Allegiance, Ability, and Achievement in the AmericanCivil War: Commander Traits and Battlefield Military
Effectiveness*
Jeffrey B. Arnold†
J. Tyson Chatagnier‡
Gary E. Hollibaugh, Jr.§
January 28, 2018
Word Count: 12,450 words
Abstract
How do the characteristics of military leaders relate to battlefield outcomes? To answerthis question, we employ original battle-level data and biographical information on hundredsof commanders in more than 250 battles in the American Civil War. We examine the rela-tionship between two common measures of battlefield success (victory and casualties) andtwo latent features of commanders—competence and loyalty—that have long been seen asimportant in the broader study of executive appointments. We find that competent comman-ders are associated with more successful battlefield outcomes, as are more loyal Confederatecommanders. More broadly, our analysis suggests that focusing on the relationship betweenmilitary appointments and battlefield outcomes—with the latter’s relatively clear definitionof “success”—allows for direct examination of the relationship between appointee traits andorganizational performance. As such, our results have implications for the study of conflict aswell as bureaucratic politics.
*Authors listed in alphabetical order, though this is observationally equivalent to the results of our most recentfoot race. All contributed equally to the paper. Thanks to Alexander Acs, Alexander Bolton, Gregory Brill, BenjaminFordham, Hein Goemans, Richard Jordan, Kerim Can Kavaklı, Lael Keiser, Dave Lewis, Michael McKoy, Chad Nelson,Paul Poast, Melinda Ritchie, Theodore Samuelson, Kathryn Spond, and Alan White for their comments, as well asattendees at the American Civil War and International Relations Workshop at the 2015 Annual Meeting of the PeaceScience Society and those at the 2016 and 2017 Annual Meetings of the Southern and Midwest Political ScienceAssociations. All errors remain our own.
†Assistant Professor, Department of Political Science, University of Washington, Box 353530, Seattle, WA 91895.Email: [email protected]
‡Assistant Professor, Department of Political Science, University of Houston, Houston, TX 77204. Email: [email protected]
§Assistant Professor, Department of Political Science, University of Notre Dame, Notre Dame, IN 46556. Email:[email protected]
1 Introduction
“In carrying out any system of policy which you may form you will require a
commander-in-chief of the army, one who possesses your confidence, understands
your views and who is competent to execute your orders by directing the military
forces of the nation to the accomplishment of the objects by you proposed.”
— Letter from Gen. George B. McClellan to President Abraham Lincoln (July 7, 1862).1
The path to victory on the battlefield has evolved over the centuries. According to Andreatta,
warfare in the ancient world—prior to the development of the Greek phalanx—was heroic in
nature, with aristocrats doing battle and masses of infantry essentially serving auxiliary roles.2
Greek democracy brought with it the democratization of warfare, in which companies of hoplites
fought one another in formation. Military engagements were further altered when Persia invaded
the Greek city-states, and Militiades’ genius was instrumental in developing a stratagem to defeat
Darius’ forces. Without his insights, the battle might well have been lost. It was at this point that
the cornerstone of the battlefield changed from the mighty warrior to the knowledgeable general.
But what is the role of the commander in ensuring victory?
While political scientists have largely shied away from analyzing leadership,3 military strategists
and historians have long recognized the critical role of command in battle outcomes. A. A. Van-
degrift, a USMC General in WWII, asserted that “positions are seldom lost because they have been
destroyed, but almost invariably because the leader has decided in his own mind that the position
cannot be held.”4 Plutarch claims that Quintus Fabius Maximus was the only man capable of
stopping Hannibal’s invasion of Italy.5 Liddell Hart traces the importance of six great comman-
1As recounted by McClellan 1887, p. 489.2Andreatta 2015.3Some important exceptions are discussed below.4United States Marine Corps 1997, p. 1.5Plutarch 1992.
1
ders in the second millennium.6 Taaffe highlights the decisions made by U.S. Army Chief of Staff
George C. Marshall in World War II in selecting officers who could defeat the Axis powers.7 And
Reiter and Wagstaff examine the performance of commanders in World War II, suggesting that the
replacement of lower-performing generals improved combat effectiveness of both the American
and German militaries.8
One conflict for which the importance of command has been particularly emphasized is the
American Civil War (hereafter, “the Civil War”), in which early Confederate victories have long been
attributed to superior leadership.9 Contemporaneously, Harper’s Weekly blamed Union officers for
the defeat at the First Battle of Bull Run, saying “[b]etter offend a thousand ambitious candidates
for military rank, than have another flight led by colonels, majors, and captains.”10 As the war
progressed, the leadership gap narrowed, and the Union exploited its advantages in population and
resources.11 This account suggests Lincoln might have appointed more competent commanders
later in the war, seeking not only those expected to be loyal,12 but those that could secure victory
on the battlefield, a notion supported by this paper’s opening epigraph.13
This shift in Lincoln’s choices is relevant, as scholars of the executive appointments process
often focus on loyalty and competence (and sometimes patronage considerations) as key traits that
presidents typically desire in their appointees.14,15 However, the relative weights presidents and
6Liddell Hart 1996.7Taaffe 2011.8Reiter and Wagstaff 2017.9See, e.g., Alexander 2007; McPherson 1988.10Cited in McPherson 1988, p. 327.11Bond 1998.12“Loyalty” as a construct is often ill-defined, and has been used in various circumstances to mean personal loyalty,
ideological affinity, shared preferences for particularized policies, and the like. As we discuss beloow, we implicitlyconceptualize loyalty as akin to support for the administration’s wartime objective. We point this out because Lincolnappointed many War Democrats to high-ranking political and military positions, many of whom disagreed with himideologically and/or politically, but nonetheless shared his broader wartime aims.
13It also implies the possibility of learning over the course of the war.14See, e.g., Edwards 2001; Hollibaugh 2016b; Hollibaugh, Horton, and Lewis 2014; Krause and O’Connell 2016; Lewis
2008.15These same traits have been emphasized by those involved in the presidential appointments process. For
example, Ed Meese, then-counselor and later Attorney General to then-Governor Reagan, devised an index cardentitled “Criteria in Selecting Appointees” that listed both “competence” and “commitment to Governor Reagan’s
2
their administrations place on these two traits often vary both over time16 and across adminis-
trations.17 Some of this variation might be due to personal or ideological preferences, but much
of it is due to an inherent conflict between the two traits; indeed, standard theories of executive
appointments suggest a tradeoff between loyalty and competence.18,19,20
Such a tradeoff has long been claimed for the Union, with many “political generals” having
been appointed with the aim of appeasing important constituencies, rather than for demonstrated
military capability.21 Indeed, Warner notes that, for the Union, “politically inspired appointments
and promotions were made without end, many to the prejudice of the lives of the men in the
ranks.”22 Even analyses written by those who served support this claim; Brigadier General Anson
Mills (U.S.A.) criticized what he saw as the overuse of brevet appointments for field purposes,
writing in his memoirs that “the conferring of brevets was so overdone by political and other
influences” that “[t]he army became dissatisfied” due to this perceived overuse.23,24
Conversely, the Confederacy is generally considered to have had an advantage in military lead-
ership.25 Even senior officers in the Union Army believed the Confederacy to be more disciplined
philosophy/policies/objectives”, in addition to “integrity”, “team work”, and “toughness.” See Rudalevige 2017.16Krause and O’Connell 2016.17On this latter point, H. R. Haldeman, President Nixon’s onetime Chief of Staff, noted during a post-election meet-
ing with then-President Nixon that “[l]oyalty [is] much more important than competence” when selecting appointees.See Rudalevige 2017.
18See, e.g., Edwards 2001; Hollibaugh 2016b; Hollibaugh, Horton, and Lewis 2014; Krause and O’Connell 2016; Lewis2008.
19Although their overuse will likely diminish organizational performance, more “political” types of appointees (e.g.,those selected primarily for reasons of loyalty or patronage) are often necessary for ensuring organizational control,counteracting inertia, and introducing new sources of information (Lewis 2009; Moe 1985).
20The existence of a tradeoff does not imply that some individuals cannot be superior on both the loyalty andcompetence scales.
21See, e.g., Goss 2003; Patterson 2014; Simpson 2000; Warner 1964; Work 2012.22Warner 1964, p. xvi.23Mills 1918, p. 209.24However, the use of “political” generals may accomplish other war objectives. Goss 2003 suggests Lincoln’s
“political generals” were instrumental in raising morale and national support for the war—and sometimes achievedsuccess on the battlefield—though they often clashed with more traditional generals when it came to the broaderdecisions of how to prosecute the war.
25However, Woodworth 1990 argues that Davis’s military experience led him to micromanage his general and flagofficers (GFOs) at both operational and tactical levels, hindering the performance of the larger Confederate military.Moreover, the question of whether the Confederacy truly had superior leadership is a matter of minor contention.Cullum 1879 notes that fewer than 25% of living United States Military Academy (USMA) graduates at the time of the
3
and to have better leadership at the start of the war.26 On this point, Owsley argued this was in
part due to Southern culture:
“The statement that the South had the better generals and better-trained soldiery at
the outbreak of the war will probably go unquestioned at the present day, for all recall
the fact that the prominent southern families took a special pride in sending their sons
to West Point and to the several well-equipped military academies in the South, and
that the mass of southern people were accustomed to the use of the rifle and pistol
and to horseback riding and other exercises closely related to military training and,
moreover, the institution of slavery tended to develop the militant temperament.”27
However, the Confederacy’s advantage in military leadership cannot be attributed solely to
Southern culture.28 Indeed, President Davis likely played some role himself, as he “looked to West
Pointers and to men he knew and trusted” when assembling the Confederate government and
military.29 Given these collective differences in culture, presidential choices, and war objectives,
the loyalty-competence tradeoff may have been less pronounced within the Confederacy, if it
existed at all.
That the particular circumstances of the Confederacy might have weakened or reversed the
empirical regularity of the tradeoff provides an opportunity for scholars of both bureaucratic
politics and military leadership. Indeed, current theories explaining the tradeoff were developed
within the context of administrative politics, and not that of military leadership in war, where
incentives are different. The Civil War provides an ideal case to examine the relationship between
the traits of leaders and their performance, as well as the generalizability and applicability of
start of the Civil War joined the Confederacy, and fewer than 25% of active-duty USMA graduates at the start of thewar did the same.
26Haughton 2000.27Owsley 1925, pp. 499–500.28The emphasis on military service within the South might have led to greater deference to authority and the
development—or reinforcement—of political views that encouraged or accepted the continuation of hierarchical insti-tutions, such as slavery. See Klingler and Chatagnier 2014 for a discussion of this phenomenon in more contemporarytimes.
29Davis 1996, p. 316.
4
extant theories of executive politics. As mentioned, one advantage of using military outcomes
to examine the relationship between appointee characteristics and organizational performance is
that military goals and outcomes, at least at the tactical level, are readily defined. This is in
stark contrast to more conventional studies of bureaucratic politics; in these contexts, measuring
broader bureaucratic performance is often difficult due to ambiguous goals30 and bad metrics.31
In using the Civil War to examine the relationship between commander traits and battlefield
success, we provide several contributions to the study of conflict. First, we employ large-scale
data on the attributes of battlefield commanders in the Civil War. We use these data to compute
measures of loyalty and competence for commanders on both sides. This is, to our knowledge,
the first systematic investigation into the attributes that define a quality commander. This is
an especially important step, as it provides researchers with a way to gauge competence and
to include it as an explanatory variable in future work. We also illustrate the utility of using
the military—and battlefield outcomes—to study bureaucratic dynamics. Finally, we assess the
respective roles that loyalty and competence play in determining battlefield outcomes, directly
testing the widely-accepted idea that leadership matters in military engagements. Our results
are broadly supportive of the traditional hypothesis, suggesting that leadership is indeed related
to success, more loyal commanders are more likely to be viewed as “political” generals in the
modern era, and that while the Confederacy began the war with a competence advantage among
its commanders, the gap narrowed considerably as the war progressed.32
30Chun and Rainey 2005a,b.31For example, while Lewis 2007 and Hollibaugh 2015a use George W. Bush-era Program Assessment Rating
Tool (PART) scores as measures of agency performance, Gilmour and Lewis 2005 find that PART scores are correlatedwith political factors, and that what PART scores measure has “little relationship with performance” (185).
32While it would be interesting to examine how the loyalty-competence tradeoff calculus at the individual levelchanged over the course of the war, such an analysis is beyond the scope of the current paper. Instead, we leave thisto future research.
5
2 Military Leadership and Battlefield Outcomes
The international relations literature explaining the outcomes of military conflict has generally
focused on the overall outcomes of wars and militarized disputes, with less attention devoted to
explaining the outcome of combat (i.e., battles) within those those conflicts.33 At both the interstate
and intrastate levels, researchers have focused on factors such as regime type, alliances, material
capabilities, state capacity, industrialization, terrain, tactics, outside intervention, and attributes
of political leadership influence war initiation and outcomes.34 Though there are some notable
exceptions,35 few studies have focused on explaining battle-level effectiveness. Yet, understanding
combat within war is essential to understanding the onset, duration, and conclusion of war,
since a complete and coherent theory of war must explain how fighting is able to resolve the
bargaining failure that caused war in the first place.36 Moreover, the features that predict success
in militarized interstate disputes are generally poor predictors of battlefield success,37 suggesting
the need for further study.
One plausible determinant of battlefield success, military leadership, has been particularly
understudied. In considering the importance of leadership, international relations research has
focused on the incentives of political leaders to initiate or participate in wars,38 rather than
military leadership. Two important exceptions warrant discussion here. First, the coup-proofing
literature focuses on the tradeoff that coup-vulnerable political leaders face when choosing be-
tween personally-loyal officers and militarily-competent commanders who can defeat domestic
and foreign enemies, but may also pose a threat to the regime.39
33Gartner 1998; Reiter 2009.34See, e.g., Balch-Lindsay, Enterline, and Joyce 2008; Biddle and Long 2004; Chatagnier and Castelli 2016; Choi 2004;
Cunningham, Gleditsch, and Salehyan 2009; DeRouen and Sobek 2004; Fortna 2012; Lake 1992; Mason, Weingarten,and Fett 1999; Reiter and Stam 1998; Wolford 2007.
35See, e.g., Biddle and Long 2004; Grauer and Horowitz 2012; Reiter and Stam 1998.36Leventoglu and Slantchev 2007.37Biddle 2006; Freedman 2005.38Bueno De Mesquita and Siverson 1995; Chiozza and Goemans 2011; Fuhrmann and Horowitz 2015; Goemans and
Fey 2009; Horowitz and Stam 2014.39Gaub 2013; Hosmer 2007; Pilster and Böhmelt 2011, 2012.
6
Second, Reiter and Wagstaff analyze how battlefield success in World War II affected decisions
to promote or remove commanders.40 They outline how tactical expertise, the ability to inspire
soldiers, the choice of competent subordinates, and the provision of better strategic options
allow leaders to affect battlefield outcomes. Whereas their analysis focuses on the effects of
battle outcomes on military leadership decisions, we examine the inverse: the effect of military
leadership on battle outcomes.
The basic idea that commanders matter is ancient.41 However, while numerous scholars
within international relations have examined the determinants of military victory in both interstate
and intrastate wars,42 measures of commander quality have been conspicuously absent. If the
commander is relevant to battlefield outcomes, then such works neglect a key factor. And while
the existence of a relationship between command and success may seem obvious, its magnitude
and the individual-specific features associated with the effect have not been subject to quantitative
analysis.
A lack of quality data on both battles and the commanders of those battles has been a
constraint on studying the effect of commanders on battle outcomes. Therefore, while we may
accept that leaders matter, we do not know which of their attributes are important. The Civil War,
however, is well documented,43 allowing us to construct a dataset of all the primary battles of
the war, the commanders of those battles, and biographical features for those commanders (see
Sections 4 and 5 of this paper).
With a few recent exceptions,44 the international relations literature has largely ignored the
Civil War. But it is a conflict that generalizes easily to more recent disputes, as it shares several
features with contemporary conflicts. First, the Civil War marks the start of the modern era of
warfare, with the introduction of technologies such as the machine gun, barbed-wire, and trench
40Reiter and Wagstaff 2017.41See, e.g., Plutarch 1992.42See, e.g., Brandt et al. 2008; Desch 2006; Gent 2008; Sullivan 2008; Wayman, Singer, and Goertz 1983.43Weiss 1966.44See, e.g., Poast 2015; Reiter 2009.
7
warfare, making it at least as relevant as World War I. Second, the real GDP per capita of the
United States at the start of the Civil War would make it a middle-income income country in the
present day. Finally, like the plurality of post-Cold War civil wars,45 the Civil War was primarily a
conventional war. Thus, the Civil War has the potential to tell us much about modern warfare.
3 Military Appointments in the American Civil War
By looking at leadership in the Civil War, we also test the generalizability of extant theories of
executive appointments. Whereas early works focused on personal traits, recent research has
examined the tradeoff between loyalty and competence.46,47 Although these works are (largely)
situated within the context of appointments to executive agencies and cabinet departments, there
is reason to believe the negative relationship between loyalty and competence is applicable to
questions of military appointments. This is especially true of the Civil War, the background
of which made loyalty a particularly important criterion.48 Indeed, there is evidence that both
Lincoln and Davis considered questions of loyalty when deciding whom to promote. After General
Don Carlos Buell was relieved for failing to defeat Braxton Bragg’s forces in Kentucky, for example,
Lincoln replaced him with Ohio-born William Rosecrans, skipping over the militarily successful
George H. Thomas, out of a reluctance to “replac[e] one Southern-born commander for another”.49
When Secretary of War Edwin Stanton expressed little confidence in Rosecrans, urging Lincoln to
replace him with Thomas, Lincoln referenced Thomas’s origin, saying, “Let the Virginian wait”,50
illustrating his willingness to trade competence for loyalty.51 Thomas was given command of the
45Kalyvas and Balcells 2010.46See, e.g., Edwards 2001; Hollibaugh, Horton, and Lewis 2014; Krause and O’Connell 2016; Lewis 2008, 2009.47This tradeoff has been found in countries outside of the U.S. as well. See, e.g., Egorov and Sonin 2011; Reuter
and Robertson 2012.48This may be due to appointee characteristics affecting the public’s trust in the government, which can affect
public support of the government during wartime. See Hollibaugh 2016a and Chatagnier 2012, respectively.49Broadwater 2009, p. 87.50Piatt and Van Boynton 1893, p. 327.51Though Rosecrans was a Democrat, this was arguably politically expedient, since his appointment would be
viewed as emblematic of national unity trumping narrow partisan interests. Moreover, he had indicated a willingness
8
Army of the Cumberland at the end of 1863, but only after he prevented Rosecrans’ defeat at
Chickamauga from turning into a disaster.
While loyalty also played an important role in the selection of Confederate commanders, Davis
arguably placed a premium on personal loyalty, rather than birthplace. Indeed, the most senior
officer in the Confederate military—outranking Generals Robert E. Lee and P. G. T. Beauregard—
was New York native Samuel Cooper, who was appointed adjutant general of the Confederate
Army, responsible only to President Davis himself. William Davis argues that Cooper was awarded
this high rank because of his friendship with the Confederate president, and his willingness to do
Davis’ bidding, allowing the latter to “solidify [his] control over his armies. Davis could act through
Cooper, and the rank insulated Cooper from question.”52
These anecdotes suggest that perceptions of loyalty influenced both Confederate and Union
decision making during the war. This does not mean, however, that loyalty was the only—or
even paramount—criterion. Indeed, relative to bureaucratic appointments, the importance of
battlefield competence and the existential threat posed to the Confederate government (as well as
uncertainty over repercussions in the event of capture or a Confederate loss) likely increased the
relative importance of competence.53 As mentioned above, the unique context of the war may
have mitigated any relationship between loyalty and competence. For example, uncertainty over
punishment in the case of a Confederate loss might have ensured that the pool of potential GFO
nominees was disproportionately loyal. In fact, current models rarely consider the pool of potential
to run on a National Union ticket in 1864 as Lincoln’s vice presidential nominee, though the Ohio Republican Partynever received written confirmation of his interest. See Lamers 1999.
52Davis 1996, p. 360.53For more on the political dynamics underlying senior military appointments, see Betts 1978. While most
Confederate officers were ultimately pardoned, several were executed for treason after the war. Additionally, uncertaintyover potential punishments at the war’s outset was high, with prominent members of Congress, like Senator BenjaminF. Wade (R-OH), stating that “the South has got to be punished and the traitors hung [sic].” See Trefousse 1963, p. 148.The passage of the Treason Law of 1862, allowing for fines and/or imprisonment for those convicted of treason and/orassisting rebel forces, likely decreased the expected cost of a loss to commanders, since it provided the judiciarywith an alternative to execution. Moreover, it became clear over the course of the war that the policy of the Lincolnadministration was one of leniency, which was captured in his second inaugural address, where he called for “malicetoward none” and “charity for all.”
9
nominees54 and none, to our knowledge, endogenize self-selection into the pool. Conceivably, the
inclusion of high penalties for failure and self-selection might affect longstanding results.
Additionally, the relationship between loyalty and competence may be mediated by cultural
factors. Indeed, while both the Union and the Confederacy were part of the United States prior
to secession, cultural differences existed between the two—and persist to this day. One of those
most relevant to our analysis is honor, which has long been valorized within the American South.55
Characterizing honor as a reputation for resolve,56 find that Southern presidents have been more
likely to initiate, continue, and win militarized disputes.57 If the same dynamic held during the
Civil War, then those commanders who most identified with the Confederacy (i.e., those most
“loyal” to the Southern cause) likely also identified most closely with Southern honor, and would
be more likely to achieve success on the battlefield due to the desire to maintain a reputation
for unwillingness to back down. This should not hold (at least not to the same extent) for Union
commanders, who generally did not grow up in a culture that placed the same emphasis on honor;
in these cases, we should expect to observe the standard loyalty-competence tradeoff, with loyalty
having neutral—or even negative—effects on outcomes.
Finally, for Confederates with military backgrounds, traits related to competence may be asso-
ciated with increased personal loyalty to Davis. Unlike Lincoln, Davis was a West Point graduate
and Mexican War veteran. As such, he maintained personal connections within the officer corps,
and staffed the government and military with USMA graduates. For Davis, some attributes associ-
ated with competence as an officer—such as training at West Point—might also have been related
to personal loyalty.
Thus, the circumstances involved in military appointments—as well as the characteristics of
the South—could result in a positive association between loyalty and competence for Confederate
54But see Hollibaugh 2015b.55Nisbett and Cohen 1996.56Dafoe and Caughey 2016.57Relatedly, Goemans defines resolve as “the total amount of resources one side is willing to expend”. See Goemans
2000, p. 29
10
commanders, in contrast to the literature on the loyalty-competence tradeoff among civilian
appointees.58 However, these considerations should be absent for Union commanders. In general,
Union commanders neither grew up in the culture of the American South, nor faced the possibility
of execution for treason, as conquest of the Union was not among the goals of the Confederacy.
4 Battle Data
To examine the effect of leadership and the relationship between loyalty and competence for
Civil War commanders, we construct two original datasets on Civil War battles and biographical
information for the commanders thereof. What constitutes a battle is not well defined. For our list
of battles, we rely upon the National Park Service’s Civil War Sites Advisory Commission’s (CWSAC)
list of the more than 300 “principal battles” of the war. These are defined as battles “of special
strategic, tactical, or thematic importance to local operations, campaigns, theaters, or to the war as
a whole”.59 From the CWSAC list, we included all battles between the Battle of Fort Sumter and the
Battle of Appomattox Court House, exclusive. We omit from the list the following types of battles:
those fought between Union forces and Native Americans, primarily naval battles,60 unopposed
captures (such as New Orleans), one-sided violence against civilians (e.g., the Lawrence Massacre),
and battle without a commander in our biographical dataset. This yields a final dataset of 294
battles for our analysis. These observations range from battles engaging armies, where the average
number of personnel on each side is greater than 53,000 (∼7% of battles), such as Gettysburg,
to those at the corps level (greater than 17,000, ∼15%), division level (greater than 5,300, ∼32%),brigade level (greater than 1,700, ∼35%), and regiment level (all others, ∼11%).
58However, whether the mechanism is due to cultivated senses of honor among Southerners (and therefore strongerreputations for resolve), greater uncertainty over possible punishment in the event of a defeat (and therefore higherlevels of resolve due to uncertainty over the expected utility of a loss), or something else entirely, is presentlyindeterminate and beyond the scope of this study.
59CWSAC 1993a,b.60We also omit those battles where the highest-ranking commander in the battle was a naval officer, since this
suggests naval primacy. However, if we omit all battles with any sort of naval component, all results are substantivelyidentical.
11
For each battle, the data comprise several features: its outcome, the strengths (number of
personnel present or engaged) of the forces on both sides, and the casualties on both sides. CWSAC
assigns a military result—one of “Union victory,” “Confederate victory,” or “Inconclusive”—for all
battles in its data.61 This classification generally follows the rule that the victorious side was the
one in control of the battlefield at the conclusion of the battle.62,63 We derive the values of each
side’s strengths (number of military personnel present or engaged in the battle) and casualties
(killed, wounded, missing, and captured) from multiple sources.64 For battles where there are
multiple sources of casualty or strength data, we use the geometric mean of the available sources.
We then estimate the remaining casualty data with multiple imputation.65,66
5 Commander Data
The commanders of each battle are those the CWSAC lists as principal commanders.67 While most
forces have one principal commander, some have up to four.68 For each principal commander, we
used several sources69 to collect biographical data including birthplace and year, length of mili-
tary service, Mexican War service, military academy attendance, experience as planters (plantation
owners), command experience, partisan affiliation, office-holding experience, and familial connec-
tions to notable individuals on both sides. We were not able to collect biographical information
for a minority of principal commanders who were not GFOs, and we dropped battles with these
61CWSAC 1993a,b.62Fox 1898; Livermore 1900.63These classifications are largely consistent with those in Bodart 1908; Fox 1898; Livermore 1900, and CAA 1991.64Bodart 1908; Fox 1898; Kennedy 1998; Livermore 1900; NPS 2012; Phisterer 1883; CAA 1991.65King et al. 2001.66Our data are missing values for Union strength in 6% of battles, Union casualties in 8%, Confederate casualties
in 17%, and Confederate strength in 29%. We impute the missing values as functions of battlefield site, the attackingparty, the date the battle began, who surrendered (if anyone), battle duration, location (measured by theater andcoordinates), battle significance, commander ranks, and outcome. The geometric means of the imputations are usedin the analyses that follow.
67CWSAC 1993a.68Not all principal commanders were generals, since some battles were regiment-level or smaller.69Allardice 2006; Eicher and Eicher 2002; Historical Data Systems 2017; Warner 1959, 1964.
12
commanders from the analysis.70
However, while we expect these biographical variables to be associated with important latent
traits (namely, loyalty and competence) that might drive battlefield results, including them all in
a regression model would likely result in high levels of multicollinearity, precluding our ability
to estimate the effects of individual predictors with any certainty. Moreover, the direction of the
relationship is unclear. That is, some characteristics might affect our latent traits, while others
might be caused by them. To avoid this potential problem—and to examine the effects of “loyalty”
and “competence” per se—we fit a structural equation model to estimate latent dimensions of
loyalty and competence.71 This is consistent with existing work that has sought to estimate these
traits from biographical characteristics.72 Variables used in fitting each trait are described in
Table 1.
Our measure of competence is derived from several variables, which we categorize as either
formative (causing competence) or reflective (caused by latent competence). Our formative indica-
tors are: (1) number of years served in the military; (2) Mexican War service; (3) number of Civil
War battles engaged in to that point; (4) United States Military Academy attendance; (5) other mil-
itary academy attendance; (6) whether they had achieved the grade of midshipman; (7) experience
as an elected official; and (8) whether the commander in question was from a Confederate state.
Variables 1 through 3 capture the influence of military experience, variables 4 through 6 capture
the influence of training, variable (7) accounts for any management experience attained through
other means (as it takes some minimal level of competence to win an election or serve in office),
and variable (8) accounts for residual socialization effects. Our reflective measures are: (1) whether
the individual had experience in command of a corps, army, or division; and (2) highest grade
attained to that point.
70This results in a loss of 13.8% of battles and 33.8% of commanders (almost entirely relegated to non-GFOs).71Typical factor analysis methods assume latent indicators are causally prior to observed indicators. As several of
our indicators are formative, factor analysis is inappropriate.72See, e.g., Krause and O’Connell 2016.
13
Our measure of loyalty is similarly derived. Our formative indicators are: (1) whether the
individual was born in a Confederate state; (2) whether the individual was born in a border
state; (3) whether the individual was born abroad; (4) Lincoln’s vote share in the 1860 presidential
election in the individual’s home state;73 (5 and 6) the difference in the number of prominent
relatives the individual had on each side—as defined by74 and disaggregated by in-laws versus
blood relatives—weighted by the coefficient of relationship,75 which is designed to capture the
closeness of kinship; (7) whether the individual was a planter before the war; and (8) whether they
had attended the United States Military Academy. Variables 1 through 4 capture the influence of
loyalty to the place of birth as well as other socialization effects, variables 5 and 6 capture the
influence of loyalty to family, variable 7 captures the influence of economic incentives (as does
2 to some extent), and variable 8 accounts for any possible loyalties to Jefferson Davis, who had
graduated from West Point in 1828.76 The reflective indicators are: (1) whether the individual
joined the Confederate military; (2) a factor variable capturing the extent of their pro-Confederate
partisan loyalties;77 (3) whether the individual had served as an elected official in a Confederate
state, a Union state, or neither; and (4) whether the individual had attended the Peace Conference
of 1861, in an attempt to avert the war.78
We use all of these variables (as measured at the commander level on the first day of each
battle in which they fought) to estimate a two-factor multiple indicators and multiple causes
73Those born abroad were coded as zero.74Eicher and Eicher 2002.75Wright 1922.76Eicher and Eicher 2002.77This variable was created using principal component analysis on variables indicating whether the person was a
Democrat, Whig, or Republican before the war, or had ever been a member of the Free Soil Party. Note that these arenot necessarily mutually exclusive because the Free Soil indicator can be used in conjunction with any of the otherpartisan indicators, and various party-Free Soil combinations might illustrate different attitudes toward abolitionismand the Confederacy. Additionally, we use principal components analysis (PCA) as opposed to including the variablesin the MIMIC model directly because, individually, the party variables exhibit too little variation and their inclusionresults in convergence problems and issues with model fit. Preprocessing them by fitting a PCA and recovering thefirst factor alleviates this problem and preserves the essential underlying variance.
78Chittenden 1864.
14
(MIMIC) model.79,80
To aid in interpretation, both traits are set to mean zero and variance one. We cluster
by commander and weight each observation by the reciprocal of the number of times each
commander appears in our sample, to so that all commanders are weighted equally and no single
commander (e.g., Robert E. Lee or Ulysses S. Grant) is driving the results. We assume a logistic
link function for dichotomous reflective indicators (Command Experience, Confederate Commander,
and Peace Conference Attendance), and an ordered logistic link function for Top Grade Reached.
The results in Table 2 demonstrate that the indicators described above generally load onto the
loyalty and competence dimensions in the expected ways. Almost all reflective indicators achieve
significance (save for Peace Conference Attendance), and all coefficients show the expected signs.
Where statistically significant, all formative indicators are also signed appropriately: time served
in the military, battle experience (including having fought in the Mexican War), and office-holding
experience all induce higher levels of latent competence; having more family in the Confederacy
versus the Union, being born in the South, being a planter, being a West Point graduate, and being
from an anti-Lincoln state (and the socialization that may entail) all contribute to pro-Confederacy
sentiments. Overall, both sets of indicators behave as expected, with results for reflective indicators
being more in line with our expectations.
Statistically, the model fits well. For both traits, the average variance explained (AVE) is above
0.5, suggesting the observed variables correlate well with each other and sufficiently explain the
latent factors. The squared inter-trait correlation is lower than the AVE for both traits, indicating
that the variables correlate more highly within traits than across traits, and suggesting discriminant
validity. Finally, all common measures of goodness of fit show a high degree of fit: the comparative
fit index (CFI) and Tucker-Lewis index are both greater than 0.95, the weighted root mean square
79Jöreskog and Goldberger 1975.80That variables are measured on the first day of battle is important, as two reflective indicators of competence—
cumulative win percentage and opponent’s cumulative casualty ratio—are measured until (but not including) the firstday of battle. This ensures that when we use these in our later analyses to predict casualties, we are not simplyregressing y on y . Also see Footnote 100.
15
Table1:Description
ofFactorsUsedin
theMIM
ICMod
el
Variable
Variable
Nam
eDescription
Com
peten
ceForm
ative
Military
Service
Num
berof
yearsserved
inthemilitary.
Mexican
War
Experience
Served
during
theMexican
War.
Battle
Experience
Num
berof
CivilWar
battlesengagedin.
UnitedStatesMilitaryAcademy
AttendedtheUnitedStates
Military
Academy.
Military
College
Attendedamilitary
college
(e.g.Th
eCitadel,V
MI,NorwichUniversity).
Midshipman
Achievedthegradeof
Midshipman.
Prew
arElectedOfficial
Served
inelectedoffi
cepriorto
theCivilWar.
Southern
Born
inaseceding
state.
Reflective
Com
mandExperience
Served
ascommanderof
corps,army,or
division
TopGrade
Reached
Highestgradeattained
tothat
point;from
“Non
e”to
“General”(forthe
Arm
yoffi
cers)and“Adm
iral”(forNaval
offices).
LogCum
ulativeWin
Percentage
Logpercentage
ofbattleswon
whenaprincipalcommander.
LogCum
ulativeOpp
onentCasualty
Percentage
Logop
ponent
casualty
ratioin
battleswhenacommanding
officer.
Loyalty
Form
ative
Southern
Born
inaseceding
state.
Border
State
Born
inaborder
state(Delaw
are,Kentucky,M
aryland,
orMissouri).
Foreign
Not
born
intheUnitedStates.
Hom
e-StateLincolnVote
Hom
estate’svote
shareforLincolnin
the1860
presidentialelection
a.
Con
federate
Relatives
Num
berof
prom
inentCon
federate
bloodrelatives,bweightedby
their
coeffi
cientof
relation
ship,cminus
theweightednu
mberof
prom
inent
Union
bloodrelatives.
Con
federate
In-law
sIbid.,bu
tcalculated
usingthenu
mberof
prom
inentCon
federate
and
Union
in-law
s.d
Planter
Planter/Plantation
ownerpriorto
thewar.
UnitedStatesMilitaryAcademy
AttendedtheUnitedStates
Military
Academy.
Reflective
Con
federate
Com
mander
Com
manderin
theCon
federate
military.
Con
federate
Partisan
Bias
Pro-Con
federate
partisan
orientations.e
Con
federate
Office-holding
Bias
1(−
1)ifheld
inan
electedoffi
cein
aCon
federate
(Union
)stateand
neverin
aUnion
(Con
federate)state;
0ifneverheld
electedoffi
ceor
served
inboth
Con
federate
andUnion
states.
PeaceCon
ferenceAttendee
AttendedthePeaceCon
ferenceof
1861.
aZero
forforeign-born
individu
als
b EicherandEicher
2002.
c Wright1922.
dEicher
andEicher
2002.
e The
first
principalcompo
nent
onvariablesforwhether
theindividu
alwas
amem
berof
theDem
ocrats,Whig,
orRepu
blican
partiesbefore
thewar,
orhadever
been
amem
berof
theFree
SoilParty.
Thefactor
isoriented
sothat
“Repub
lican”and“FreeSoil”
have
negative
loadings.AlsoseeFootno
te77.
16
Table 2: A MIMIC Model of Loyalty and Competence
ConfederateCompetence Loyalty
Formative IndicatorsSouthern −0.054 0.427∗∗∗
(0.083) (0.103)United States Military Academy 0.055† 0.073∗
(0.030) (0.032)Military Service 0.357∗∗
(0.043)Battle Experience 0.612∗∗
(0.034)Passed Midshipman 0.007
(0.014)Other Military Academy −0.010
(0.019)Mexican War Experience 0.053∗
(0.022)Prewar Elected Official 0.088∗∗
(0.023)Border State 0.117
(0.072)Foreign −0.058
(0.073)Confederate Bias in Weighted Coefficient of Relationship (Blood Relatives) 0.096∗
(0.046)Confederate Bias in Weighted Coefficient of Relationship (In-Laws) 0.190∗∗
(0.044)Planter 0.129†
(0.071)Home-State Lincoln Vote −0.328∗∗
(0.106)Reflective Indicators
Log Cumulative Win Percentage 0.572∗∗
(0.042)Log Cumulative Opponent Casualty Percentage 0.347∗∗
(0.062)Command Experience 0.788∗∗
(0.038)Top Grade Reached 0.699∗∗
(0.054)Confederate Commander 0.914∗∗
(0.033)Confederate Partisan Bias 0.867∗∗
(0.058)Confederate Office-holding Bias 0.859∗∗
(0.069)Peace Conference Attendance −0.202
(0.177)
Average Variance Explained (AVE) 0.647 0.704Squared Inter-trait Correlation 0.006Comparative Fit Index 0.961Tucker-Lewis Index 0.952Weighted Root Mean Square Residual 0.858Root Mean Square Error of Approximation 0.017Number of Commanders 260Number of Observations 721
Note: Bootstrapped standard errors clustered on commanders in parentheses. Two-tailed tests: ∗∗p < 0.01, ∗p <0.05, †p < 0.1.
17
residual (WRMSR) is less than 0.9, and the root mean squared error of approximation (RMSEA) is
less than 0.06,
Figure 1: Confederate Advantage in Competence Over Time
Brigadier Generals, Colonels, and Commodores
High−Ranking Commanders
1861
−07−
01
1861
−10−
01
1862
−01−
01
1862
−04−
01
1862
−07−
01
1862
−10−
01
1863
−01−
01
1863
−04−
01
1863
−07−
01
1863
−10−
01
1864
−01−
01
1864
−04−
01
1864
−07−
01
1864
−10−
01
1865
−01−
01
1865
−04−
01
0.00
0.25
0.50
0.75
−0.2
0.0
0.2
0.4
Date
Con
fede
rate
Adv
anta
ge in
Com
man
der
Com
pete
nce
As we note above, most contemporary analyses of the war indicate that during the early years
of the war, the Confederacy possessed superior leadership.81 However, as the war progressed,
the Union reduced the Confederacy’s initial advantage, exploiting its own advantage in terms
of resources and manpower.82 Our estimates speak to this understanding. Figure 1 plots the
competence of the active commanders, those who fought in at least one battle and omitting those
who left the military or died, over the course of the war.83 The top panel presents estimates for
81Alexander 2007.82Bond 1998, p. 5.83The line is the LOESS curve through the point estimates. The y-axis indicates the difference in terms of
standard deviations, so a value of 1 at a particular time would indicate that, on average, Confederate commanders
18
two-star officers or higher (those who could command divisions), while the bottom presents the
same for all other commanders, including Brigadier Generals, colonels, and commodores who were
pressed into command without having achieved (or being brevetted into) a GFO grade. In general,
these results are consistent with the common belief that the Confederate States started the war
with an an advantage in military leadership, but the advantage declined over the course of the
war. By the end of the war, the Confederate advantage declined to one-third of its initial value.
Intriguingly, the Confederate States began with a much smaller advantage in the competence of
lower-ranked commanders, which fluctuated over the course of the war.
We can also check the face validity of our loyalty measure.84 To do this, we leverage the
fact that many commanders were deemed “political generals” and arguably appointed primarily
for political purposes. Presumably, these commanders would have higher levels of loyalty, on
average. We examine the relationship between loyalty and being a “political general” by regressing
indicators of political general status on commanders’ mean lifetime-level loyalty scores.85 We use
two indicators of “political general” status—whether a Union commander appeared on86 list of
Lincoln’s “political generals,” and whether a commander appeared on Wikipedia’s list of “political
generals” for the American Civil War.87 In both logistic regression models, the coefficient for loyalty
is positive and significant (βWork ≈ 0.708, σWork ≈ 0.201; βWikipedia ≈ 0.442, σWikipedia ≈ 0.165), in-
dicating that more loyal generals are more likely to be seen as “political generals” in contemporary
times, regardless of which source is used. Figure 2 illustrates this relationship graphically, with
of the indicated grade were about one standard deviation higher on the competence trait measure than Unioncommanders.
84Before estimating the model, we should note that, given the directions of the loadings, the estimated factorfor “loyalty” is better described as Confederate Loyalty. Therefore, we separately re-normalize the estimated positionson this factor for Confederate and Union commanders, multiply these normalized factors by −1 for all Unioncommanders, and re-normalize post hoc the resulting estimates to ensure the new factor has mean zero and varianceone. This renormalized measure will be used in all later analyses.
85An individual’s “lifetime-level” score for a particular trait is defined as the mean of all estimates for a particulartrait-individual combination. This lifetime-level aggregation is used here because the model in Table 2 allows fortrait-level estimates to vary over time.
86Work 2012.87The permanent URL for the Wikipedia page is https://en.wikipedia.org/w/index.php?title=
Political_general&oldid=802825806#American_Civil_War_2.
19
Figure 2(a) showing that the probability of being perceived as a “political general” increases by
about two-to-twelve percentage points when shifting loyalty from one standard deviation below its
mean to one above, and Figure 2(b) showing the distributions of the two.88 Between these results
and those presented in Figure 1, the congruence between our results and the historical narrative
suggests face validity, providing us with confidence that our estimated factors tap into underlying
latent dimensions of loyalty and competence. These factors—or traits—in hand, we proceed to a
series of empirical models.
●Wikipedia (2017)
Work (2012;Union Only)
0.00
0.03
0.06
0.09
Predicted Change in Probabilityof Being a Political General
Sou
rce
List
(a) Effects of Loyalty
●
●
●
●
Work (2012; Union Only) Wikipedia (2017)
−2 0 2 4 6 −2 0 2 4 6
No
Yes
Mean Lifetime Loyalty(Number of Standard Deviations from Mean)
List
ed a
s P
oliti
cal G
ener
al?
(b) Distributions of Loyalty
Figure 2: Loyalty and “Political Generals”
6 Latent Traits and Battlefield Outcomes
Before moving to more sophisticated regression models, we look at the correlations between our
estimated traits. We estimate two models: our “dynamic” model includes correlations between
loyalty and competence for all active officers for each day of the war, and our “initial” model
only includes correlations between loyalty and competence, where each officer’s contribution is
88In Figure 2(a), the thin lines represent the 95% confidence intervals, and the thick lines represent 90% intervals.In Figure 2(b), the dark circles represent the means, and the lines represent the mean +/- one standard deviation.
20
his level of loyalty and competence at the time of attaining the grade in question.89 We do this
because the “dynamic” model—as well as Figure 1—uses indicators of aggregate levels of loyalty
and competence over time, and does not necessarily stem from considerations at the time of
appointment, since it could reflect individual commanders learning over the course of the war.
Thus, the “initial” model is a better measure of the relationships between indicators at the time
that commanders were appointed.
The results in Figure 3 suggest that the presence of the loyalty-competence tradeoff is condi-
tional on side.90 For the Union, the results are consistent with historical accounts of the sustained
use of “political generals”,91 especially for higher-ranking commanders. If the determinants of
Union appointments were partially political in nature, where some generals some selected for
political reasons and others for demonstrated military expertise, then the observed distribution
of commander traits should be similar to those predicted by extant theories of the appointments
process, with a negative relationship between loyalty and competence.
Figure 3: Correlations Between Latent Traits
●
●
●
●
Confederacy Union
−0.4 0.0 0.4 0.8 −0.4 0.0 0.4 0.8
Brigadier Generals, Colonels,and Commodores
High−RankingCommanders
Correlation Between Loyalty and Competence
Offi
cer
Ran
king
Correlation Type
●
Initial
Dynamic
89That is, each commander appears up to twice in our “initial” version—once for the “Higher-Ranking Comman-ders” set, and once for the “Lower-Ranking Commanders” set.
90Spearman’s ρ is presented with 90% (thick line) and 95% (thin line) confidence intervals. “Dynamic” observationsare at the day level.
91Work 2012.
21
For the Confederacy, the results are somewhat different. Indeed, in stark contrast to the
negative correlation among Union commanders, the correlation between loyalty and competence
for the Confederacy’s commanders is positive and significant at the 95% level in our “dynamic”
version, and is significant at the 90% level for lower-ranking commanders. While the effect for
Confederate officers is contrary to existing theories of executive appointments, it is consistent with
the aforementioned idea of Southern honor, which has been documented to affect international
politics in other areas.92
However, these results do not directly reflect performance on the battlefield. Thus, to examine
the relationship between traits and battlefield outcomes, we estimate a series of regression mod-
els. Battlefield success can be measured using either categorical indicators of “victory”,93 or the
relative casualties between sides.94 There are advantages and drawbacks to both. Inflicting greater
casualties does not directly equate to military success, but it is easier to measure. While classifying
the results of battles as victories and defeats may incorporate more information about the context
of the battle, it may also require more subjective decisions. In our analyses we use both victory
and casualties as measures of battlefield success, though the two measures are associated in the
case of our sample.95
Our analyses focus on the relationship between commander characteristics and the realized
battlefield results, whether victory or casualties. We make no claim as to whether civilian leaders
at the time of the Civil War used either metric in making their assessments about who should
be hired, fired, or promoted.96 That said, it is likely that both tactical victories and casualties are
related to such decisions.92Dafoe and Caughey 2016.93Grauer and Horowitz 2012; Reiter and Stam 1998.94Biddle 2006; Biddle and Long 2004.95An ordered logit regression of Battlefield Outcome on the logged loss-exchange ratio predicts 67% of battle
outcomes in-sample. The loss-exchange ratio is the ratio of casualty ratios on each side: casualtiesC S AstrengthC S A
/casualtiesU S AstrengthU S A
. We
add one to each term in order to prevent divide by zero errors. While the loss-exchange ratio is not the exact measureof casualties we use in our later analyses, it is clear that casualties and battle outcomes are reasonably correlated.
96This differs from Gartner 1999, who focuses on assessments of leadership.
22
We first examine Battlefield Outcome. As discussed in Section 4, this is an ordinal variable, for
which Confederate Victory < Inconclusive < Union Victory. For models with Battlefield Outcomes
as the response, we estimate an ordered logistic regression model. Our key independent variables
are the latent traits extracted above. However, our unit of analysis is the battle, and in any given
battle, either side can have multiple commanders.97 Thus, for commander-level latent traits, we
use the battle-level means for the Confederate and Union commanders as covariates, which are
denoted Confederate Competence, Union Competence, Confederate Loyalty, and Union Loyalty.98 In
order to capture the effects of military strength, we include the strengths of each combatant (in
thousands), denoted as Confederate Strength and Union Strength. We control for location, which
affects knowledge of the terrain as well as (potentially) the willingness to fight. The variable
Confederate Battle is an indicator that takes a value of one if the battle took place in a Confederate
state; additionally, we interact it with our Loyalty variables, as Southern honor may matter more
when defending one’s home. Finally, the variable Union Attacker, is an indicator variable for
whether the Union was the attacking side.
Our results are presented in Table 3. Competence has the expected effect. Higher levels of
97The maximum in our data is four.98Using the maxima and minima of the relevant commander-level traits produce results similar to those pre-
sented here (see Appendix). Additionally, to account for the fact that larger battles will likely have greater unit-levelheterogeneity—that is, variance in terms of whether they are regiment-level, division-level, corps-level, or army-level—we also replicate all of our analyses (except those including fixed effects due to small sample sizes and computationalissues) by subsetting the data in terms of whether the total combined forces on each side were above or below themean, and then running separate models on both sets. Generally, the results are in line with those presented in themain paper, with the major exceptions of Union Loyalty increasing Confederate casualty rates and the probabilityof Union victory in smaller battles on Union soil (which is a particularly small set of battles), and the latent traitshaving minimal effect on tactical outcomes in smaller battles. In many other respects, however, the results alignwith those here, and we can be reasonably confident that our results are robust to major differences in battle-levelunit heterogeneity. Finally, to account for the possibility that Southern honor might translate differently dependingon whether the individual in question is fighting for the Union or the Confederacy, we estimate a series of modelswhere we account for the mean level of loyalty for “crossover” commanders (see Appendix). Our results suggest thatmore loyal “crossover” Confederate commanders suffer proportionally fewer casualties, and more loyal “crossover”Union commanders are less likely to win battles on Union territory. The former result might be explained due to theConfederate commanders in question seeking to prove themselves to others for whom honor is highly valued (e.g.,other Southerners), whereas the latter result is likely a statistical artifact (there is only one instance in our data of a“crossover” Union commander leading the charge in Union territory—Virginia-born George H. Thomas at the Battle ofMill Springs, a Union victory). This result aside, “crossover” Union generals seem to be indistinguishable from thosenot born in the Confederacy.
23
Table 3: Latent Traits and Battle Outcomes
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6
Confederate Commander Competence −0.385∗∗ −0.352∗ −0.352∗ −0.344∗ −0.368∗∗ −0.179(0.132) (0.139) (0.139) (0.140) (0.142) (0.203)
Union Commander Competence 0.628∗∗ 0.574∗ 0.605∗∗ 0.602∗∗ 0.641∗∗ 0.342(0.210) (0.230) (0.232) (0.232) (0.234) (0.286)
Confederate Commander Loyalty −0.105 −0.101 −0.098 −0.045 −0.032 0.104(0.116) (0.117) (0.117) (0.246) (0.248) (0.273)
Union Commander Loyalty −0.058 −0.051 −0.045 −0.205 −0.209 −0.281(0.112) (0.113) (0.113) (0.289) (0.290) (0.318)
Confederate Strength −0.043∗ −0.044∗ −0.043∗ −0.051∗∗ −0.057∗
(0.019) (0.019) (0.019) (0.020) (0.024)Union Strength 0.022∗ 0.023∗ 0.023∗ 0.028∗ 0.028∗
(0.011) (0.011) (0.011) (0.012) (0.014)Confederate Battle −0.329 −0.324 −0.346 −0.118
(0.292) (0.294) (0.294) (0.341)Confederate Commander Loyalty × Confederate Battle −0.094 −0.104 0.021
(0.279) (0.281) (0.311)Union Commander Loyalty × Confederate Battle 0.203 0.208 0.294
(0.315) (0.316) (0.350)Union Attacker −0.317 −0.402
(0.252) (0.281)Cutpoint 1 −1.043∗∗ −1.050∗∗ −1.312∗∗ −1.309∗∗ −1.503∗∗ −1.579∗∗
(0.147) (0.190) (0.303) (0.305) (0.343) (0.438)Cutpoint 2 −0.167 −0.158 −0.418 −0.412 −0.602† −0.570
(0.134) (0.180) (0.293) (0.296) (0.334) (0.428)
AIC 606.613 604.887 605.604 608.868 609.277 622.299BIC 628.715 634.356 638.756 649.388 653.480 765.959Log Likelihood −297.307 −294.444 −293.802 −293.434 −292.638 −272.150Likelihood Ratio Test 12.964∗ 18.690∗∗ 19.973∗∗ 20.709∗ 22.300∗ 63.278∗∗
Number of Observations 294 294 294 294 294 294
Note: Ordered logit coefficients. Observations are at the battle level. The dependent variable is an ordered factor with the levels being, inorder, Confederate Victory, Inconclusive, and Union Victory. Negative [positive] coefficients indicate the covariate is associated with higherprobabilities of Confederate [Union] victories. Model 6 includes commander-level fixed effects for those who fought in at least seven battles.Standard errors in parentheses. Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
competence among Confederate [Union] generals are associated with higher probabilities of stale-
mates or Confederate [Union] victories. Additionally, higher levels of Confederate [Union] strength
are associated with higher probabilities of Confederate [Union] victories (or stalemates). These
dynamics are illustrated in Figure 4, which shows the expected relationship between competence
and outcomes—more competent commanders are associated with greater probabilities of victory
for their respective sides. In this figure, we examine how the probabilities of each outcome change
as we move commander traits from one standard deviation below the combatant-specific mean to
24
one above.99 Increasing the competence of Confederate [Union] commanders from one standard
deviation below the mean to one above increases the probability of victory by about 4–32 [7–23]
percentage points, and decreases the probability of defeat by about 4–24 [7–34] percentage points.
Figure 4: Competence and the Probability of Battle Outcomes
●
●
●
●
●
●
Confederate Victory Inconclusive Union Victory
−0.2 0.
00.
2−0
.2 0.0
0.2
−0.2 0.
00.
2
Competence
Loyalty
Predicted Change in Probability of Outcome
Trai
t
Combatant
●
Confederacy
Union
Interestingly, we note that loyalty has no effect on the probability of battlefield success in
any model. This could mean that loyalty does not matter at all when it comes to battlefield
performance, or it might be that loyalty plays a role apart from determining tactical success on
the battlefield. Instead, it might affect how commanders fight on the battlefield. To unpack
this, we examine how loyalty and competence affect the ratio of the combatants’ casualties.100
While this does not translate into victory per se, it is not dependent upon subjective classification.
Furthermore, it distinguishes the magnitude of victory.101 Therefore, we estimate a series of quasi-
binomial regression models (with logistic link functions) with the same independent variables
99To calculate the effects in Figure 4, we set all continuous independent variables to their mean values and allowcompetence and loyalty to vary. Our predictions, which include 90% (thick line) and 95% (thin line) confidenceintervals, are based on the coefficient estimates in Model 1, as it had the lowest BIC of those models estimated.
100Importantly, the casualties used in Table 4 are not the same as those used in the estimation of the latent traits.The latent traits are estimated at the outset of each battle, while the casualties used as the dependent variable hereare those obtained over the course of the battle. As such, there is no overlap, and we are not merely regressing y ona function of y . At best, the recovered latent traits can be considered to be a function of a lagged dependent variable,albeit dependent on several other covariates not included in the models in Table 4.
101Biddle and Long 2004, p. 535.
25
used in Table 3. The dependent variables in all analyses are the proportion of casualties from the
Confederacy.102
Table 4: Latent Traits and Relative Confederate Casualty Rates
Model 7 Model 8 Model 9 Model 10 Model 11 Model 12
Confederate Commander Competence −0.217∗∗ −0.235∗∗ −0.245∗∗ −0.243∗∗ −0.245∗∗ −0.122(0.048) (0.055) (0.056) (0.056) (0.056) (0.085)
Union Commander Competence 0.213∗∗ 0.211∗∗ 0.252∗∗ 0.247∗∗ 0.239∗∗ 0.191(0.078) (0.080) (0.084) (0.084) (0.085) (0.124)
Confederate Commander Loyalty −0.255∗∗ −0.254∗∗ −0.256∗∗ 0.108 0.102 0.261(0.053) (0.054) (0.054) (0.213) (0.213) (0.252)
Union Commander Loyalty −0.064 −0.053 −0.048 −0.005 −0.008 0.074(0.054) (0.056) (0.056) (0.160) (0.160) (0.197)
Confederate Strength 0.001 0.001 0.000 0.002 0.009(0.004) (0.004) (0.004) (0.004) (0.006)
Union Strength 0.000 0.000 0.000 −0.001 −0.002(0.003) (0.003) (0.003) (0.003) (0.004)
Confederate Battle −0.210 −0.107 −0.112 −0.186(0.129) (0.144) (0.144) (0.188)
Confederate Commander Loyalty × Confederate Battle −0.384† −0.378† −0.422†
(0.217) (0.217) (0.248)Union Commander Loyalty × Confederate Battle −0.047 −0.047 0.001
(0.171) (0.171) (0.222)Union Attacker 0.072 0.073
(0.114) (0.122)Constant 0.160∗∗ 0.118 0.309∗ 0.229 0.195 0.326
(0.058) (0.079) (0.142) (0.151) (0.161) (0.218)
Quasi-AIC 305.965 307.940 308.965 310.980 311.969 339.305Wald Test 76.259∗∗ 76.296∗∗ 78.790∗∗ 81.746∗∗ 81.848∗∗ 128.819∗∗
Number of Observations 294 294 294 294 294 294
Note: Quasi-binomial logistic coefficients. Observations are at the battle level. The dependent variable is the proportion of battlecasualties from the Confederacy. Positive [negative] coefficients indicate that the covariate is associated with higher [lower] ratios ofConfederate casualties. Model 12 includes commander-level fixed effects for those who fought in at least seven battles. Standarderrors in parentheses. Two-tailed tests: ∗∗∗p < 0.01, ∗∗p < 0.05, †p < 0.1.
The results presented in Table 4 are more varied than those in Table 3, but tell the same story
102We use the binomial regression framework as these are count models where the maximum possible count isknown, and thus the dependent variable is the fraction of the possible maximum count. This allows the estimatorto make use of the variation in the maximum count, which allows for more precise estimates than standard countmodels while accounting for the heterogeneity in total casualties. In our case, the maximum casualty count is the sumof the casualties on both sides, and thus our analysis can be construed as measuring the relative Confederate casualtyrate. We estimate quasi-binomial models because of overdispersion. This is analogous to using a negative binomialmodel instead of a Poisson model when using count data where the maximum count is unknown or undefined; inpractice, however, our model is simply a weighted logit where the dependent variable is the battle-level proportionof casualties from the Confederacy, with each observation weighted by the total number of casualties (both Unionand Confederate). Note that we maintain the Confederate Strength and Union Strength independent variables in partbecause the number of possible casualties is limited by the strength brought to the battle.
26
in terms of competence, as Confederate [Union] Competence, where significant, is associated with
decreased [increased] rates of Confederate casualties. The effect of Loyalty is more ambiguous: the
results suggest that more loyal Confederate commanders generally suffer lower rates of Confederate
casualties, while loyalty has little impact for Union commanders.
That said, our model is a quasi-binomial model, and direct interpretation of the results might
be difficult. Therefore, we present changes in predicted Confederate casualty rates, varying mean
competence and loyalty. In particular, we examine how the proportion of casualties from the
Confederate side changes as we move commander traits from one standard deviation below the
combatant-specific mean to one above. Results are presented in Figure 5, which includes 90%
(thick line) and 95% (thin line) confidence intervals.103
Figure 5: Commander Traits and Predicted Effects on Confederate Versus Union Casualties
●
●
Competence
Loyalty
−0.1 0.0 0.1
Predicted Change in Proportion of Casualties from Confederate Side
Trai
t
Combatant
●
Confederacy
Union
Figure 5 illustrates dynamics similar to those presented in Figure 4, with more competent
commanders associated with better outcomes. Generally, the effect is consistent with the results
from the outcome models. The competence of Confederate [Union] commanders is associated with
lower ratios of Confederate to Union [Union to Confederate] casualties. More specifically, a shift
from one standard deviation below the mean to one standard deviation above the mean in the103We use the coefficient estimates from Model 7, as it had the lowest Quasi-AIC of those estimated. As with
Figure 4, we set continuous variables to their means.
27
competence of Confederate [Union] commanders reduces [increases] the proportion of casualties
from the Confederate side by about 4–14 [3–13] percentage points. The plot suggests that the
effect of loyalty is strong, but only apparent for Confederate commanders, and is associated with
better outcomes for the Confederacy. These results support the mechanism of Southern honor
for Confederate commanders, with Confederate soldiers fighting harder since most battles take
place at home.104 Alternatively, it may indicate greater familiarity with the terrain, as Models 10–12
suggest the loyalty effect is limited to Confederate battlefields (though those models exhibit poorer
fit than the more parsimonious Model 7).
Overall, these results reinforce our findings from Figure 3, Table 3, and Figure 4. Competence
is associated with better outcomes, regardless of how it is measured. Loyalty is associated with
lower casualty ratios for the Confederacy, and appears irrelevant to the Union, despite the pres-
ence of the loyalty-competence tradeoff implied by Figure 3. Broadly speaking, the Confederate
results are consistent with theories of resolve and Southern honor discussed earlier, whereas the
Union results are more consistent with traditional theories of executive appointments, which posit
tradeoffs between loyalty and competence. Collectively, the results challenge the generalizability
of standard theories of executive appointments. Moreover, they should motivate scholars to bet-
ter conceptualize the linkages between individual characteristics and organizational performance,
since, despite the presence of a tradeoff for the Union, we do not generally observe worse Union
performance when loyalty is high.
7 The Importance of Leadership: Generals Lee and Grant
Finally, we illustrate our results by applying them to the case Generals Ulysses S. Grant and Robert
E. Lee. With our model, we estimate the leadership effect of the two generals, by comparing
the observed results of the battles which they commanded to the counterfactual of an “average”
104Recall that the dependent variable is effectively a proportion, so causing comparatively more Union casualtieswould serve to drive down the proportion of Confederate casualties.
28
commanders.105 To do this, we simulate 10,000 sets of predicted probabilities from Model 5 in
Table 3 using the observed values; this gives us distributions of possible battle outcomes under
Grant and Lee’s leadership. We then replace their traits with those of the average commander and
simulate another 10,000 sets of predicted probabilities. We plotted the distributions in Figures 6
and 7.106
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Change in Probabilityof Confederate Victory
Change in Probabilityof Inconclusive Outcome
Change in Probabilityof Union Victory
−0.6
−0.4
−0.2 0.
0−0
.1 0.0
0.1
0.0
0.2
0.4
0.6
Cumberland ChurchThird PetersburgWhite Oak Road
Fort StedmanDarbytown & New Market Roads
Chaffin's FarmGlobe Tavern
Second Deep BottomCrater
Jerusalem Plank RoadSecond Petersburg
Cold HarborTotopotomoy Creek
North AnnaSpotsylvania Court House
WildernessMine Run
Second Rappahannock StationWilliamsport
GettysburgSalem Church
ChancellorsvilleFredericksburg
AntietamSouth Mountain
Second Bull RunMalvern Hill
Glendale/White Oak SwampGaines's Mill
Beaver Dam CreekOak Grove
Cheat Mountain
Effect of Replacing General Lee with an"Average" Confederate Commander
Bat
tle
Figure 6: Outcome Probabilities Under General Lee and Average Confederate Commanders
Looking first at Lee’s results in Figure 6, we note that in most cases, replacing General Lee
with an average Confederate commander would have proved disastrous for the rebel cause. In
105An average commander is defined as commander with mean lifetime-level traits.106For Figures 6 and 7, the estimates are the median changes in predicted probabilities for each outcome over
10,000 simulations, assuming Confederate forces commanded by General Lee or Grant were instead commanded bycommanders from the same side who possessed average traits. We include 90% (thick line) and 95% (thin line)confidence intervals.
29
most cases, it would have resulted in a 10–40-percentage-point decrease in the probability of a
Confederate victory, with commensurate increases in the probabilities of Union victories. Results
for General Grant (Figure 7) illustrate similar dynamics.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Change in Probabilityof Confederate Victory
Change in Probabilityof Inconclusive Outcome
Change in Probabilityof Union Victory
−0.3 0.
00.
30.
6−0
.3 0.0
0.3
0.6
−0.3 0.
00.
30.
6
Third PetersburgJerusalem Plank Road
Second PetersburgCold Harbor
Totopotomoy CreekNorth Anna
Spotsylvania Court HouseWilderness
Missionary RidgeVicksburg
Champion HillJackson, Mississippi
Port GibsonShiloh
Fort DonelsonFort Henry
Belmont
Effect of Replacing General Grant with an"Average" Union Commander
Bat
tle
Figure 7: Outcome Probabilities Under General Grant and Average Union Commanders
We can unpack these results further by examining the results of two of the most consequential
battles of the war. For Lee, we used the Confederate victory at Fredericksburg. For Grant, we used
the Union victory at the Third Battle of Petersburg in 1865, which ended the Siege of Petersburg
and immediately led to the capture of Richmond. Ternary plots of simulated outcomes for these
battles are presented in Figure 8. Each point represents one set of predicted probabilities, and the
rings represent the 50%, 90%, and 95% confidence intervals.
Figure 8 suggests that leadership was critical to Confederate and Union fortunes at Fredericks-
burg and the Third Battle of Petersburg, respectively. At Fredericksburg, under Lee, the predicted
probability of a Confederate victory was about 35.6% (95% CI: [27.1%, 44.9%]); under an average
Confederate commander, the probability of a Confederate victory would have been about seven-
teen percentage points lower, at 18.6% (95% CI: [12.4%, 26.6%]). The results for Grant at the Third
30
Figure 8: Simulated Outcomes Under Lee, Grant, and Average Commanders
20
40
60
80
100
20
40
60
80
100
20 40 60 80 100
Inconclusive
CSA USA
20
40
60
80
100
20
40
60
80
100
20 40 60 80 100
Inconclusive
CSA USA
20
40
60
80
100
20
40
60
80
100
20 40 60 80 100
Inconclusive
CSA USA
20
40
60
80
100
20
40
60
80
100
20 40 60 80 100
Inconclusive
CSA USA
Lee at Fredericksburg Grant at Third PetersburgO
bservedC
ounterfactual
Battle of Petersburg are similar and suggest he might have helped to turn a probable Confederate
victory into a likely Union triumph. Under an average Union commander, the predicted probability
of a Confederate victory would have been about 79.4% (95% CI: [49.8%, 93.8%]), with only a 9.7%
chance (95% CI: [2.8%, 28.4%]) of a Union victory. However, under Grant, the Confederate win
probability declined to 36.1% (95% CI: [16.9%, 61.4%]), and the Union win probability increased to
42.2% (95% CI: [19.7%, 68.0%]), suggesting a more likely Union victory (and the resulting Union
victory was critical to the war’s outcome).
Overall, these results reinforce our previous finding on the importance of leadership, and
31
illustrate that different leadership during critical stages of the war might have had far-reaching
consequences for its ultimate outcome.
8 Discussion and Conclusion
Military leadership matters. This dictum is oft-heard but rarely verified. While previous scholarship
has examined leadership in particular cases, it has not sought broader evidence for the claim. By
obtaining commander-level data and identifying traits relevant to leadership quality, we conducted
one of the first large-scale quantitative studies of commander quality in combat.107 Our analysis
suggests a number of empirically-observable factors that underlie leader competence, and shows
that higher-quality leaders generate better outcomes on the battlefield. That our results comport
with findings of in-depth qualitative studies of the conflict provides face validity.
We also find a surprising relationship between loyalty and battlefield outcomes. In particular,
we uncover a negative association between Union loyalty and competence (at least for high-
ranking generals)—in line with extant theories of the loyalty-competence tradeoff—but a positive
relationship between Confederate loyalty and competence. Moreover, we find no relationship be-
tween Union loyalty and same-side casualty rates, and a negative relationship between Confederate
loyalty and the same. Given the dominant paradigm, this is unexpected and suggests that our
understanding of executive appointments might need to be refined. Indeed, there might exist
situations where the tradeoff might be turned on its head, with loyalty positively associated with
competence. While we cannot ascertain the the mechanism at this point, scholars of executive
appointments might consider self-selection into the pool of potential nominees, large penalties for
failure to implement particular policy that bring a de facto end to the policy, and outside career
options of the responsible agent and associated reputational effects. Further exploration would
shed more light on the mechanisms driving our results, as well as the appointments process more
107But see also (Reiter and Wagstaff 2017).
32
generally.
Our results point to two additional paths for future work. First, the techniques we used
to analyze the relationships between loyalty, competence, and victory can be applied to other
situations, given data availability. It would be useful to explore other conflicts, to ensure that
our results are not an aberration, and that leadership matters in other disputes. We encourage
scholars with similar data on other conflicts to conduct such analyses. A natural starting point
might be World War II, using the data analyzed by Reiter and Wagstaff. In fact, explaining
the outcome of the Ardennes campaign has been a focus of much of the operations research
literature, though such studies have neglected the role of military leadership.108 Bringing in this
crucial variable should help to improve those explanations. However, our technique is equally
applicable to internal conflicts. Indeed, analyses of the loyalty and competence levels of officers
appointed to the upper echelons of a country’s military can tell us much about the concerns of
the country’s leadership—whether threats are expected to be internal or external—and whether
the leader is engaging in coup-proofing. This can inform us about the leader’s motivations, as
well as the degree to which power—usually measured in terms of raw material capabilities—can
be leveraged in international conflict bargaining scenarios.
Second, though we note the importance of individual commanders in this article, we omit
consideration of the political contexts surrounding their appointments. Given our findings regard-
ing the seeming breakdown of the loyalty-competence tradeoff for Confederate commanders and
the irrelevance of it for Union commanders, it would be useful to examine which commander
qualities were emphasized during the initial appointments process. This is of particular interest
because, unlike other many studies of executive appointments, the pool of potential nominees to
GFO status is reasonably well-defined for both sides, and likely consist of lower-ranking GFOs and
other high-ranking officers.109 Moreover, such an analysis would illuminate whether our aggre-
108See, e.g., Bracken 1995; Fricker 1998; Lucas and Turkes 2004.109But see Hollibaugh and Rothenberg 2017.
33
gate findings here about the loyalty-competence tradeoff generally hold for all appointments, or
whether the relative emphases on each changed during the war, along with the perceived chances
of victory. This could allow scholars to determine the extent to which experiential learning110
influenced the presidents’ appointment decisions.
More generally, while providing additional context to the findings here, examinations of mil-
itary appointments would help scholars of executive politics to understand how nominee pool
constraints affect the appointments process more broadly. In the present case, since our findings
point to differences in the relationship between loyalty and competence for the two sides, we
surmise that pools of potential GFOs may have been constrained in different ways for the two
combatants.111 Given the different roles played by generals and civilian bureaucrats, it would be
interesting to determine whether appointment politics stop at the water’s edge.
110See, e.g., Krause and O’Connell 2016.111Notably, our results suggest there may be cases where the loyalty-competence tradeoff, while present, might be
irrelevant for organizational performance. While we find evidence of a tradeoff among Union commanders, we findno evidence of loyalty negatively affecting battle outcomes for the Union. Whether this is unique to the AmericanCivil War or holds more broadly is yet to be determined.
34
References
Alexander, Bevin (2007). How the South could Have Won the Civil War: The Fatal Errors that Led to
Confederate Defeat. New York: Crown Publishers.
Allardice, Bruce S. (2006). More Generals in Gray. Baton Rouge: Louisiana State University Press.
Andreatta, Filippo (2015). Potere militare e arte della guerra. Trento, Italy: FBK Press.
Balch-Lindsay, Dylan, Andrew J. Enterline, and Kyle A. Joyce (2008). “Third-Party Intervention and
the Civil War Process”. In: Journal of Peace Research 45.3, pp. 345–363.
Betts, Richard (1978). Soldiers, Statesmen, and Cold War Crises. Cambridge: Harvard University Press.
Biddle, Stephen (2006). Military Power: Explaining Victory and Defeat in Modern Battle. English.
Princeton: Princeton University Press.
Biddle, Stephen and Stephen Long (2004). “Democracy and Military Effectiveness: A Deeper Look”.
en. In: Journal of Conflict Resolution 48.4, pp. 525–546.
Bodart, Gaston (1908). Militär-historisches Kriegs-Lexikon (1618-1905). ger. Vienna: C. W. Stern.
Bond, Brian (1998). The Pursuit of Victory: From Napoleon to Saddam Hussein. Oxford: Oxford
University Press.
Bracken, Jerome (1995). “Lanchester Models of the Ardennes Campaign”. In: Naval Research Logistics
(NRL) 42.4, pp. 559–577.
Brandt, Patrick T. et al. (2008). “When and How the Fighting Stops: Explaining the Duration and
Outcome of Civil Wars”. In: Defence and Peace Economics 19.6, pp. 415–434.
Broadwater, Robert P. (2009). General George H. Thomas: A Biography of the Union’s “Rock of Chicka-
mauga”. Jefferson, NC: McFarland.
Bueno De Mesquita, Bruce and Randolph M. Siverson (1995). “War And the Survival of Political
Leaders: A Comparative Study of Regime Types and Political Accountability”. In: American
Political Science Review 89.4, pp. 841–855.
35
Chatagnier, J. Tyson (2012). “The effect of trust in government on rallies ’round the flag”. In: Journal
of Peace Research 49.5, pp. 631–645.
Chatagnier, J. Tyson and Emanuele Castelli (2016). “The Arc of Modernization: Economic Structure,
Materialism, and the Onset of Civil Conflict”. In: Political Science Research & Methods. Advance
Online Publication. doi: http://doi.org/10.1017/psrm.2016.24.
Chiozza, Giacomo and H. E. Goemans (2011). Leaders and International Conflict. Cambridge: Cam-
bridge University Press.
Chittenden, L. E. (1864). Report of the Debates and Proceedings in the Secret Session of the Con-
ference Convention, for Proposing Amendments to the Constitution of the United States, held at
Washington, D.C., in February, A.D. 1861. New York: Appleton & Company.
Choi, Ajin (2004). “Democratic Synergy and Victory in War, 1816–1992”. In: International Studies
Quarterly 48.3, pp. 663–682.
Chun, Young Han and Hal G. Rainey (2005a). “Goal Ambiguity and Organizational Performance in
U.S. Federal Agencies”. In: Journal of Public Administration Research and Theory 15.4, pp. 529–
557.
— (2005b). “Goal Ambiguity in U.S. Federal Agencies”. In: Journal of Public Administration Research
and Theory 15.1, pp. 1–30.
Cullum, George W. (1879). Biographical Register of the Officers and Graduates of the U.S. Military
Academy at West Point. 2d. Vol. I. New York: D. Van Nostrand.
Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan (2009). “It Takes Two: A
Dyadic Analysis of Civil War Duration and Outcome”. In: Journal of Conflict Resolution 53.4,
pp. 570–597.
Dafoe, Allan and Devin Caughey (2016). “Honor and War: Southern U.S. Presidents and the Effects
of Concern for Reputation”. In: World Politics 68.2, pp. 341–381.
Davis, William C. (1996). Jefferson Davis: The Man and His Hour. Baton Rouge: Louisiana State
University Press.
36
DeRouen, Karl R. and David Sobek (2004). “The Dynamics of Civil War Duration and Outcome”.
In: Journal of Peace Research 41.3, pp. 303–320.
Desch, Michael C. (2006). “Democracy and Victory: Fair Fights or Food Fights?” In: Democracy 28.1,
pp. 180–194.
Edwards III, George C. (2001). “Why Not the Best? The Loyalty-Competence Trade-off in Presidential
Appointments”. In: The Brookings Review 19.2, pp. 12–16.
Egorov, Georgy and Konstantin Sonin (2011). “Dictators And Their Viziers: Endogenizing The
Loyalty-Competence Trade-Off”. In: Journal of the European Economic Association 9.5, pp. 903–
930.
Eicher, John H. and David J. Eicher (2002). Civil War High Commands. English. Palo Alto: Stanford
University Press.
Fortna, Virginia Page (2012). “Do Terrorists Win? Rebels’ Use of Terrorism and Civil War Outcomes”.
In: International Organization 69.3, pp. 1–38.
Fox, William Freeman (1898). Regimental Losses in the American Civil War, 1861–1865. A treatise on
the extent and nature of the mortuary losses in the Union regiments, with full and exhaustive
statistics compiled from the official records on file in the state military bureaus and at Washington.
Albany: Albany Publishing Company.
Freedman, Lawrence (2005). “A Theory of Battle or a Theory of War?” In: Journal of Strategic Studies
28.3, pp. 425–435.
Fricker, Ronald D (1998). “Attrition Models of the Ardennes Campaign”. In: Naval Research Logistics
45.1, pp. 1–22.
Fuhrmann, Matthew and Michael C. Horowitz (2015). “When Leaders Matter: Rebel Experience and
Nuclear Proliferation”. In: Journal of Politics 77.1, pp. 72–87.
Gartner, Scott Sigmund (1998). “Opening Up the Black Box of War”. en. In: Journal of Conflict
Resolution 42.3, pp. 252–258.
— (1999). Strategic Assessment in War. Yale University Press.
37
Gaub, Florence (2013). “The Libyan Armed Forces Between Coup-Proofing and Repression”. In:
Journal of Strategic Studies 36.2, pp. 221–244.
Gent, Stephen E. (2008). “Going in When It Counts: Military Intervention and the Outcome of Civil
Conflicts”. In: International Studies Quarterly 52.4, pp. 713–735.
Gilmour, John B. and David E. Lewis (2005). “Assessing Performance Budgeting at OMB: The
Influence of Politics, Performance, and Program Size”. In: Journal of Public Administration
Research and Theory 16.2, pp. 169–186.
Goemans, H. E. (2000). War and Punishment: The Causes of War Termination and the First World
War. Princeton: Princeton University Press.
Goemans, H. E. and Mark Fey (2009). “Risky but Rational: War as an Institutionally Induced
Gamble”. In: Journal of Politics 71.01, pp. 35–54.
Goss, Thomas Joseph (2003). The War within the Union High Command: Politics and Generalship
during the Civil War. Lawrence, KS: University Press of Kansas.
Grauer, Ryan and Michael C. Horowitz (2012). “What Determines Military Victory? Testing the
Modern System”. In: Security Studies 21.1, pp. 83–112.
Haughton, Andrew R. B. (2000). Training, Tactics and Leadership in the Confederate Army of Ten-
nessee: Seeds of Failure. New York: Routledge.
Historical Data Systems (2017). American Civil War Research Database.
Hollibaugh Jr., Gary E. (2015a). “Naïve Cronyism and Neutral Competence: Patronage, Performance,
and Policy Agreement in Executive Appointments”. In: Journal of Public Administration Research
and Theory 25.2, pp. 341–372.
— (2015b). “Vacancies, vetting, and votes: A unified dynamic model of the appointments process”.
In: Journal of Theoretical Politics 27.2, pp. 206–236.
— (2016a). “Presidential Appointments and Public Trust”. In: Presidential Studies Quarterly 46.3,
pp. 618–639.
38
Hollibaugh Jr., Gary E. (2016b). “The Incompetence Trap: The (Conditional) Irrelevance of Agency
Expertise”. In: Journal of Public Administration Research and Theory. 27.2, pp. 217–235.
Hollibaugh Jr., Gary E., Gabriel Horton, and David E. Lewis (2014). “Presidents and Patronage”. In:
American Journal of Political Science 58.4, pp. 1024–1042.
Hollibaugh Jr., Gary E. and Lawrence S. Rothenberg (2017). “The When and Why of Nominations:
Determinants of Presidential Appointments”. In: American Politics Research 45.2, pp. 280–303.
Horowitz, Michael C. and Allan C. Stam (2014). “How Prior Military Experience Influences the
Future Militarized Behavior of Leaders”. In: International Organization 68.3, pp. 527–559.
Hosmer, Stephen T. (2007). Why The Iraqi Resistance to the Coalition Invasion Was So Weak. Santa
Monica, CA: Rand Corporation.
Jöreskog, Karl G. and Arthur S. Goldberger (1975). “Estimation of a Model with Multiple Indica-
tors and Multiple Causes of a Single Latent Variable”. In: Journal of the American Statistical
Association 70.351a, pp. 631–639.
Kalyvas, Stathis N. and Laia Balcells (2010). “International System and Technologies of Rebellion:
How the End of the Cold War Shaped Internal Conflict”. In: American Political Science Review
104.03, pp. 415–429.
Kennedy, Frances H. (1998). The Civil War Battlefield Guide. Boston: Houghton Mifflin.
King, Gary et al. (2001). “Analyzing Incomplete Political Science Data: An Alternative Algorithm for
Multiple Imputation”. English. In: American Political Science Review 95.1, pp. 49–69.
Klingler, Jonathan D. and J. Tyson Chatagnier (2014). “Are You Doing Your Part? Veterans’ Political
Attitudes and Heinlein’s Conception of Citizenship”. In: Armed Forces & Society 40.4, pp. 673–
695.
Krause, George A. and Anne Joseph O’Connell (2016). “Experiential Learning and Presidential
Management of the US Federal Bureaucracy: Logic and Evidence from Agency Leadership
Appointments”. In: American Journal of Political Science 60.4, pp. 914–931.
39
Lake, David A. (1992). “Powerful Pacifists: Democratic States and War”. In: American Political Science
Review 86.1, pp. 24–37.
Lamers, William M. (1999). The Edge of Glory: A Biography of General William S. Rosecrans, U.S.A.
Baton Rouge: Louisiana State University Press.
Leventoglu, Bahar and Branislav L. Slantchev (2007). “The Armed Peace: A Punctuated Equilibrium
Theory of War”. In: American Journal of Political Science 51.4, pp. 755–771.
Lewis, David E (2007). “Testing Pendleton’s Premise: Do Political Appointees Make Worse Bureau-
crats?” In: Journal of Politics 69.4, pp. 1073–1088.
Lewis, David E. (2008). The Politics of Presidential Appointments: Political Control and Bureaucratic
Berformance. Princeton: Princeton University Press.
— (2009). “Revisiting the Administrative Presidency: Policy, Patronage, and Agency Competence”.
In: Presidential Studies Quarterly 39.1, pp. 60–73.
Liddell Hart, Basil Henry (1996). Great Captains Unveiled. London: Da Capo Press.
Livermore, Thomas Leonard (1900). Number and Losses in the Civil War in America, 1861–65. Boston:
Houghton Mifflin.
Lucas, Thomas W. and Turker Turkes (2004). “Fitting Lanchester Equations to the Battles of Kursk
and Ardennes”. In: Naval Research Logistics (NRL) 51.1, pp. 95–116.
Mason, T. David, Joseph P. Weingarten, and Patrick J. Fett (1999). “Win, Lose, or Draw: Predicting
the Outcome of Civil Wars”. In: Political Research Quarterly 52.2, pp. 239–268.
McClellan, George B. (1887). McClellan’s Own Story: The War for the Union, the Soldiers who Fought It,
the Civilians who Directed it and His Relations to It and to Them. New York: Charles L. Webster
& Company.
McPherson, J. M. (1988). Battle Cry of Freedom: The Civil War Era. Oxford History of the United
States. New York: Oxford University Press.
Mills, Anson (1918). My Story. Washington, DC: Press of Byron S. Adams.
40
Moe, Terry M. (1985). “The Politicized Presidency”. In: The New Direction in American Politics. Ed. by
John E. Chubb and Paul E. Peterson. Washington: Brookings Institution Press.
National Park Service (2012). “Civil War Sites Advistory Commission Report Update & Resurvey”.
In:
Nisbett, Richard E. and Dov Cohen (1996). Culture Of Honor: The Psychology Of Violence In The
South. Boulder: Westview Press.
Owsley, Frank Lawrence (1925). “Local Defense and the Overthrow of the Confederacy: A Study in
State Rights”. In: The Mississippi Valley Historical Review 11.4, pp. 490–525.
Patterson, Benton Rain (2014). Lincoln’s Political Generals: The Battlefield Performance of Seven Con-
troversial Appointees. Jefferson, NC: McFarland.
Phisterer, Frederick (1883). Statistical Records of the Armies of the United States. New York: C. Scrib-
ner’s Sons.
Piatt, Donn and Henry Van Boynton (1893). General George H. Thomas: A Critical Biography. Cincin-
nati, OH: Robert Clarke & Co.
Pilster, Ulrich and Tobias Böhmelt (2011). “Coup-proofing and military effectiveness in interstate
wars, 1967–99”. In: Conflict Management and Peace Science 28.4, pp. 331–350.
— (2012). “Do Democracies Engage Less in Coup-Proofing? On the Relationship between Regime
Type and Civil–Military Relations”. In: Foreign Policy Analysis 8.4, pp. 355–372.
Plutarch (1992). Lives of Noble Grecians and Romans. Trans. by John Dryden. New York: Random
House.
Poast, Paul (2015). “Lincoln’s Gamble: Fear of Intervention and the Onset of the American Civil
War”. In: Security Studies 24.3, pp. 502–527.
Reiter, Dan (2009). How Wars End. English. Princeton: Princeton University Press.
Reiter, Dan and Allan C. Stam (1998). “Democracy and Battlefield Military Effectiveness”. en. In:
Journal of Conflict Resolution 42.3, pp. 259–277.
41
Reiter, Dan and William A. Wagstaff (2017). “Leadership and Military Effectiveness”. In: Foreign
Policy Analysis. Advance Online Publication. doi: https://doi.org/10.1093/fpa/orx003.
Reuter, Ora John and Graeme B. Robertson (2012). “Subnational Appointments in Authoritarian
Regimes: Evidence from Russian Gubernatorial Appointments”. In: Journal of Politics 74.4,
pp. 1023–1037.
Rudalevige, Andrew (2017). The Cabinet was the easy part. Staffing (and steering) the bureaucracy
takes much more work. https://www.washingtonpost.com/news/monkey-cage/wp/
2017/01/10/the-cabinet-was-the-easy-part-staffing-the-bureaucracy-takes-
much-more-work. Blog.
Simpson, Brooks D. (2000). “Lincoln and His Political Generals”. In: Journal of the Abraham Lincoln
Association 21.1, pp. 63–77.
Staff of the Civil War Sites Advisory Commission (1993a). “Civil War Sites Advisory Commission
Report on the Nation’s Civil War Battlefields”. In:
— (1993b). “Civil War Sites Advisory Commission Report on the Nation’s Civil War Battlefields:
Technical Volume II: Battle Summaries”. In:
Sullivan, Patricia L. (2008). “At What Price Victory? The Effects of Uncertainty on Military In-
tervention Duration and Outcome”. In: Conflict Management and Peace Science 25.1, pp. 49–
66.
Taaffe, Stephen R. (2011). Marshall and His Generals: US Army Commanders in World War II.
Lawrence, KS: University Press of Kansas.
Trefousse, Hans L. (1963). Benjamin Franklin Wade: Radical Republican From Ohio. New York:
Twayne Publishers.
United States Marine Corps (1997). Warfighting. MCDP 1. USMC.
U.S. Army Concepts Analysis Agency (1991). “Database of Battles-Version 1990 (Computer Diskette).”
In:
42
Warner, Ezra J. (1959). Generals in Gray: Lives of the Confederate Commanders. Baton Rouge:
Louisiana State University Press.
— (1964). Generals in Blue: Lives of the Union Commanders. Baton Rouge: Louisiana State University
Press.
Wayman, Frank W., J. David Singer, and Gary Goertz (1983). “Capabilities, Allocations, and Success
in Militarized Disputes and Wars, 1816–1976”. In: International Studies Quarterly 27.4, pp. 497–
515.
Weiss, Herbert K. (1966). “Combat Models and Historical Data: The U.S. Civil War”. In: Operations
Research 14.5, pp. 759–790.
Wolford, Scott (2007). “The Turnover Trap: New Leaders, Reputation, and International Conflict”.
In: American Journal of Political Science 51.4, pp. 772–788.
Woodworth, Steven E. (1990). Jefferson Davis and His Generals: The Failure of Confederate Command
in the West. Lawrence, KS: University Press of Kansas.
Work, David (2012). Lincoln’s Political Generals. Urbana, IL: University of Illinois Press.
Wright, Sewall (1922). “Coefficients of Inbreeding and Relationship”. In: American Naturalist 56.645,
pp. 330–338.
43
Appendix A: Robustness Checks(NOT FOR PRINT PUBLICATION)
Appendix Table A-1: Latent Traits and Battle Outcomes (Minimum Traits Used)
Model A-1 Model A-2 Model A-3 Model A-4 Model A-5 Model A-6
Confederate Commander Competence −0.380∗∗ −0.377∗∗ −0.377∗∗ −0.365∗∗ −0.392∗∗ −0.243(0.127) (0.137) (0.138) (0.139) (0.142) (0.184)
Union Commander Competence 0.599∗∗ 0.531∗ 0.562∗ 0.553∗ 0.597∗∗ 0.329(0.209) (0.223) (0.225) (0.225) (0.228) (0.274)
Confederate Commander Loyalty −0.106 −0.104 −0.103 −0.026 −0.010 0.109(0.117) (0.118) (0.118) (0.241) (0.243) (0.272)
Union Commander Loyalty −0.066 −0.059 −0.055 −0.208 −0.208 −0.305(0.112) (0.113) (0.113) (0.281) (0.282) (0.311)
Confederate Strength −0.046∗ −0.046∗ −0.045∗ −0.054∗∗ −0.057∗
(0.019) (0.019) (0.019) (0.020) (0.024)Union Strength 0.026∗ 0.027∗ 0.026∗ 0.032∗∗ 0.029∗
(0.011) (0.011) (0.011) (0.012) (0.014)Confederate Battle −0.325 −0.324 −0.350 −0.115
(0.292) (0.296) (0.297) (0.343)Confederate Commander Loyalty × Confederate Battle −0.125 −0.139 −0.122
(0.275) (0.277) (0.302)Union Commander Loyalty × Confederate Battle 0.199 0.198 0.302
(0.308) (0.308) (0.344)Union Attacker −0.336 −0.406
(0.253) (0.281)Cutpoint 1 −1.025∗∗ −0.989∗∗ −1.249∗∗ −1.246∗∗ −1.454∗∗ −1.494∗∗
(0.147) (0.188) (0.301) (0.305) (0.345) (0.436)Cutpoint 2 −0.147 −0.093 −0.350 −0.345 −0.548 −0.484
(0.134) (0.179) (0.293) (0.297) (0.335) (0.427)
AIC 605.734 603.262 604.007 607.181 607.402 621.572BIC 627.836 632.730 637.159 647.701 651.605 765.231Log Likelihood −296.867 −293.631 −293.004 −292.591 −291.701 −271.786Likelihood Ratio Test 13.843∗∗ 20.315∗∗ 21.570∗∗ 22.396∗∗ 24.175∗∗ 64.005∗∗
Number of Observations 294 294 294 294 294 294
Note: Ordered logit coefficients presented. Observations are at the battle level. The dependent variable is an ordered factor with the levelsbeing, in order, Confederate Victory, Inconclusive, and Union Victory. Negative [positive] coefficients indicate the covariate is associated withhigher probabilities of Confederate [Union] victories. Standard errors in parentheses. Model A-6 includes commander-level fixed effects forthose who fought in at least seven battles. Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
44
Appendix Table A-2: Latent Traits and Relative Confederate Casualty Rates (Minimum Traits Used)
Model A-7 Model A-8 Model A-9 Model A-10 Model A-11 Model A-12
Confederate Commander Competence −0.151∗∗ −0.169∗∗ −0.178∗∗ −0.174∗∗ −0.173∗∗ −0.100(0.046) (0.056) (0.056) (0.056) (0.057) (0.074)
Union Commander Competence 0.223∗∗ 0.221∗∗ 0.258∗∗ 0.249∗∗ 0.251∗∗ 0.230∗
(0.081) (0.081) (0.084) (0.084) (0.086) (0.115)Confederate Commander Loyalty −0.283∗∗ −0.281∗∗ −0.285∗∗ 0.040 0.041 0.191
(0.055) (0.055) (0.055) (0.198) (0.198) (0.239)Union Commander Loyalty −0.036 −0.028 −0.027 −0.047 −0.047 0.035
(0.051) (0.054) (0.054) (0.157) (0.157) (0.193)Confederate Strength 0.001 −0.000 −0.001 −0.001 0.009
(0.004) (0.004) (0.004) (0.004) (0.006)Union Strength 0.001 0.001 0.001 0.001 −0.002
(0.003) (0.003) (0.003) (0.003) (0.004)Confederate Battle −0.234† −0.138 −0.137 −0.171
(0.126) (0.139) (0.140) (0.182)Confederate Commander Loyalty × Confederate Battle −0.346† −0.347† −0.401†
(0.201) (0.202) (0.233)Union Commander Loyalty × Confederate Battle 0.022 0.022 0.040
(0.166) (0.166) (0.215)Union Attacker −0.019 0.055
(0.114) (0.121)Constant 0.068 0.033 0.244† 0.171 0.180 0.292
(0.056) (0.080) (0.139) (0.147) (0.158) (0.212)
Quasi-AIC 306.011 307.983 309.020 311.037 312.019 339.457Wald Test 79.852∗∗ 79.544∗∗ 82.971∗∗ 85.868∗∗ 85.606∗∗ 137.530∗∗
Number of Observations 294 294 294 294 294 294
Note: Quasi-binomial logistic coefficients presented. Observations are at the battle level. The dependent variable is the proportion ofbattle casualties from the Confederacy. Positive [negative] coefficients indicate the covariate is associated with higher [lower] ratios ofConfederate casualties. Model A-12 includes commander-level fixed effects for those who fought in at least seven battles. Standarderrors in parentheses. Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
45
Appendix Table A-3: Latent Traits and Battle Outcomes (Maximum Traits Used)
Model A-13 Model A-14 Model A-15 Model A-16 Model A-17 Model A-18
Confederate Commander Competence −0.270∗ −0.223† −0.221† −0.216† −0.229† −0.032(0.119) (0.123) (0.123) (0.123) (0.124) (0.194)
Union Commander Competence 0.494∗∗ 0.479∗ 0.507∗ 0.506∗ 0.532∗ 0.295(0.190) (0.218) (0.219) (0.219) (0.221) (0.279)
Confederate Commander Loyalty −0.075 −0.071 −0.068 −0.060 −0.052 0.106(0.106) (0.107) (0.107) (0.244) (0.246) (0.268)
Union Commander Loyalty −0.058 −0.049 −0.043 −0.210 −0.219 −0.259(0.109) (0.110) (0.110) (0.293) (0.294) (0.321)
Confederate Strength −0.043∗ −0.044∗ −0.043∗ −0.050∗ −0.057∗
(0.019) (0.019) (0.019) (0.020) (0.024)Union Strength 0.020† 0.020† 0.020† 0.025∗ 0.028∗
(0.011) (0.011) (0.011) (0.012) (0.014)Confederate Battle −0.325 −0.322 −0.339 −0.150
(0.291) (0.291) (0.292) (0.340)Confederate Commander Loyalty × Confederate Battle −0.030 −0.033 0.181
(0.272) (0.273) (0.306)Union Commander Loyalty × Confederate Battle 0.204 0.214 0.270
(0.317) (0.318) (0.350)Union Attacker −0.267 −0.381
(0.249) (0.281)Cutpoint 1 −0.993∗∗ −1.051∗∗ −1.309∗∗ −1.306∗∗ −1.467∗∗ −1.674∗∗
(0.143) (0.189) (0.300) (0.301) (0.338) (0.434)Cutpoint 2 −0.126 −0.169 −0.424 −0.420 −0.578† −0.662
(0.130) (0.178) (0.291) (0.292) (0.328) (0.424)
AIC 610.836 608.784 609.513 612.964 613.810 621.526BIC 632.938 638.252 642.665 653.483 658.013 765.186Log Likelihood −299.418 −296.392 −295.756 −295.482 −294.905 −271.763Likelihood Ratio Test 8.741† 14.793∗ 16.064∗ 16.613† 17.767† 64.051∗∗
Number of Observations 294 294 294 294 294 294
Note: Ordered logit coefficients presented. Observations are at the battle level. The dependent variable is an ordered factor with the levelsbeing, in order, Confederate Victory, Inconclusive, and Union Victory. Negative [positive] coefficients indicate the covariate is associated withhigher probabilities of Confederate [Union] victories. Standard errors in parentheses. Model A-18 includes commander-level fixed effects forthose who fought in at least seven battles. Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
46
Appendix Table A-4: Latent Traits and Relative Confederate Casualty Rates (Maximum Traits Used)
Model A-19 Model A-20 Model A-21 Model A-22 Model A-23 Model A-24
Confederate Commander Competence −0.210∗∗ −0.209∗∗ −0.213∗∗ −0.213∗∗ −0.218∗∗ −0.041(0.043) (0.046) (0.047) (0.047) (0.047) (0.075)
Union Commander Competence 0.138∗ 0.143∗ 0.166∗ 0.165∗ 0.155∗ 0.066(0.067) (0.071) (0.076) (0.076) (0.077) (0.116)
Confederate Commander Loyalty −0.185∗∗ −0.188∗∗ −0.189∗∗ 0.144 0.128 0.287(0.047) (0.048) (0.048) (0.224) (0.225) (0.254)
Union Commander Loyalty −0.070 −0.070 −0.065 0.035 0.028 0.119(0.056) (0.057) (0.058) (0.166) (0.166) (0.198)
Confederate Strength 0.001 0.001 0.001 0.004 0.010(0.004) (0.004) (0.004) (0.005) (0.006)
Union Strength −0.001 −0.001 −0.001 −0.003 −0.001(0.003) (0.003) (0.003) (0.003) (0.004)
Confederate Battle −0.124 −0.033 −0.043 −0.174(0.134) (0.151) (0.152) (0.192)
Confederate Commander Loyalty × Confederate Battle −0.346 −0.335 −0.343(0.228) (0.228) (0.252)
Union Commander Loyalty × Confederate Battle −0.110 −0.107 −0.077(0.177) (0.177) (0.224)
Union Attacker 0.136 0.081(0.116) (0.124)
Constant 0.210∗∗ 0.210∗ 0.323∗ 0.251 0.189 0.343(0.059) (0.082) (0.147) (0.158) (0.167) (0.220)
Quasi-AIC 305.704 307.672 308.669 310.672 311.676 339.156Wald Test 57.420∗∗ 57.215∗∗ 57.821∗∗ 60.071∗∗ 61.199∗∗ 120.183∗∗
Number of Observations 294 294 294 294 294 294
Note: Quasi-binomial logistic coefficients presented. Observations are at the battle level. The dependent variable is the proportion of battlecasualties from the Confederacy. Positive [negative] coefficients indicate the covariate is associated with higher [lower] ratios of Confederatecasualties. Model A-24 includes commander-level fixed effects for those who fought in at least seven battles. Standard errors in parentheses.Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
47
Appendix Table A-5: Latent Traits and Battle Outcomes (Smaller Battles)
Model A-25 Model A-26 Model A-27 Model A-28 Model A-29
Confederate Commander Competence −0.533∗ −0.434 −0.425 −0.426 −0.423(0.266) (0.274) (0.276) (0.280) (0.289)
Union Commander Competence 0.553 0.241 0.297 0.374 0.372(0.368) (0.400) (0.405) (0.413) (0.416)
Confederate Commander Loyalty −0.111 −0.134 −0.132 −0.169 −0.171(0.150) (0.153) (0.154) (0.295) (0.295)
Union Commander Loyalty −0.074 −0.143 −0.143 −0.957∗ −0.960∗
(0.155) (0.163) (0.165) (0.457) (0.458)Confederate Strength −0.269∗ −0.262∗ −0.288∗ −0.287∗
(0.113) (0.113) (0.115) (0.118)Union Strength 0.147† 0.163∗ 0.133 0.132
(0.080) (0.081) (0.083) (0.084)Confederate Battle −0.487 −0.428 −0.427
(0.383) (0.396) (0.396)Confederate Commander Loyalty × Confederate Battle −0.117 −0.116
(0.355) (0.356)Union Commander Loyalty × Confederate Battle 1.021∗ 1.024∗
(0.490) (0.491)Union Attacker 0.014
(0.365)Cutpoint 1 −0.921∗∗ −0.779 −1.081† −1.280∗ −1.272†
(0.265) (0.539) (0.592) (0.616) (0.650)Cutpoint 2 −0.022 0.174 −0.118 −0.280 −0.272
(0.252) (0.536) (0.586) (0.606) (0.640)
AIC 287.049 282.146 282.507 280.057 282.056BIC 304.436 305.328 308.588 311.933 316.830Log Likelihood −137.524 −133.073 −132.254 −129.029 −129.028Likelihood Ratio Test 5.471 14.374∗ 16.012∗ 22.462∗∗ 22.464∗
Number of Observations 134 134 134 134 134
Note: Ordered logit coefficients presented. Observations are at the battle level. The dependent variable is an ordered factorwith the levels being, in order, Confederate Victory, Inconclusive, and Union Victory. Negative [positive] coefficients indicatethe covariate is associated with higher probabilities of Confederate [Union] victories. Battles under analysis are those wherethe total forces on both sides are at or below the median in our dataset. Standard errors in parentheses. Two-tailed tests:∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
48
Appendix Table A-6: Latent Traits and Relative Confederate Casualty Rates (Smaller Battles)
Model A-30 Model A-31 Model A-32 Model A-33 Model A-34
Confederate Commander Competence 0.177 0.154 0.154 0.151 0.154(0.144) (0.143) (0.143) (0.140) (0.142)
Union Commander Competence 0.344 0.211 0.254 0.304 0.296(0.209) (0.220) (0.226) (0.221) (0.225)
Confederate Commander Loyalty −0.045 −0.096 −0.102 −0.347 −0.349(0.095) (0.095) (0.096) (0.224) (0.224)
Union Commander Loyalty 0.059 0.028 0.041 −1.005∗∗ −1.005∗∗
(0.086) (0.085) (0.086) (0.338) (0.339)Confederate Strength −0.180∗∗ −0.185∗∗ −0.188∗∗ −0.186∗∗
(0.051) (0.052) (0.051) (0.052)Union Strength 0.003 0.005 −0.016 −0.018
(0.042) (0.042) (0.042) (0.042)Confederate Battle −0.227 0.105 0.102
(0.238) (0.263) (0.265)Confederate Commander Loyalty × Confederate Battle 0.274 0.275
(0.243) (0.244)Union Commander Loyalty × Confederate Battle 1.140∗∗ 1.140∗∗
(0.351) (0.352)Union Attacker 0.042
(0.197)Constant 0.157 0.636∗ 0.842∗ 0.637† 0.614
(0.136) (0.300) (0.372) (0.379) (0.394)
Quasi-AIC 145.333 147.668 148.662 151.051 152.012Wald Test 11.375∗ 23.712∗∗ 24.502∗∗ 36.884∗∗ 36.675∗∗
Number of Observations 134 134 134 134 134
Note: Quasi-binomial logistic coefficients presented. Observations at the battle level. The dependent variable is the proportionof battle casualties from the Confederacy. Positive [negative] coefficients indicate the covariate is associated with higher [lower]ratios of Confederate casualties. Battles under analysis are those where the total forces on both sides are at or below themedian in our dataset. Standard errors in parentheses. Two-tailed tests: ∗∗∗p < 0.01, ∗∗p < 0.05, †p < 0.1.
49
Appendix Table A-7: Latent Traits and Battle Outcomes (Larger Battles)
Model A-35 Model A-36 Model A-37 Model A-38 Model A-39
Confederate Commander Competence −0.345∗ −0.286† −0.288† −0.319† −0.336∗
(0.157) (0.163) (0.163) (0.165) (0.166)Union Commander Competence 0.630∗ 0.650∗ 0.648∗ 0.697∗ 0.801∗
(0.285) (0.313) (0.313) (0.317) (0.322)Confederate Commander Loyalty −0.140 −0.137 −0.135 −0.421 −0.336
(0.181) (0.184) (0.184) (0.678) (0.681)Union Commander Loyalty −0.030 −0.012 −0.011 0.604 0.563
(0.169) (0.172) (0.172) (0.556) (0.572)Confederate Strength −0.036† −0.036† −0.038∗ −0.057∗∗
(0.019) (0.019) (0.019) (0.022)Union Strength 0.014 0.014 0.015 0.028∗
(0.012) (0.012) (0.012) (0.014)Confederate Battle −0.114 −0.148 −0.231
(0.483) (0.504) (0.511)Confederate Commander Loyalty × Confederate Battle 0.333 0.231
(0.701) (0.703)Union Commander Loyalty × Confederate Battle −0.748 −0.692
(0.588) (0.602)Union Attacker −0.818∗
(0.388)Cutpoint 1 −1.056∗∗ −1.195∗∗ −1.290∗∗ −1.354∗∗ −1.888∗∗
(0.202) (0.265) (0.485) (0.505) (0.575)Cutpoint 2 −0.198 −0.316 −0.411 −0.464 −0.977†
(0.185) (0.249) (0.475) (0.496) (0.561)
AIC 331.775 331.557 333.501 334.674 332.029BIC 350.263 356.209 361.234 368.570 369.006Log Likelihood −159.887 −157.779 −157.751 −156.337 −154.014Likelihood Ratio Test 8.089† 12.306† 12.362† 15.189† 19.835∗
Number of Observations 161 161 161 161 161
Note: Ordered logit coefficients presented. Observations are at the battle level. The dependent variable is an ordered factorwith the levels being, in order, Confederate Victory, Inconclusive, and Union Victory. Negative [positive] coefficients indicatethe covariate is associated with higher probabilities of Confederate [Union] victories. Battles under analysis are those wherethe total forces on both sides are at or above the median in our dataset. Standard errors in parentheses. Two-tailed tests:∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
50
Appendix Table A-8: Latent Traits and Relative Confederate Casualty Rates (Larger Battles)
Model A-40 Model A-41 Model A-42 Model A-43 Model A-44
Confederate Commander Competence −0.236∗∗ −0.258∗∗ −0.271∗∗ −0.269∗∗ −0.273∗∗
(0.060) (0.070) (0.070) (0.070) (0.070)Union Commander Competence 0.143 0.142 0.189† 0.173 0.163
(0.101) (0.104) (0.108) (0.109) (0.110)Confederate Commander Loyalty −0.287∗∗ −0.284∗∗ −0.285∗∗ 0.328 0.319
(0.069) (0.070) (0.070) (0.348) (0.350)Union Commander Loyalty −0.107 −0.094 −0.090 0.138 0.134
(0.072) (0.075) (0.075) (0.208) (0.209)Confederate Strength 0.001 0.000 −0.001 0.001
(0.005) (0.005) (0.005) (0.006)Union Strength 0.001 0.001 0.001 −0.000
(0.003) (0.003) (0.003) (0.004)Confederate Battle −0.253 −0.096 −0.101
(0.166) (0.198) (0.199)Confederate Commander Loyalty × Confederate Battle −0.631† −0.623†
(0.351) (0.353)Union Commander Loyalty × Confederate Battle −0.260 −0.259
(0.223) (0.224)Union Attacker 0.079
(0.147)Constant 0.224∗∗ 0.171 0.405∗ 0.289 0.253
(0.079) (0.112) (0.191) (0.208) (0.219)
Quasi-AIC 170.186 172.164 173.180 175.200 176.190Wald Test 59.346∗∗ 58.975∗∗ 61.090∗∗ 65.087∗∗ 64.895∗∗
Number of Observations 161 161 161 161 161
Note: Quasi-binomial logistic coefficients presented. Observations at the battle level. The dependent variable is the propor-tion of battle casualties from the Confederacy. Positive [negative] coefficients indicate the covariate is associated with higher[lower] ratios of Confederate casualties. Battles under analysis are those where the total forces on both sides are at or abovethe median in our dataset. Standard errors in parentheses. Two-tailed tests: ∗∗∗p < 0.01, ∗∗p < 0.05, †p < 0.1.
51
Appendix Table A-9: Latent Traits and Battle Outcomes (With “Crossover” Variables)
Model A-45 Model A-46 Model A-47 Model A-48 Model A-49 Model A-50
Confederate Commander Competence −0.388∗∗ −0.355∗ −0.355∗ −0.353∗ −0.376∗∗ −0.191(0.132) (0.138) (0.139) (0.140) (0.142) (0.203)
Union Commander Competence 0.630∗∗ 0.575∗ 0.607∗∗ 0.600∗∗ 0.638∗∗ 0.360(0.210) (0.230) (0.232) (0.233) (0.235) (0.288)
Confederate Commander Loyalty −0.154 −0.148 −0.151 −0.151 −0.141 0.092(0.166) (0.167) (0.167) (0.347) (0.349) (0.468)
Union Commander Loyalty −0.026 −0.010 −0.002 −0.106 −0.115 −0.186(0.121) (0.122) (0.122) (0.294) (0.296) (0.340)
Crossover Confederate Commander Loyalty 0.110 0.108 0.121 0.354 0.352 0.131(0.241) (0.243) (0.243) (0.535) (0.540) (0.708)
Crossover Union Commander Loyalty −0.240 −0.294 −0.304 −8.410∗∗ −8.214∗∗ −9.961∗∗
(0.312) (0.313) (0.314) (0.161) (0.161) (0.172)Confederate Strength −0.045∗ −0.045∗ −0.045∗ −0.053∗∗ −0.058∗
(0.019) (0.019) (0.019) (0.020) (0.024)Union Strength 0.023∗ 0.024∗ 0.024∗ 0.029∗ 0.029∗
(0.011) (0.011) (0.011) (0.012) (0.014)Confederate Battle −0.345 −0.406 −0.429 −0.150
(0.292) (0.340) (0.341) (0.429)Confederate Commander Loyalty × Confederate Battle −0.015 −0.021 0.038
(0.397) (0.398) (0.489)Union Commander Loyalty × Confederate Battle 0.153 0.164 0.260
(0.325) (0.326) (0.384)Crossover Confederate Commander Loyalty × Confederate Battle −0.293 −0.294 −0.140
(0.602) (0.606) (0.738)Crossover Union Commander Loyalty × Confederate Battle 8.110∗∗ 7.912∗∗ 9.711∗∗
(0.161) (0.161) (0.172)Union Attacker −0.305 −0.393
(0.253) (0.281)Cutpoint 1 −1.064∗∗ −1.068∗∗ −1.346∗∗ −1.394∗∗ −1.582∗∗ −1.605∗∗
(0.159) (0.200) (0.311) (0.343) (0.378) (0.516)Cutpoint 2 −0.185 −0.174 −0.448 −0.492 −0.676† −0.592
(0.147) (0.190) (0.302) (0.334) (0.368) (0.507)
AIC 609.782 607.756 608.349 614.325 614.858 628.727BIC 639.251 644.592 648.869 669.579 673.795 787.121Log Likelihood −296.891 −293.878 −293.175 −292.162 −291.429 −271.363Likelihood Ratio Test 13.795∗ 19.821∗ 21.228∗ 23.252∗ 24.719∗ 64.850∗
Number of Observations 294 294 294 294 294 294
Note: Ordered logit coefficients presented. Observations are at the battle level. “Crossover” variables indicate the battle-level mean competence for“crossover” commanders of the indicated side. The dependent variable is an ordered factor with the levels being, in order, Confederate Victory, Incon-clusive, and Union Victory. Negative [positive] coefficients indicate the covariate is associated with higher probabilities of Confederate [Union] victories.Battles under analysis are those where the total forces on both sides are at or above the median in our dataset. Model A-50 includes commander-levelfixed effects for those who fought in at least seven battles. Standard errors in parentheses. Two-tailed tests: ∗∗p < 0.01, ∗p < 0.05, †p < 0.1.
52
Appendix Table A-10: Latent Traits and Relative Confederate Casualty Rates (With “Crossover” Variables)
Model A-51 Model A-52 Model A-53 Model A-54 Model A-55 Model A-56
Confederate Commander Competence −0.210∗∗ −0.228∗∗ −0.238∗∗ −0.233∗∗ −0.233∗∗ −0.107(0.048) (0.055) (0.055) (0.056) (0.056) (0.084)
Union Commander Competence 0.188∗ 0.197∗ 0.243∗∗ 0.237∗∗ 0.237∗∗ 0.290∗
(0.077) (0.080) (0.083) (0.084) (0.085) (0.128)Confederate Commander Loyalty 0.064 0.068 0.077 0.413 0.413 1.179∗∗
(0.093) (0.093) (0.093) (0.265) (0.266) (0.373)Union Commander Loyalty −0.015 −0.008 0.005 0.066 0.066 0.028
(0.060) (0.061) (0.061) (0.167) (0.167) (0.205)Crossover Confederate Commander Loyalty −0.485∗∗ −0.499∗∗ −0.514∗∗ −1.334† −1.334† −2.076∗∗
(0.116) (0.118) (0.118) (0.737) (0.739) (0.795)Crossover Union Commander Loyalty −0.125 −0.086 −0.115 0.155 0.155 0.109
(0.116) (0.124) (0.124) (1.058) (1.060) (1.044)Confederate Strength 0.004 0.003 0.003 0.003 0.003
(0.004) (0.004) (0.004) (0.005) (0.006)Union Strength −0.001 −0.001 −0.001 −0.001 0.002
(0.003) (0.003) (0.003) (0.003) (0.004)Confederate Battle −0.247† −0.101 −0.101 0.017
(0.127) (0.166) (0.167) (0.224)Confederate Commander Loyalty × Confederate Battle −0.384 −0.384 −0.721∗
(0.283) (0.283) (0.355)Union Commander Loyalty × Confederate Battle −0.070 −0.070 0.166
(0.180) (0.180) (0.244)Crossover Confederate Commander Loyalty × Confederate Battle 0.875 0.876 1.302
(0.746) (0.748) (0.790)Crossover Union Commander Loyalty × Confederate Battle −0.258 −0.258 −0.266
(1.066) (1.069) (1.062)Union Attacker −0.001 −0.033
(0.114) (0.125)Constant 0.005 −0.042 0.175 0.059 0.059 0.022
(0.069) (0.087) (0.142) (0.167) (0.175) (0.249)
Quasi-AIC 308.213 310.198 311.244 315.210 316.192 343.638Wald Test 92.161∗∗ 92.978∗∗ 96.965∗∗ 97.067∗∗ 96.721∗∗ 145.674∗∗
Number of Observations 294 294 294 294 294 294
Note: Quasi-binomial logistic coefficients presented. Observations at the battle level. “Crossover” variables indicate the battle-level mean competencefor “crossover” commanders of the indicated side. The dependent variable is the proportion of battle casualties from the Confederacy. Positive[negative] coefficients indicate the covariate is associated with higher [lower] ratios of Confederate casualties. Battles under analysis are those wherethe total forces on both sides are at or above the median in our dataset. Model A-56 includes commander-level fixed effects for those who fought inat least seven battles. Standard errors in parentheses. Two-tailed tests: ∗∗∗p < 0.01, ∗∗p < 0.05, †p < 0.1.
53