The Nonmartial Origins of the “Martial Races”

Introduction

Responding to the reverse wave of authoritarianism that engulfed newly decolonized countries in Africa, Asia, and the Middle East in the 1950s and 1960s, scholars of civil–military relations contended that the military was best poised to promote nation building in these countries in the absence of pluralist civil society organizations. This consensus in favor of the military as a tool for nation building gave way to a new consensus in the 1970s and 1980s that militaries were as prone to ethnic particularisms as the civilian regimes they had replaced. The narrow ethnic identity of Idi Amin’s regime in Uganda was held up as a prime embodiment of this thesis. What had undermined the military’s cohesiveness, the new scholarship argued, was the colonial “martial race” doctrine. ¹

The doctrine, which was first adopted in British India after the Sepoy Mutiny of 1857, justified the overrepresentation of select communities in the colonial security apparatus on the grounds that “one of the essential differences between the East and West” is “in the East, with certain exceptions, only certain clans and classes can bear arms,” whereas “The others have not the physical courage for the warrior.” ² However, as noted by Donald Horowitz in his seminal work, in practice, the status of martialness was attributed to those groups who could be relied upon to provide a steady stream of loyal recruits and withheld from groups who had developed sophisticated forms of military and political organization in the precolonial period. ³ While this recruitment strategy was consistent with the colonial army’s primary mission of maintaining internal security—it was rarely used as a fighting force—it bequeathed a debilitating legacy of split ethnic domination of military and political institutions after independence since the groups that dominated the armed forces lacked the political skills and the strength of numbers to prevail in electoral politics. Thus, far from being the pivotal step toward nation building, military intervention in politics was merely a pretext for furthering ethnopolitical agendas.

The key insight that emerged from this constructivist reading of the martial race doctrine was that the military could be a successful vehicle for nation building in the postcolonial context only if it is recruited by conscription. ⁴ In other words, the ethnic composition of the military should mirror the ethnic composition of the civilian population. Thus, a case that this literature would cite approvingly is the Singapore Armed Forces (henceforth SAF) after independence. The SAF had been an overwhelmingly Malay-dominated body prior to independence, but a policy of conscription, adopted after independence, reduced the proportion of Malays and increased the proportion of ethnic Chinese to bring the share of these groups in the armed forces in line with their population shares.

It is instructive to note that, in recent years, the origins of the colonial martial races have once again become a contested terrain. In particular, two alternative narratives have emerged as rivals to the constructivist narrative. The first narrative restates the primordialist view of colonial historiographers that the martial races were precolonial military powers and not products of colonial discourse, whereas the second narrative takes the more relativist position that colonial recruitment practices were idiosyncratic—the martial race was a constructed identity in some cases, but not in others.

The primordialist view is most succinctly articulated in Sean Walsh’s study of the SAF. ⁵ According to Walsh, Singapore’s postindependence policy of conscripting its armed forces has severely undermined their combat effectiveness because the ethnic balance of the forces shifted against the minority Malay community that “culturally looks highly on military service.” ⁶ The extent of this change was such that the Malays went from being heavily overrepresented in colonial military and police forces on the eve of Singapore’s independence to being what one researcher described as “exasperated for not being called to serve” by the mid-1970s. ⁷ Building on this premise, Walsh implicitly advocates for a return to the colonial status quo of voluntary military service, which would allow Malays to enlist in large numbers.

In contrast to the first narrative that emphasizes the primordial essence of the martial races, the second narrative combines the primordialist and constructivist positions, as illustrated by Zirker et al.’s analysis of the ethnic group-like features of postcolonial militaries. ⁸ Comparing the outrightly ethnic character of Uganda’s postindependence military forces to the relatively more professional outlook of Fiji’s postindependence forces, the study attributes this difference to differences in colonial recruitment practices, which persisted after independence. The indigenous Fijian community, which had been overrepresented in Fiji’s security forces under colonial rule, had a precolonial warrior tradition, but in Uganda, the Nubians, who rose to prominence under Idi Amin, were purely products of colonial discourse. In other words, colonial recruitment practices constructed militarily inclined ethnic groups in Uganda but were more attuned to precolonial military realities in Fiji.

Did colonial rule invent the martial race? If so, how pervasive was the use of this discourse? Despite the obvious differences that have emerged between the older constructivist readings of the martial race doctrine and the more recent interpretations, and the high stakes involved, these arguments have not been subjected to systematic empirical testing. This article fills this gap using an original data set on the colonial status of 181 ethnic groups in twenty-nine ex-British colonies to examine the determinants of ethnic bias in colonial security forces. Pairing my colonial legacies data set with data on the precolonial characteristics of modern ethnic groups from George Murdock’s Ethnographic Atlas, ⁹ I find that, consistent with the predictions of the earlier constructivist writings, groups that had a greater propensity to organize militarily in the precolonial period were associated with lower rates of incorporation into colonial security forces. In particular, I find robust support for the following propositions: (i) ethnic groups that had developed state-like structures in the precolonial period were significantly underrepresented in colonial security forces compared to ethnic groups that were stateless in the precolonial period and (ii) ethnic groups that provided stiff resistance against attempts by the colonial state to extend its power were significantly underrepresented compared to ethnic groups that offered weak resistance or simply acquiesced to colonial domination.

These findings suggest that, contra Walsh, postcolonial armies will be no more effective in combat if they are raised through voluntary enlistment than through conscription. Ethnic groups that had achieved high levels of military and political organization in the precolonial period were heavily underrepresented in the colonial security apparatus, whereas ethnic groups that had achieved minimal levels of military and political organization in the precolonial period were heavily overrepresented. The net result was that any differences in the propensity for military service across ethnic groups had been erased by the time the colonies gained independence. On one hand, the groups that were incorporated into colonial armies did not acquire fighting skills because they were mainly employed for internal security purposes. On the other hand, groups that had developed fighting skills in the precolonial period experienced gradual erosion of their fighting capacities because they bore the brunt of colonial “pacification” campaigns.

This article also contributes to an ongoing methodological debate in the social sciences on the value of constructivist approaches to the study of ethnic politics. In an influential critique of what they call the “construction-by-discourse” view in the field of ethnic violence, Fearon and Laitin (henceforth FL) argue that most studies using this perspective fail to establish a causal relationship between discourses and violence. ¹⁰ According to FL, the construction-by-discourse argument implies that “one can analyze or discern the logic of the discourse or symbolic system that constructs individuals and groups, and make predictions from this as to the likelihood of a range of practices, including violence.” The brunt of their critique is that both of these propositions are difficult to prove. First, discourses are complex and open to multiple interpretations. Second, discourses are long-standing structures, whereas ethnic violence is episodic.

FL’s critique can readily be extended to constructivist theorizing on the martial race doctrine. Drawing primarily on single case studies or small-N comparisons, this literature had paid inadequate attention to the question of causality. As a result, it failed to establish conclusively that a general discourse of the martial races existed and that this discourse left a legacy of interethnic competition in postcolonial security forces that would have been absent sans such a discourse. This article addresses the first shortcoming using quantitative data to demonstrate that the martial race discourse was consistently applied in a wide range of contexts with similar effects.

The rest of this article is structured as follows. The second section provides a brief survey of the constructivist research on the martial race doctrine and deduces a set of testable hypotheses regarding the origins of martialness. The third section describes the data and methods used in the study. The fourth section presents the main empirical results. The final section concludes.

Theory and Hypotheses

By most accounts, the concept of the martial race originated in India during the Sepoy Mutiny of 1857, and then diffused to other British colonies via administrators who had been transferred from India. In the course of the mutiny, which had occurred in the ranks of the Bengal Army stationed in eastern India, Punjabi and Pathan soldiers from northwestern India remained loyal to their colonial superiors and played a critical role in suppressing the dissenters, who came disproportionately from the local catchment areas. The mutiny, despite being a sporadic and short-lived affair, led to a profound reorganization of colonial rule in India as the juridical powers of the British East India Company passed directly into the hands of the Crown in England. One of the most important aspects of this reorganization was that Punjabis and Pathans began to be recruited to the colonial army in numbers that were highly disproportionate to their share of the colonized population. Further, the recruitment of these groups was not just limited to the army. In the course of time, they also came to be highly valued as policemen. ¹¹ Moreover, as soldiers and policemen, they were not just deployed locally, but sent to far-flung corners of the empire. ¹²

This recruitment strategy, which was soon to create for Punjab its reputation as the “Sword Arm” of British India, was justified on the grounds that the peoples of this region were the only ones in the subcontinent who had the capacity to bear arms. Within a short span of time, the idea also found its scholarly interlocutors. Sir Philip Mason, a civil servant, argued that the division of the Indian people between martial and nonmartial classes was not a British invention, but recognition of something already implicit in the social system, and therefore it was common sense to not waste effort in turning moneylenders into soldiers. ¹³

In trying to understand the wave of praetorianism that gripped the countries that emerged from colonial rule in the 1950s and 1960s, a body of constructivist research developed in the 1970s and 1980s, which argued that colonial martialness was unrelated to precolonial military prowess. From the perspective of this scholarship, groups that showed their loyalty under situations of stress were given the status of martial, and conferred benefits and advantages to ensure their continued loyalty. Yet, despite these broad lines of consensus, there are also important differences among these scholars in terms of what they perceive to be the attributes that made some groups more loyal than others. While some highlight the degree to which a group was integrated with the colonial administration as a determinant of loyalty, ¹⁴ others emphasize precolonial slavery, ¹⁵ still others point to the importance of precolonial political organization, ¹⁶ and finally, some highlight the importance of actual resistance to the colonial power. ¹⁷

The first variant of the loyalty argument is that groups that lived in remote areas, away from the centers of colonial power, were likely to be recruited in disproportionate numbers. This view is stated most cogently by Horowitz:

It is more than coincidence that martial characteristics were often discerned among groups that resided in mountainous or arid border areas remote from the administrative and economic centers where the colonial impact was most profound—in other words, far from the areas that, as time passed, also became centers of anti-colonial activity and potential disorder. ¹⁸

To examine the relationship between remoteness from colonial administrative centers and martialness, I test the following hypothesis:

A second variant of the loyalty thesis points to intergroup differences in the incidence of slavery in precolonial times. As noted by S. C. Ukpabi, the initial recruits for British armies in West Africa were predominantly ex-slaves who enlisted in large numbers to escape from their masters in the hinterlands. ¹⁹ These slaves made good recruits since colonial employment protected them from persecution by their former masters, and in return, they demonstrated unquestioning obedience. However, the system of recruiting exclusively from the slave class also made colonial forces extremely unpopular in the hinterlands where they were considered social outcasts. As a consequence, when colonial administrators sought to expand the manpower base of their armies to enlist freeborn men, very few recruits were forthcoming. This meant that, in order to maintain their forces at full strength, they had to resort to expedients such as reenlisting soldiers who would have normally been discharged on account of old age. Thus, the disproportionate representation of ex-slaves in western African armies continued up to the eve of independence although the abolition of the slave trade had become highly effective in the hinterlands by the end of the nineteenth century. To examine the relationship between precolonial slavery and martialness, I test the following hypothesis:

The third variant of the loyalty thesis is that groups that lacked sophisticated forms of precolonial political organization were likely to be highly sought after as soldiers and policemen. In his analysis of Britain’s military recruitment practices in Ghana during World War II, the historian David Killingray notes how most recruits came from the Northern Territories where local chiefs could easily round up recruits from their villages. ²⁰ In return, chiefs like Sandemanab of the Kanjarga ethnic group, who provided the largest number of recruits, were awarded the King’s Medal for Chiefs. By contrast, in the southern region, local chiefs could furnish fewer recruits since they were accountable to higher political authorities and could not arbitrarily send their men to the battlefront. According to Cynthia Enloe, what was recognized within decentralized communities such as the Kanjarga “was a mode of social organization that would enhance enlistments.” ²¹ Specifically, local chiefs could be persuaded, with sufficient inducements, to “draw upon existing bonds of obligation and credibility for the sake of bringing their communal subordinates into the army,” and the men who were enlisted in this way turned out to be highly reliable because they fought for the honor of their chiefs. To examine the relationship between precolonial political sophistication and martialness, I test the following hypothesis:

While Hypotheses 1–3 point to the geographic and cultural determinants of loyalty, the final variant of the loyalty thesis points to the importance of actual military resistance during the expansionary phase of colonial rule. As noted by Anthony Kirk-Greene, colonial military administrators privately admired “plucky losers,” but such groups were nonetheless heavily underrepresented because it was perceived that “What they possessed in valour was not compensated for by their reaction to discipline.” ²² To examine the relationship between resistance to colonial military expansion and martialness, I test the following hypothesis:

Data and Methods

The empirical analysis that follows is restricted to the application of the martial race doctrine in ex-British colonies. However, the doctrine was also adopted by other European powers, most remarkably, in French colonies. Myron Echenberg provides an excellent account of how the ostensibly conscripted colonial armies of French West Africa were disproportionately recruited from the Mossi ethnic group or Bambara-speakers generally. ²³ Yet, since the official French policy was to assimilate colonial subjects as “Frenchmen,” it is difficult to glean systematic comparative data on the ethnic composition of overseas military forces from the French archives. ²⁴ By contrast, the ethnic composition of overseas military and police forces was closely monitored and recorded by the Colonial Office in London. While the onset of World War II significantly attenuated London’s oversight of the ethnic composition of military recruits in the colonies, ²⁵ the ethnic composition of police forces continued to be centrally monitored until independence.

One of the key challenges that researchers who seek to measure ethnic outcomes face is to identify a representative sample of ethnic groups. Constructing such a list is difficult because of the situational nature of ethnicity—an individual may self-identify her ethnicity differently depending on the context. Using rigorous classification criteria that account for this endogeneity problem, ²⁶ I selected a sample of 181 ethnic groups in twenty-nine ex-British colonies. Table 1 provides the full list of the enumerated ethnic groups.

OPEN IN VIEWER

Table 1.

Sample of Ethnic Groups by Colony of Origin.

Colony	Groups
Bahrain	Asian, Bahrain Arabs
Belize	Garifuna, Maya Indian, Negros
Botswana	Kalanga, Kgalagadi, Lete, San Bushman, Yei, Tswana
Brunei	Asian Indian, Chinese, Dusun, Iban, Malay
Burma	Arakan, Chinese, Hill Tribal, Chin, Indian, Kachin, Karen, Mon, Shan, Burmese
Cyprus	Turkish Cypriots, Greek Cypriots
Fiji	Fijian, Indian, European, Pacific Islander
Gambia	Diola-Jola, Fulani, Mandingo, Mandjak, Serahuli, Serer, Wolof
Ghana	Akan, Ewe, Yoruba, Ga-Adamgbe, Guan,
	Konkonba, Mande-Busanga, Mossi-Dagomba
Guyana	Amerindians, Indians, Mixed, Negros
India	Assamese, Bengali, Bodo, Gujarati, Hindi-speaker, Muslim,
	Kannada, Malayalee, Marathi, Mizo, Naga, Oriya, Punjabi-Sikhs,
	Scheduled Castes & Scheduled Tribes, Tamil, Telugu, Tripuri
Iraq	Kurd, Turkoman, Muslims
Kenya	Kisii, Somali, Kalenjin, Kamba, Kikuyu, Luhya, Luo,
	Maasai, Mijikenda, Rendille, Turkana
Lesotho	Zulu, Basotho
Malawi	Chewa, Lomwe, Mananja-Nyanja, Yao, Ngoni, Nkonde-Tonga
Malaysia	Chinese, Dusun, Iban, Indian, Malay
Mauritius	Chinese, French, Mauritius Indians
Nigeria	Edo, Hausa, Ibibio-Efik-Ijaw, Ibo, Kanuri, Middle Belt, Ogoni, Yoruba
Pakistan	Ahmadi, Baluchi, Mohajjir, Hindu, Pashtun, Punjabi, Sindhi, Bengali Muslims
Qatar	Iranian, Qatar Arabs
Sierra Leone	Creole, Kissi, Kono, Koranko, Limba,
	Loko, Mende, Sherbro, Susu, Temne
Singapore	Indian, Malay, Singapore Chinese
Sri Lanka	Indian Tamil, Moor, Sri Lankan Tamil, Sinhalese
Sudan	Fur, Northern Arab, Nuba, Southerner
Swaziland	Zulu, Swazi
Tanzania	Chagga, Gogo, Ha, Haya, Hehe, Iramba, Nyakusa, Nyamwesi, Nyassa, Pare,
	Rufiji, Rukwa, Rwanda-Rundi, Sukuma, Asian, Yao, Zanzibari, Zigula
Trinidad and Tobago	Black, Indian, Mixed
Uganda	Acholi, Alur, Baganda, Banyankole, Banyarwanda, Banyoro, Basoga, Gisu,
	Kakwa, Karamojong, Kiga, Lango, Lugbara, Padhola, Konjo, Teso, Toro
Zambia	Barotse, Bemba, Lunda-Kaonde, Nyanja-speaker, Tonga

Note: Police data are missing for ethnic groups in boldface.

Dependent Variable

The dependent variable of my study, which is denoted as martialness, is simply the difference between a group’s percentage share of the police manpower in its colony of residence on the eve of independence and its percentage share of the colony’s population. I use the ethnic composition of police forces to measure the colonial preference for the martial races because comparative data on the ethnic composition of armies is missing from the records of the Colonial Office in London after World War II.

At one level, the choice of the ethnic composition of colonial police forces as a proxy measure for the ethnic composition of colonial military forces seems almost axiomatic. As was noted earlier, the martial races dominated the colonial police force in India. Similarly, David Anderson and David Killingray note in the context of British West Africa that the martial race doctrine used for military recruitment was also applied to police recruitment. ²⁷ Yet, there is also a pervasive, though unsubstantiated, view in the extant literature that colonial police forces were less ethnically skewed than colonial military forces. ²⁸ A simple consideration of the distribution of martialness should be sufficient to dispel this view. Figure 1 provides a quantile plot of martialness. The plot demonstrates that roughly 65 percent of the observations have martialness values of 0 or less. In other words, only a small fraction of observations are demographically overrepresented, which is exactly as we would expect if some sort of screening criterion were being used to favor some groups at the expense of others.

OPEN IN VIEWER

Figure 1.

Quantile plot of martialness.

To further test the validity of the police ethnicity measure, I calculated the simple correlation coefficient between military and police representation for twenty-five ethnic groups across five colonies—Burma, Kenya, Nigeria, Guyana, and Sierra Leone—for whom I was able to collect military ethnicity data from secondary sources. The coefficient is 0.84, which implies that the extent to which a group was overrepresented in the police strongly predicts the extent to which it was overrepresented in the military, and conversely, the extent to which a group was underrepresented in the police strongly predicts the extent to which it was underrepresented in the military.

Independent Variables

The main source of the independent variables in my estimations is the Ethnographic Atlas (henceforth EA). The data set, coded by George Murdock, is an excellent resource for researchers who are interested in the precolonial past of ethnic groups. It presents vast amounts of anthropological data collected by colonial ethnographers throughout the nineteenth and early twentieth century in a format that is easily amenable to quantitative analysis. Most importantly for our purposes, the group attributes that are coded by the data set capture the socioeconomic context of the group at the point at which its members came into contact with colonial ethnographers, and by extension, with the colonial state.

To test Hypothesis 1, namely, whether the degree of remoteness from the center of colonial power had an effect on colonial warrior status, I constructed a measure of urbanization based on data from colonial census reports. This measure, which is denoted as urbanprop, codes the percentage share of each group in the population of the capital or the largest urban area of its territory of residence prior to independence.

I test Hypothesis 2 using EA’s Type of Slavery variable. The variable assigns a value of 1 to groups with no incidence of slavery in the precolonial period, 2 to groups who had experienced incipient or nonhereditary slavery, 3 to groups with unidentified enslavement type, and 4 to groups who had experienced hereditary and socially significant slavery. Since the hypothesis does not distinguish between the effects of incipient and hereditary slavery, I coded the binary variable enslaved such that enslaved takes the value 1 if Type of Slavery > 1 and 0 otherwise.

To test Hypothesis 3, I constructed alternative measures of the extent of precolonial political centralization based on EA’s Jurisdictional Heirarchy variable. The variable assigns a value of 1 to groups who lacked any political authority beyond the local community, 2 to groups who had developed petty chiefdoms, 3 to groups who had developed large chiefdoms, 4 to groups who had developed small states, and 5 to groups who had developed large states. My primary measure of centralization follows the natural differentiation in EA’s data between chiefdoms and states. Specifically, I coded the binary variable centralized such that centralized takes the value 1 if Jurisdictional Heirarchy >3 and 0 otherwise. To test for robustness, a second version of this variable was also coded in which the treatment category is restricted to groups who had developed large states only.

Finally, I test for the effect of resistance to colonial expansion by coding the dummy variable war, which takes the value 1 when the group had been involved in a major armed confrontation with the colonial state and 0 otherwise. Data on domestic armed conflict are from the Correlates of War extrastate war data set version 4.0. ²⁹ The data set defines an extrastate war as a conflict that involved sustained combat, with organized forces fighting on both sides, resulting in a minimum of 1,000 battle-related fatalities within a twelve-month period.

Table 2 provides descriptive statistics for all the variables used in the estimations. The reported figures suggest that the average group was nearly proportionately represented in colonial police forces, had a largely rural background, was enslaved in the precolonial period, had decentralized political authority structures, and was unlikely to have been engaged in a sustained armed conflict with the colonial state.

OPEN IN VIEWER

Table 2.

Descriptive Statistics.

Variables	Obs	Mean	SD	Min	Max
Centralized	140	0.392	0.490	0	1
Enslaved	132	0.689	0.464	0	1
Martialness	131	−1.381	13.129	−58.7	48.1
Urbanprop	168	13.449	22.335	0	100
War	181	0.104	0.307	0	1

Results

The results of my estimations are presented in Table 3. The dependent variable is martialness and the reported coefficients are ordinary least squares estimates. The first model in the table is the baseline regression estimated on groups with nonmissing observations only. The second model is estimated on the entire sample of ethnic groups.

OPEN IN VIEWER

Table 3.

Explaining Overrepresentation in Colonial Police Forces.

Dependent variable: Martialness	(1)		(2)
Independent variable	Groups with nonmissing observations only	SE	All groups (missing observations imputed)	SE
Urbanprop	−0.115	0.096	-0.092	0.077
Enslaved	1.552	2.989	0.639	2.964
Centralized	−5.194	2.797	−5.212	2.208
War	−10.999	4.661	−10.884	4.761
Constant	1.853	3.406	2.445	3.074
N	107		181
Colonies	20		29
R 2	.192		.182

Note: Robust standard errors clustered on colony; coefficients in boldface denote p value less than .10.

The first major finding of this study is that the regression coefficients are always signed in the direction predicted by the earlier constructivist literature. Urbanization, precolonial centralization, and resistance to colonial expansion are all negatively related to overrepresentation, whereas experience of precolonial slavery varies positively with overrepresentation.

Yet, this finding alone does not provide conclusive evidence for the constructivist arguments. In particular, the effects of urbanization and precolonial slavery turn out to be statistically insignificant, which seem to support the relativist view that colonial recruitment practices were idiosyncratic, varying on a country-by-country basis. The coefficient of urbanprop in the baseline regression implies that a one percentage point increase in urbanization lowers the level of overrepresentation (increases underrepresentation) by roughly one-tenth of a percentage point; the coefficient of enslaved implies that a change in group status from not being enslaved to being enslaved only increases the level of overrepresentation (lowers underrepresentation) by 1.5 percentage points.

However, the relativist view is invalidated by the robust performance of the centralization and resistance measures. The coefficient of centralized in the baseline regression implies that a change in group status from precolonial “statelessness” to precolonial “stateness” lowers the level of overrepresentation (increases underrepresentation) by at least 5 percentage points. Moreover, this difference remains unchanged when missing values are imputed. I also find a systematic effect of resistance to colonial rule on overrepresentation. The coefficient of war in the baseline regression implies that a change in group status from not having resisted colonial expansionism to having resisted colonial expansionism lowers the level of overrepresentation (increases underrepresentation) by at least 11 percentage points and the difference remains unchanged even when missing values are imputed. Finally, as a further test of robustness, the predicted effects are similar when the treatment category of centralized is restricted to include only the groups who had developed large states. ³²

To summarize, the central empirical result of this article can be stated as follows. Ethnic groups who had achieved high levels of military and political sophistication in the precolonial period were systematically excluded from the colonial security apparatus. Conversely, the so-called martial races were precisely the groups who had few vestiges of stateness in the precolonial period and did not resist the expansion of the colonial state.

Conclusion

In recent years, scholarship on military manpower recruitment practices in postcolonial settings, particularly the work of Walsh and Zirker et al., has reopened the question of the origins of the colonial martial races. Although an older tradition of constructivist research had argued that the martial races had nonmartial origins, the newer scholarship suggests that the martial races had a cultural inclination to serve in the military, which was reified by colonial recruitment practices. These divergent views have very different implications for the optimal choice of military manpower systems in these settings. While the constructivist view favors the use of conscription, the primordialist view favors the use of all-volunteer forces. This article discriminates between these arguments using an original data set on the colonial status of 181 ethnic groups in twenty-nine ex-British colonies to examine the determinants of colonial martialness. My analysis demonstrates that, consistent with the predictions of the older constructivist literature, colonial recruitment practices systematically distorted precolonial military and political realities. In sum, the martial races were, irreducibly, the products of colonial discourse.