels are perfectly safe) to 10.4% (i.e., higher than the mortality rate observed in the Antebellum South, per subsection ‘Dueling was relatively safe’). As the mortality rate increases from 0, duels become more of a deterrent, reducing libel (albeit at the expense of some deaths). As the mortality rate increases further still, agents are more hesitant to issue challenges. Thus, the deterrent effect begins to evaporate, and the level of libel observed in equilibrium increases (while the number of deaths/contest decreases). ¹⁹ OPEN IN VIEWER

While we are agnostic how society should trade off deaths from dueling against lower libel levels, the optimality of an intermediate level of deadliness from dueling holds regardless of how this tradeoff is resolved. In Figure 3, mortality rates corresponding to the shaded region, labeled “Optimal Mortality,” Pareto dominate points outside of the shaded region. This is because for any point not in the shaded region, the same level of dueling deaths can be obtained in the shaded region, with lower libel. To be clear, Figure 3 represents a particular numerical example, but the same intuition would hold even for different model parameters.

Sixth, the effectiveness of a duel in reducing libel is a key determinant of the frequency of duels and the level of libel. If dueling is ineffective at reducing libel, then a duel is not a credible threat capable of deterring libel. As seen in the previous section, dueling was widely mocked in the North, and thus was unavailing as a method of restoring one’s honor, even in the face of vicious libel. On the other hand, as seen in the subsection ‘Dueling was afforded widespread legitimacy in the South’, in the absence of credible legal institutions, Southerners saw dueling as a legitimate means of honorably ending conflicts, which thus established a duel as a credible threat, capable of deterring libelous behavior.

Seventh, and finally, the existence of multiple equilibria offers context for why dueling persisted in the U.S. South long after it fell out of favor elsewhere. In our model, which equilibrium obtains depends on the model’s underlying parameters. For instance, if duels are likely to be very deadly, no one will ever issue a challenge, and the threat of a duel will not deter libelous conduct. Similarly, if duels are held in low regard by the citizenry, they will be ineffective at ending conflicts, and will likely be ineffective at reducing the effects of libel, in which case agents will be hesitant to issue a challenge. In either case an unconstrained equilibrium is more likely to result. A dueling equilibrium is more likely to result when public’s acceptance of dueling is sufficiently high, but the deadliness is not too great. If (consistent with the historical record) a duel needed to be at least somewhat deadly in order to satisfactorily resolve a contest, then the effectiveness of dueling may have depended on duels being somewhat, but not overly, deadly.

Empirical support

In this section, we provide empirical analyses with our data sets to buttress some of our theory and simulation results. As described in the introduction, our first data set collects anecdotal evidence of dueling by U.S. Senators. While historical accounts will be incomplete, the prominence of the participants most likely means that records will be more complete than affairs of honor involving the public at large.

Our second data set is the universe of dueling incidents reported by two large newspapers from 1861 to 1865: the New York Times and the Richmond Daily Dispatch. ²⁰ The two newspapers were selected for study because, to our knowledge, they are one of the few, if not the only, newspapers with extensive digitized archives that go back to the Antebellum South. Also, the two newspapers seem to employ separate correspondents. As such, accounts of duels do not overlap. We initially extracted over 600 articles that contain variations on the word “duel” and “affairs of honor.” Of these, we found a large majority did not pertain to the duels we were interested in. Many were descriptions of battles between the North and the South (“artillery duels”), rumors of duels scheduled or finished with no additional details, or other unrelated materials. After excluding irrelevant and unusable items, we were left with 130 articles.

Figure 4 shows the trend in participation in duels reported by the newspapers. This figure is analogous to Figure 1, which shows a large number of senators participating in duels prior to the run-up of the Civil War. After the war, this number plummeted. Similarly, the number of accounts of dueling in newspapers is much higher in 1861 and declines precipitously throughout the war period. These trends imply, first, that duels were on the decline over the long run. This may have been due to the strengthening of institutions to settle disputes or simply that dueling for honor was becoming less in vogue. Second, dueling was concentrated in the South, and as the Civil War became an existential struggle, the luxury of engaging in affairs of honors seems to have disappeared.OPEN IN VIEWER

Table 1 shows summary statistics for our newspaper database. Several features of the data stand out. First, over half of the articles describe duels featuring prominent public figures. As newspapers report on what the public would find interesting, this likely induces selection into which of many duels are reported in newspapers. In particular, it is likely that there were many unreported duels featuring non-prominent participants and with motivations and results not of particular interest to the newspaper-reading public. In addition, the probability of death from the newspaper articles is approximately 0.27, which is more than three times what was reported in Schwartz et al. (1984). This again implies that duels by themselves were not necessarily newsworthy.

OPEN IN VIEWER

Table 1.

Summary statistics from articles appearing the New York Times and Richmond Daily Dispatch, 1860-1865.

		Mean	Std. Dev
Participants	Prominent Figures	0.55	0.50
	Average citizens	0.31	0.46
Outcome	Death	0.27	0.45
	Major injury	0.09	0.29
	Minor injury	0.16	0.37
	No harm	0.15	0.36
	Ended peacefully	0.21	0.41
Reason for duel	Political	0.18	0.38
	Military	0.08	0.27
	Personal insult	0.06	0.24
	Financial	0.03	0.17
	Miscellaneous	0.17	0.38
	Unknown	0.48	0.50
Weapon	Dueling pistol	0.25	0.44
	Unspecified	0.53	0.50
Location	South	0.44	0.50
	North	0.15	0.36
	Foreign/Unknown	0.41	0.49
Observations			130

Over half of the duels in the newspaper articles end with minor to no injuries or without actual shots being fired. Descriptions of conclusions of such incidents are straightforward and matter of fact, with no negative sentiment found. This supports the assumption underpinning our model that the issuing of the challenge is the important event, and the outcome is somewhat secondary.

We find that among duels with known causes, the most common reason for the duel was political arguments. However, the largest category of reasons is unknown, or at least unspecified in the relevant newspaper article. This may imply that the reason for the duel was already widely known to the public, and the reporter did not feel the need to describe it. The most common weapon used was the standard dueling pistol. However, as with reasons for the duel, the largest category of weapons is unspecified. It is likely that the vast majority of these duels used dueling pistols, and these weapons were not described by the article as their presence was unremarkable. Finally, as expected, a large majority of duels were fought in the South (among known locations). The imbalance in the number of duels is interesting as the number of articles from the New York Times is approximately double the number of articles from the Richmond Daily Dispatch.

Table 2 breaks down the outcome of the duels into South and North. The most salient differences between duels in the North and the South is the expected level of violence. Death and dismemberment were almost half of the expected outcomes in Southern duels, whereas the conflict ending without an actual exchange of gunfire or other forms of violence was the most likely outcome for Northern duels.

OPEN IN VIEWER

Table 2.

Dueling outcome by region.

By region	South		North
	Mean	Std. Dev	Mean	Std. Dev
Death/major injury	0.46	0.50	0.35	0.49
Minor injury/no harm	0.39	0.49	0.25	0.44
Ended peacefully	0.11	0.31	0.40	0.50
				Observations 77

Table 3 shows linear probability model regression results where the outcome is that a duel takes place after the issuance of a challenge. Model 1 includes information only on the reason for the duel. Model 2 adds the social status of the participants. Model 3 adds interaction terms between the status of the participants and whether the duel was due to a political dispute. ²¹ Finally, Model 4 includes variables for weapons used, location, and whether the challenge was issued prior to the Civil War. We caution that these results should be interpreted in the context of a selected sample. As mentioned previously, it is likely that many challenges that ended peacefully were under-reported. The sample contains reports of incidents where, although no physical violence occurred, the newspaper still deemed it worthy of print.

OPEN IN VIEWER

Table 3.

Linear probability regression results, outcome variable is whether a duel takes place.

Variables	(1)	(2)	(3)	(4)
Political dispute	−0.278**(0.110)	−0.298**(0.119)	0.0105(0.167)	0.0124(0.158)
Military dispute	0.0465(0.130)	0.0329(0.130)	0.0722(0.139)	0.0517(0.132)
Personal insult	−0.0258(0.119)	−0.0483(0.124)	−0.0281(0.112)	−0.00919(0.111)
Prominent Figures		0.0709(0.0931)	0.0901(0.0956)	0.0534(0.0944)
Private citizen		0.130(0.0904)	0.150(0.0980)	0.0967(0.0976)
Prominent x political			−0.447**(0.219)	−0.403*(0.206)
Citizen x political			0.0551(0.286)	0.0814(0.268)
Dueling pistol used				0.111*(0.0651)
South				0.115*(0.0674)
Pre-civil war years				−0.0240(0.0822)
Constant	0.840***(0.0390)	0.767***(0.0808)	0.748***(0.0846)	0.717***(0.0991)
Observations	130	130	130	130
R-squared	0.068	0.086	0.124	0.159

The negative parameters on political dispute (in Models 1 and 2), and on the interaction term between prominent citizens and political dispute (in Models 3 and 4) may seem odd, when one of our central claims is that political disputes play a central role in dueling in the Antebellum South. However, the implication of the results is that in political disputes, especially among the social elite, a challenge is much more likely to be concluded without trading gunfire, and someone being injured or killed. This again implies that the issuance of the challenge and its acceptance are the salient features of the duel, and the violence arising from the duel itself may not be necessary for mutual satisfaction. ²² As we saw in Table 2, challenges are more likely to be resolved through a violent duel in the South.

Table 4 shows linear probability model regression results for the sub sample of articles where a duel did not terminate peacefully. The dependent variable is that the duel ends with the death or major injury of at least one combatant. The regression specifications are built up in the same way as described for Table 3. Although not statistically significant, a suggestive finding is that a political dispute is more likely to end in death or major injury compared to other reasons. Prominent public citizens are more likely to suffer grave bodily harm if they are involved in a duel. Private citizens in general may be more likely to suffer harm, although a puzzling finding is that if the dispute is political in nature, they are much less likely to be seriously injured or killed. This may be due to our selected sample. The public will be more interested in the details, including the cause of the dispute, when prominent citizens are killed in duels. At the same time, prominent citizens may be more likely to argue and duel over political matters compared to the average citizen. Then, for average citizens, even in the few cases of duels arising from political arguments, newspapers may be less inclined to report on these details due to the lack of interest from the public.

OPEN IN VIEWER

Table 4.

Linear probability regression results, dependent variable is whether a duel resulted in a death.

Variables	(1)	(2)	(3)	(4)
Political dispute	0.185(0.148)	0.112(0.172)	0.165(0.263)	0.183(0.276)
Military dispute	−0.120(0.181)	−0.146(0.185)	−0.246(0.182)	−0.260(0.185)
Personal insult	−0.116(0.204)	−0.159(0.209)	−0.176(0.208)	−0.175(0.219)
Prominent Figures		0.217*(0.116)	0.275**(0.115)	0.281**(0.118)
Private citizen		0.219*(0.122)	0.332***(0.124)	0.336***(0.127)
Prominent x political			0.243(0.293)	0.248(0.296)
Citizen x political			−0.897**(0.343)	−0.943**(0.395)
Dueling pistol used				−0.0666(0.110)
South				0.119(0.104)
Pre-civil war years				0.00406(0.115)
Constant	0.449***(0.0570)	0.274***(0.0900)	0.215**(0.0872)	0.168(0.125)
Observations	103	103	103	103
R-squared	0.021	0.063	0.113	0.129

The parameter estimate on prominent figures speaks to the real social harm that dueling can cause: serious disputes that cannot be resolved before the actual execution of the duel can result in the death or dismemberment of even social and political elites. Somewhat surprisingly, disputes of military matters or grave personal insults are not associated with higher likelihood of traumatic outcomes.

Finally, we use a standard sentiment analysis tool — spaCy ²³ — to assess the tone associated with articles in both newspapers. The algorithm, which is trained on a manually classified corpus of movie reviews appearing in newspapers, scores each article’s tone on a scale of 0 to 1, with 0 being the most negative, and one being the most positive. We augment the algorithmic scoring with our own manual review, in which we flagged sarcastic and/or condemnatory articles against dueling. ²⁴ This step was necessary because the machine learning algorithm we employed cannot reliably assess sarcasm. We found that 10% (8 out of 80) New York Times articles used sarcasm or strong condemnatory language, while zero of the 50 articles in the Richmond Daily Dispatch used such language.

Table 5 and Table 6 present correlation matrices describing how various factors — whether a challenge was peacefully concluded (labeled No duel), whether the participants or a duel were prominent, whether the duel’s cause was political, and whether a duel took place in the U.S. South — are correlated with article sentiment, and with each other. We find that article sentiment is positively correlated with peaceful conclusion in the northern New York Times, but uncorrelated with no violence taking place in the Southern Richmond Daily Dispatch. Sentiment is roughly equally correlated with a duel’s political nature in both papers. A duel taking place in the South is negatively correlated with sentiment in the Times but positively correlated in the Dispatch. Finally, sentiment is much more positively correlated with prominent participants in the Times than in the Dispatch, but this appears to be mainly due to higher correlation between “No duel” and “Prominent” and between “Political” and “Prominent” in the Times, relative to the Dispatch.

OPEN IN VIEWER

Table 5.

Correlation between article sentiment and other factors, New York Times.

	Sentiment	No duel	Prominent	Political	South
Sentiment	1.00
No duel	0.16	1.00
Prominent participants	0.16	0.22	1.00
Political origin	0.19	0.34	0.56	1.00
South	−0.06	−0.25	−0.25	−0.16	1.00

OPEN IN VIEWER

Table 6.

Correlation between article sentiment and other factors, Richmond Daily Dispatch.

	Sentiment	No duel	Prominent	Political	South
Sentiment	1.00
No duel	0.00	1.00
Prominent participants	−0.12	0.03	1.00
Political origin	0.20	−0.11	0.17	1.00
South	0.09	−0.02	0.10	−0.08	1.00