Ethnic Geography and the Colonial Design of Administrative Units in Sub-Saharan Africa

Abstract

Subnational administrative units are fundamental to territorial states and their political topography, but we know little about how their borders are designed. I argue that indirect rulers engage in preservation by ethnically aligning administrative borders, which empowers peripheral actors. In contrast, centralizing governments disrupt ethnic groups and their ability for collective action by splitting groups (dismemberment) and/or creating diverse units (suffocation). I test this argument by studying colonial administrative unit designs in Sub-Saharan Africa. I contrast indirect British with more direct French colonial rule and use new historical data on administrative borders and ethnic geography. Modeling subnational borders with a probabilistic spatial partition model, I find strong positive associations with ethnic boundaries under British compared to French rule, which realized more extensive ethnic dismemberment but not suffocation. The results shed light on colonial administrative unit designs, thus highlighting the importance of unit endogeneity more broadly.

The design of administrative units is fundamental to states’ political geography. How administrative borders are drawn determines the number, shape, and demography of subnational governance units. The choice between few or many, ethnically homogeneous or diverse administrative units is far from innocuous. It affects, for example, local development (Alesina & Zhuravskaya, 2011; Grossman, Pierskalla, & Boswell Dean, 2017), state capacity (Henn, 2023; Müller-Crepon, 2023), as well as violent conflict (Cunningham & Weidmann, 2010; Juon, 2025). While the design of regions and districts is thus inherently political, empirical political science literature mostly treats them as exogenously given (Soifer, 2019). This article addresses this gap with a focus on the ethnic underpinnings of colonial administrative geographies in Sub-Saharan Africa. Studying administrative unit design with new data and methods, I contribute to understanding the origins of political topographies, particularly in Africa.

A small literature examines administrative geographies as politically determined, focusing primarily on unit proliferation. Following up on Green (2010), this literature examines the political drivers of unit splits along preexisting lower-level borders (see also Grossman & Lewis, 2014; Hassan, 2016; Resnick, 2017). However, the literature’s focus on relatively marginal changes of administrative geographies leaves unexplored the causes of the more fundamental overall partitioning of state territories into administrative units. ¹ Two main factors can account for this oversight. First, we lack explicit theories of administrative unit design. Second, any test of related arguments must overcome thorny problems in modeling the determinants of administrative partitionings, as neither the number nor shapes of units are known ex ante. I address the first problem by theorizing administrative divisions as strategic choices made in response to ethnic geographies. Newly collected data analyzed with a probabilistic spatial partition model allow me to test the argument’s observable implications.

I argue that governments can draw administrative borders to either facilitate or disrupt ethnic groups which harbor significant collective action potential arising from their often-strong social networks and political institutions. Administrative unit designs can preserve ethnic groups, their networks and institutions by aligning administrative borders with their geography. This facilitates schemes of indirect rule but hampers centralized governance. In contrast to preservation, centralizing governments can disrupt ethnic groups in two ways. Borders that cut through their settlement areas can dismember groups while designing diverse units can suffocate them. ² Both strategies disrupt ethnic networks and institutions, consequently lowering groups’ ability for collective action. While they require greater investment by the state, these strategies consolidate political power and facilitate direct rule.

My empirical analysis tests this argument with a focus on colonial rule in Sub-Saharan Africa. ³ I compare the effect of ethnic geography on administrative unit designs under relatively decentralized, indirect British rule with its impact under more centralized, direct French rule. To do so, I combine three innovations in empirical measurement and methods.

First, I measure administrative partitionings and their change as the main outcome of interest. I therefore collect administrative maps from across British and French colonies throughout the colonial period. These show the evolution of territorial governance from few, imprecisely designed administrative units towards the late-colonial setup of districts, cercles, and regions which partially persisted until today.

Second, I present new data on historical ethnic geography that remedy the low resolution and likely reverse causality in Murdock’s (1959) data on ethnic settlement areas. The new measure of ethnic geography is based on 49 newly digitized ethnic maps produced in the first half of the century. While less comprehensive in coverage of the African continent than Murdock (1959), the data improve on spatial detail, approximation of local ethnic diversity, and reflection of uncertainty over relevant ethnic groupings. Additionally, maps from the early colonial period allow me to investigate subsequent administrative border changes and mitigate the risk of reverse causality.

Third, I model the effect of ethnic geography on administrative borders using the Probabilistic Spatial Partition Model (PSPM) developed by Müller-Crepon et al. (2025). The PSPM models administrative units as the partitioning of a spatial network of points. Encoding covariates on the network’s edges, the model estimates the effect of ethnic boundaries on administrative borders conditional on confounding spatial features (e.g., rivers or watersheds) that may cause both. Moving beyond the original model, I present new estimators that capture the effects of macro-level strategies of ethnic dismemberment and suffocation. The extended model helps to empirically separate the two strategies that are negatively correlated, yet theoretically distinct.

The empirical results support the main argument. I find that ethnic boundaries are significantly associated with a substantively higher probability of district borders and border change in British colonies. Ethnic boundaries are weakly associated with the borders of French colonial cercles and do not explain their change over time. These results are robust to the inclusion of a lagged dependent variable that accounts for important sources of reverse causality bias and permutations of the empirical strategy. Distinguishing between ethnic dismemberment and suffocation, I find that the British engaged in significantly less dismemberment than the French, with no significant difference in the degree of suffocation. Lastly, I draw on post-colonial data on administrative borders and show long-term persistence of the ethno-geographic roots of administrative units in former British colonies as compared to former French colonies.

Ethnic Geography and the Design of Administrative Units

With the exception of the literature on administrative unit proliferation, political scientists mostly treat administrative divisions as relatively stable political institutions. ⁴ Yet, once understood as the outcome of political choices, questions arise on the strategies of partitioning state territories into administrative units of an ex ante unknown number and shape.

The theoretical point of departure is the claim that the spatial decentralization of administrations is a ‘dual use’ governance tool. On the one hand, decentralized governance units aid the state in bridging the spatial gap between its center and the population. However, administrative units can also facilitate collective action among peripheral populations either due to a continuous influence of pre-existing actors or, more dynamically, by aiding the mobilization of society. Importantly, whether decentralized governance units are structurally predisposed to serve the center or periphery hinges on their congruence with local socio-political topographies, in particular ethnic geography. ⁵ As a result, the balance of power between the center and periphery affects how administrative borders are drawn: Where local populations and elites enjoy a power advantage and states rule indirectly, administrative divisions are drawn along ethnic lines. Where states are stronger and aim to rule directly, units will be designed to break pre-existing institutions and prevent local mobilization against the state. While empowering the state in the long-run, such misalignment is costly in the short-run as it disrupts prior local governance arrangements.

The creation of administrative outposts is a crucial instrument by which territorial states increase their “infrastructural power” (Mann, 1984). Decreasing the physical distance between state agents and the population, administrative decentralization facilitates greater and more spread-out control of, extraction of taxes from, and provision of services to citizens. As a result, most states’ administrations are spatially organized with hierarchically nested administrative tiers reaching down from the center to local populations. While there are decreasing or even negative returns to administrative fragmentation (Grossman et al., 2017), larger states tend to have more units arranged in more extensive hierarchies.

At the same time, a large literature highlights the promise of administrative decentralization to empower local populations by increasing citizens’ control of local policies and thus better aligning governance supply and demand (Tommasi & Weinschelbaum, 2007; Grossman et al., 2017). Such empowerment is dependent on local capacities for collective action, which is crucially shaped by ethnicity for at least two reasons. First, due to general homophily, geographic concentration and segregation, as well as common language use, ethnic social networks facilitate information sharing as a crucial precondition for collective action (Larson & Lewis, 2017). Second, ethnic groups oftentimes harbor distinct political institutions that facilitate intra-group coordination (e.g. Fortes and Evans-Pritchard, 1940). Even where political institutions lack geographic scale as in acephalous ethnic groups, these institutions bear greater similarity within groups than across them, which facilitates political integration and mobilization under a common roof. ⁶

As a result, the empirical literature finds that the benefits of decentralization materialize in units that are spatially congruent with socio-political institutions (Wilfahrt, 2022) and ethnically homogeneous (Alesina et al., 1999; Habyarimana et al., 2007; Miguel & Gugerty, 2005). Beyond mobilizing for better governance and public services, the capacity of peripheral collective action can be used to counter the central government more broadly. The design of ethnically delineated administrative units has thus also been linked to successful secessions (Griffiths, 2016; Roeder, 2012), lower political stability and generalized social trust (Alesina, Easterly, & Matuszeski, 2011), as well as stronger ethnic identities (Müller-Crepon, 2025; Robinson, 2020).

Building on the ambiguous effect of administrative borders, we can delineate two distinct designs of administrative units: one preserving ethnic groups and the other disrupting them through spatial misalignment, using administrative borders to dismember and/or suffocate their communities (Figure 1).

Figure 1.

Preservation, dismemberment and suffocation Note. In (b), ethnic settlement areas in color and administrative borders in black

The first strategy of preservation aligns the borders of administrative units with the geography of ethnic groups. In particular when coupled with the cooptation of local elites, designs that create relatively homogeneous administrative divisions allow states to quickly and effectively build their reach on the cheap. This comes at the cost of maintaining or even strengthening local capacity for collective action and sharing power and revenue with local elites. The result is a fragmentation of states’ institutional landscape which can hamper state-wide governance.

More disruptive, but expensive, are administrative designs that undermine the functioning of ethnic groups, their social networks and institutions through strategies of dismemberment and suffocation (see also Geertz, 1963; Englebert et al., 2002). In an attempt to break the organization of local societies, governments can impose administrative borders that partition groups such that they are dismembered across multiple units. This disrupts groups’ internal organization and mobilization capacity while decreasing governments’ economies of scale. Alternatively, states can suffocate competing powers and prevent the rise of future ones by creating diverse administrative units that include several groups. The resulting units will be internally divided, which reduces their usefulness as vehicles for mobilization and increases central governments’ ability to rewire local society. However, the governance of internally fragmented units is relatively inefficient, as administrators have to respond to varied local circumstances.

Governments likely trade off the potential long-term benefits of disruption against the potentially steep short-term cost of local resistance and inefficiencies in local governance. Centralizing governments then choose the mix of dismemberment and suffocation based on the geographic feasibility and costs of either strategy. As conceptualized in Figure 1, combining dismemberment and suffocation likely yields maximal disruption. Yet, either method on its own weakens ethnic groups and might therefore be sufficiently disruptive to achieve a targeted level of state centralization. This is particularly important as both strategies cannot always be easily combined: Suffocation tends to come with larger units, while dismemberment is fostered by smaller units. As a result, combining dismemberment and suffocation tends to yield less compact units. ⁷ The resulting impracticalities suggest that dismemberment and suffocation might be strategic substitutes.

The institutional constraints underlying states’ administrations affect the degree to which administrative units are designed in (mis)alignment with ethnic geography. Administrative preservation of ethnic groups is congruent and indeed complementary to strategies of indirect rule and a comparatively weak central state: ⁸ The cooptation of preexisting political institutions to build state reach works best if their social foundations and territorial expanse remain intact and spatially consistent with the new administrative units. Even where groups are only weakly organized, ethnically aligned administrative divisions can augment groups’ internal organization and facilitate their integration into the state. Any imposed territorial discontinuities through either dismemberment or suffocation lead to a change in local governance arrangements. This contravenes indirect rulers’ credo of “if it ain’t broke, don’t fix it” (Gerring et al., 2011, p. 385) and creates costs the relatively weak center is unable or unwilling to bear:

Preservation : Indirect rule increases the alignment of administrative borders with ethnic geography compared to direct rule.

In contrast, strong states use direct rule to break and replace local governance arrangements. This is aided by disruptive administrative designs that dismember and/or suffocate ethnic groups, their social networks and institutions. Akin to dynamics of “divide and rule”, strategies of dismemberment and suffocation produce political tensions and conflict within and between groups, thereby limiting collective action capacities and facilitating the imposition of rule by the center.

Dismemberment : Direct rule increases the division of ethnic groups by administrative borders compared to indirect rule.

Suffocation : Direct rule increases the internal division of administrative units by ethnic boundaries compared to indirect rule.

In general, it is important to note that the strategic incentives for governments to (not) align administrative borders with ethnic geographies exist irrespective of other administrative unit design choices such as units’ average size and compactness. While various design aspects are interrelated for practical reasons, they are not fully co-determined. Furthermore, my theoretical argument explains only one source of variation in administrative unit designs. Concerns beyond ethnicity and (in)direct rule shape administrative designs, in particular administrative, physical, and resource exigencies. For example, administrative efficiency will favor compact units that are smaller in densely populated areas. The physics of transportation underlying states’ territorial reach increase the chance that administrative borders run along rivers and mountain ridges. And states’ revenue imperatives likely lead them to place more and smaller units into resource-rich areas. While the following abstracts from such parallel sources of district designs, the empirical discussion returns to them as far as they risk biasing the analysis.

British and French Colonial Administrative Unit Designs

I turn to British and French rule over large parts of Sub-Saharan Africa to investigate my argument that indirect rule leads to ethnically aligned administrative borders, while direct rule produces administrative borders designed to disrupt ethnic groups. The following contextualizes my theoretical argument in the relevant historical literature. In particular, I discuss the spatial character of ethnic groups and pre-colonial political institutions, their encounter with rigid European conceptions of bounded territoriality, and differences between more indirect British and direct French colonial rule.

Understanding ethnic administrative unit designs in Africa not only sheds light on colonial strategies of rule and their ethno-political legacies (Ali et al., 2019; Neupert-Wentz & Müller-Crepon, 2024). The continent also promises three advantages over studying administrative border design elsewhere. ⁹ First, territorial statehood was imposed by the colonizers on the continent comparatively late at the turn of the century (Asiwaju, 1983; Paine et al., 2025). This implies that we can study the introduction of administrative borders as a governance revolution occurring over a few decades rather than as the longer process it was elsewhere. Second, while local actors clearly constrained and shaped colonial policies, the general French-British difference in (in)direct rule originated in the empires’ ideology and overall capabilities rather than within colonies (Müller-Crepon, 2020). ¹⁰ This is, again, different from historical shifts from indirect to direct rule driven by strategic center-periphery interactions (Hechter, 1975; Tilly, 1990). Third, the comparative haphazardness of colonial border designs (Herbst, 2000) ¹¹ facilitates the study of administrative borders within colonial empires. Elsewhere, selection bias complicates analyses as some subnational borders over time become state borders (Carter & Goemans, 2011) which are affected by ethnic geography (Müller-Crepon et al., 2025; Cederman et al., 2025).

While focusing on Sub-Saharan Africa, I expect the theoretical argument to capture patterns of subnational unit designs elsewhere. For example, post-unification Germany in 1871 remained administratively divided along the borders of the comparatively powerful kingdoms in the South. This contrasts with the redrawing of administrative borders in post-unification unitary Italy which did not follow pre-existing political borders (cf. Ziblatt, 2004, 2006). A similar contrast consists in the drawing of new borders in post-revolution France and Russia. The former case featured an almost grid-like homogeneous design with few compromises. In contrast, substate borders in Russia under Lenin cut across pre-existing units but were roughly aligned with the prevailing ethnic geography (e.g., Hirsch, 2000). While I expect the findings to travel to other cases of internal or externally imposed governance revolutions, the generalizability of the empirical findings remains speculative until empirically tested.

Political Topographies, Territoriality, and the Introduction of Borders

Defining the state and its subordinate administrative divisions via its territory demarcated by borders is integral to our contemporary idea of statehood (e.g., Mann, 1984; Weber, 1919). Yet, the use of widthless lines to territorially bound political entities is a fundamentally modern phenomenon that “was virtually unknown in most places in Africa during the period before the European partition” (Asiwaju, 1983, p. 45). Instead, frontier zones characterized the spaces between the political cores of states where they existed, mirroring the absence of interstate competition over abundant and sparsely populated territory (Herbst, 2000; Wilfahrt, 2025).

Not dissimilar to premodern Europe (Ruggie, 1983), precolonial rule oftentimes featured spatially overlapping and non-aligned jurisdiction over territory and people, with control over people being more valuable than control over territory (Herbst, 2000; Wilks, 1975). In practice, this implied strong political control over the core areas of states and only irregular and weak reach into their peripheries. The power of rulers was thus radiating outwards, gradually ranging from tight control in the center, via tributary rule, to mere raiding of peripheries. The rarity of precolonial maps only underscores the absence of demarcated boundaries (Herbst, 2000). Evidently, the concept of territorially bounded political entities was even more foreign to acephalous societies where political power was not centralized. After all, the political boundary comes only to life as a separating line between political entities (Kristof, 1959).

The prevailing diversity of political topographies changed rapidly with the colonial conquest in the late and early century. Not only did the European conquerors carve up the continent into empires and colonies, but they also partitioned those colonies into administrative units to create the basic infrastructure needed to establish ‘effective control.’ Thus shifting from rule over people to territorial rule, the elements of the hierarchical governance chain – the “thin white line” (Kirk-Greene, 1980) – presided over whoever happened to reside in their territorial unit. The creation of non-overlapping and neatly bounded administrative divisions – regions, districts, and subdistricts – was thus as much of a governance revolution as the drawing of international borders. Both replaced the precolonial variety of governance arrangements with sharp lines that delimited the territorial scope of states and their subnational entities across the entire African landmass (see Asiwaju, 1983).

To make matters worse, the introduction of the concept of bounded, non-overlapping units extended to the colonial conceptualization of ethnic geography. Historiographies describe the predominant colonial mindset as expecting individuals to be nested in tribes, “discrete, bounded groups, whose distribution could be captured on an ethnic map” (Young, 1985, p. 74). Incidentally, this understanding of ethnic geography – and the production of ethnic maps based on it – was intimately linked to ethno-nationalist discourses that colonizers brought from Europe (Berman, 1998).

Yet, the idea of geographically fixed, bounded, and non-overlapping groups met a reality where multiple groups often settled in the same environmental niche, as for example pastoralists and sedentary agriculturalists inhabiting the same savanna (Cohen & Middleton, 1970, p. 11). Even the identification of groups themselves could prove difficult, in particular where identities and associated loyalties were “complex, flexible and amorphous, sometimes overlapping, sometimes complementary, and did not add up to clearly demarcated tribes” (Lentz, 1995, p. 317; Southall, 1970). Consequently, ethnic groups as perceived by colonial governments did not correspond neatly to political entities, which in turn often included ethnic minorities among their populations (e.g. Colson, 1969; Wilfahrt, 2022).

While there was often no one-to-one correspondence between ethnic groups and precolonial institutions, the anthropological literature highlights that cultural groups tended to have distinctive political institutions, in particular but not only in cases with centralized political authorities (e.g. Fortes and Evans-Pritchard, 1940). For example, ethnic minorities ruled by a larger state often received special political treatment (e.g. Colson, 1969; Wilfahrt, 2022), indicating institutional differentiation along ethnic lines. As Ochonu (2014) shows for the case of Hausa governance over non-Muslim ethnic groups in the Nigerian middle-belt, (British) colonial rulers often built on these precolonial ethno-political topographies. Even acephalous ethnic groups featured more institutional homogeneity within than between groups. In turn, the creation of aligned administrative units around acephalous groups could effectively create group-level political organization and contribute to their identity formation (MacArthur, 2012, 2013; Müller-Crepon, 2025).

Local Rule and Administrative Unit Designs: British Versus French Styles

For British colonies in particular, there is ample case study evidence that colonial administrators often aimed at drawing district boundaries along the ‘tribal’ boundaries they perceived. Because the dominant strategy of indirect rule declared ‘tribes’ as ‘natural’ social units of local governance, ethnic settlement areas or ‘tribal homelands’ were destined to become administrative units (e.g., Asiwaju, 1970; Crowder, 1968; Miles, 1994; Spear, 2003).

While the British application of indirect rule has been widely studied, its impact on administrative unit design is not precisely documented. Writing on the internal borders of the British Gold Coast, today’s Ghana, Lentz (2006, p. 53) notes that the colonial government was able to make administrative borders roughly but not fully consistent with the prevailing, complex, and fuzzy ethnic geography. Instead, pragmatism coupled with administrative and geographic exigencies ultimately determined the precise location of borders (see also Howard, 2005). Yet, Sharpe (1986) describes how local chronicles were used in Northern Nigeria to determine ruling elites, their groups, and delimit administrative areas. Evidencing similar bottom-up processes, local populations and elites in the Gold Coast at times successfully mobilized for border change that would increase the alignment between social and administrative geographies (Bening, 1999). While Müller-Crepon (2020) notes that British rule was more direct in pre-colonially acephalous groups without centralized institutions due to creation of new subnational governments, the resulting governance arrangements were nevertheless ethnicized by grouping coethnic areas with similar local institutions together into a common administrative unit.

The historical case of French colonies is different for their more centralized governance. Although French administrations relied on local intermediaries as well, they tended to crush pre-existing institutions, replace them with more uniform institutions of their own making, and hand less power to local rulers (e.g., Cohen, 1971; Conklin, 1997; Crowder, 1964; Müller-Crepon, 2020). This strategy of direct rule likely originated in the higher level of administrative capacity available to the French and their preference for more centralized governance, following a Republican blueprint that despised hereditary and other forms of traditional rule (see, e.g. Müller-Crepon, 2020).

Strategies of administrative unit designs among French governments are again less studied. The available evidence at least partially points towards a more fundamental disregard of precolonial ethno-political geographies. Pourtier (1989, p. 288) notes for the case of Gabon the French goal of establishing a tight administrative hierarchy modeled after Republican France that directly opposed tribalism and did not regard ethnic geography. Bernier (1976) similarly concludes that French colonial cercles did not have roots in ethnic or institutional geography, at least not at the end of the colonial period. This pattern has also been noted by Crowder (1968, p. 175) across French West Africa (see also Guillaume, 1999; Suret-Canale, 1966). In contrast, Lefebvre (2019) for the case of Niger notes that the colonial administration did indeed, presumably because of their weakness in this most remote part of French West Africa, aim at harnessing the social power of local elites and therefore modeled administrative units after their reach.

Yet, the evidence cited above is based on relatively few cases which may not generalize. The case studies also do not typically account for potentially biasing influence of geographic features that simultaneously affect ethno-political geographies and colonial administrative borders. Given the impossibility of a perfect alignment between fuzzy ethnic topographies and sharp administrative borders, it remains unclear to what extent colonizers chose to preserve or disrupt ethnic geographies.

Observable Implications

Clear expectations arise from the theoretical argument, its application to British and French (in)direct rule, and the suggestive historical evidence. In particular, I expect that ethnic boundaries caused the drawing of aligned administrative borders in British colonies, yielding high levels of preservation of ethnic groups and their social networks and institutions. Consequently, and conditional on districts’ average size, I expect relatively low levels of dismemberment and suffocation. The effect of ethnic boundaries should be smaller if not absent in French colonies where disruption of ethnic groups was the theoretically expected strategy. Because extensive dismemberment and suffocation might be strategic substitutes, it is ex ante unclear what mix of increased administrative dismemberment and suffocation drove this disruption.

It is important to note the temporal dynamics of the increasing administrative partitioning of colonial territories. At the most basic level, the number of administrative units tended to steadily increase in lockstep with the expansion of the colonial state apparatus. But the redrawing of administrative borders could also be driven by bottom-up pressures such as those in Nigeria and the Gold Coast as well as top-down decisions by colonial governments driven by learning from failures and information accumulation. In line with the theoretical argument, I therefore expect ethnic geography to shape administrative border changes in British colonies more than in French colonies.

Historical Data on Administrative and Ethnic Geographies

I test whether colonial governments aligned administrative geographies with prevailing ethnic geographies using newly digitized administrative and ethnographic maps from the colonial period. The following presents the spatial data structure and the main variables of interest, providing the grounds for introducing the Probabilistic Spatial Partition Model (PSPM) thereafter.

Colonial Territories as Planar Graphs

The approach to modeling administrative borders follows Müller-Crepon et al. (2025) in understanding geographic space as a planar network of points. The network approach discretizes the otherwise infinite number of possible outcomes. Through the PSPM, it accounts for spatial interdependencies and observed covariates, thereby improving previous approaches to inferring the drivers of spatial partitionings.

The structure of the main graph takes into account geographic area and the heavily skewed population distribution of the African continent. ¹² I do so by spatially sampling the graph’s vertices with a probability proportional to the natural logarithm of local population counts estimated for the year 1880 by Goldewijk et al. (2010). ¹³ The edges of the planar network are derived through a Delaunay triangulation, which connects points closest to each other without creating overlapping edges. ¹⁴ I sample 400 vertices per million square kilometers, which yields an average edge-length of approximately 50 kilometers. Figure 2 plots the full graph and Figure 5 shows more detail for a subgraph of the Gold Coast (Ghana). Appendix A.10 shows robustness of the main results to varying spatial graph structures.

Figure 2.

Former British and French colonies included in the main sample as planar graphs

After constructing one graph for the entire continent, I split it into the British and French empires. Doing so avoids spatial overlap between the two samples. Throughout, I assess all intra-empire administrative borders of districts and cercles, including those that coincide with borders between colonies, since these were often changed and likely follow similar strategic concerns as administrative borders within colonies. ¹⁵

Administrative Geographies

To measure administrative borders, I extend the spatial and temporal coverage of existing data from British (Müller-Crepon, 2020) and French sources (Huillery, 2010). For the main analysis, I focus on British districts and French cercles. ¹⁶ Administrative borders changed with relative frequency (e.g. Bening, 1999; Wilfahrt, 2022), refining them over time. Beyond cross-sectional information, the panel data is particularly useful for studying border change with a lagged dependent variable model that addresses concerns over reverse causality.

The data collection aimed at mapping British and French administrative geographies in the early (ca. 1900-1920), mid (ca. 1920-1940), and late (1940-independence) colonial periods. In total, with the help of my research assistants, I have digitized 53 new administrative maps. Joined with the existing data, the final database spans 82 unique colony-years in 26 colonies with a total of 1’858 administrative units. ¹⁷

The data clearly shows an iterative process of refining administrative units in the British and French colonies, not unlike that of higher-level colonial borders (Paine et al., 2025). As shown in Figure 3, units’ average size roughly halves over time due to a parallel doubling of the number of administrative units. At the same time, borders’ ‘squiggliness,’ or fractal dimension, increases steadily as borders are drawn with increasing knowledge of social and natural geography. ¹⁸ Because French colonies cover more desert areas that are barely populated, their administrative borders are, on average, straighter.

Figure 3.

The refinement of colonial administrative partitionings over time Note. In (b), the fractal dimension of straight borders is 1. Its value increases towards 2 for very squiggly lines.

Because the full data covers some colony-periods multiple times and others not at all, the analysis draws on a trimmed version of the dataset. It uses only one map per period, dropping colony-periods without coverage (see Figure 4). I define periods’ start and end years flexibly (±3 years) to maximize coverage. In additional analyses that explore the persistent effects of ethnic geography post-independence, I employ the time-series of post-colonial regional borders from Müller-Crepon (2023). ¹⁹

Figure 4.

Administrative data used in the main analysis by colony and year. Transparent dots denote data that is dropped to avoid duplicate coverage of colony-periods. Note. Rows indicate the ISO-codes of colonies’ successor states

I intersect the data from every colony-period with the main graph introduced above. As the main outcome of the analysis shown in Figure 5(a), I code the administrative unit within which every node falls. At the level of edges, I code whether an edge, in a given period, crosses an administrative boundary or not.

Figure 5.

Data illustration: British Gold Coast (today’s Ghana). Note. Darker edges denote edges that cross district borders in (a) and greater ethnic differences in (b). The measure in (b) is computed as the fraction of maps that cover a given area on which an edge crosses an ethnic boundary, while discounting ethnic settlement areas that overlap on the same map

Historical Ethnic Geography

While colonial administrative borders are comparatively well-documented on maps that were crucial for the functioning of the colonial state, data on ethnic geographies is scarcer and of worse quality. Even more importantly, ethnographic maps are affected by the colonial conceptualization of ethnicity itself.

First, as noted above, the boundaries of groups themselves were oftentimes fuzzy, with mutually unintelligible languages separated by dialect continua, further complicating the “ethnic grouping problem” which highlights the importance of the choice of granularity at which groups are conceptualized (Posner, 2004). Second, groups often settled and continue to live in a spatially interspersed manner (Lentz, 1995). Third, individuals’ interethnic heritage and frequent multilingualism further complicate measurement (Buzasi, 2016). A fourth problem consists in historical reverse causality biases by which administrative borders shaped (perceptions of) ethnic groups and their geography (e.g. Müller-Crepon, 2025; Singh & Hau, 2016).

While clearly important, the above concerns should not deflect from the fact that individuals across Sub-Saharan Africa spoke and still speak a vast diversity of languages, which differ to varying degrees and cluster geographically. In other words, categorical classifications of ethnic groups tend to be more informative in the center of groups’ main settlement regions and less so in the linguistic and geographical space between them. The approach to measuring ethnic geography employed in this paper builds on this understanding while aiming to address the problems identified above. In particular, a collection of 49 newly digitized historical ethnographic maps drawn throughout the colonial period overcomes some of the problems associated with existing data on ethnic geography.

In the following, I first discuss problems relating to the quality and timing of the widely-used ethnographic map by George Murdock (1959). These limit the map’s usefulness for explaining administrative borders. I then present the new collection of colonial ethnographic maps. They are of higher quality, together capture local ethnic mixing and uncertainty, and allow for analyzing administrative border changes occurring after their date of production thus mitigating reverse causality.

Murdock’s Classic Map of Ethnic Groups and Its Shortcomings

The ethnic map produced by Murdock (1959) is based on earlier secondary sources, was digitized by Nunn & Wantchekon (2011), and has been very influential in the burgeoning quantitative historical literature on the continent. The map shows the approximate settlement areas of 842 ethnic groups across the whole continent. Yet, while its coverage is extensive, it features three important drawbacks. First, the map is of low resolution (1:10 million or 1 cm = 100 km) which leads to low spatial detail. Second, Murdock depicted groups’ settlement areas as non-overlapping, neatly bounded, as well as smoothly and regularly shaped. This suggests either significant noise or, worse, bias in measurement of ethnic boundaries. Lastly, it is unclear how Murdock triaged between potentially diverging information on relevant ethnic groupings and their geography from secondary sources. The latter two caveats point to the danger of reverse causality by which administrative regions might have influenced Murdock’s mapping of ethnic geographies.

Murdock’s map has the advantage of full and relatively uniform coverage of all of Africa, which makes it suitable for many research designs that do not rely on high spatial precision in the mapping of ethnic groups. Yet for explaining administrative borders, the map’s low resolution, lack of detail, and risk of reverse causality motivate a new collection of ethnographic maps that predate Murdock’s summary map.

Colonial Ethnographic Maps

The new collection of ethnographic maps from across Sub-Saharan Africa combines 49 newly digitized historical maps. The resulting measure of ethnic boundaries captures the information on ethnic geographies and the uncertainties and ambiguities associated with it at the time when administrative borders were drawn.

The maps were found through a systematic search in online map repositories and major library catalogues. ²⁰ All maps were produced prior to that of Murdock (1959). Most maps follow the typical “polygon-style” also used by Murdock, though many depict ethnic settlement areas as overlapping. A few maps depict uncertainty directly by showing the names of some ethnic groups without any spatial delimiters, thus denoting fuzzy and unbounded ethnic settlement areas. As visible in Figure 6, the maps have a much higher resolution than the one by Murdock, since they focus on regions, and often even on (parts of) single colonies.

Figure 6.

Ethnic settlement areas from 49 historical ethnic maps

The main drawback of using historical ethnographic maps as sources to measure ethnic geography consists in potentially biased measurement. Given the maps’ potential impact on policy making, diverging colonial ideologies might have produced maps to justify, for example, administrative unit designs, thus introducing the risk of reverse causality and omitted variable bias (cf. Lawrence & Sajid, 2025). A dedicated empirical strategy explained in more detail below addresses this issue by using temporally lagged ethnic maps to explain subsequent border changes. In sum then, the digitization of this ethnographic data allows for capturing local ethnic diversity and colonial ethnographers’ uncertainty about ethnic groupings and their geographies. I encode this information on the graph by computing for each edge the fraction of maps on which an edge crosses an ethnic boundary (see Figure 5(b)). Where groups are depicted only as “limitless” labels, I associate vertices only with a group if it falls on the label.

Probabilistic Spatial Partition Model

I use a recently developed Probabilistic Spatial Partition Model (PSPM) to estimate the effect of ethnic geography on the partitioning of the spatial graph into administrative units. The model, presented in detail by Müller-Crepon et al. (2025), captures the realization probability of any partitioning P = p _i among all possible contiguous partitionings $\mathbb{P}$ of a graph as a Boltzmann distribution

$Pr(P=p_i)=\frac{e^{-{ϵ}_i}}{{\sum }_{i=1}^{|\mathbb{P}|}{e}^{-{ϵ}_i}},$

(1)

where the likelihood of a given partitioning p _i decreases with the total “energy” ϵ _i associated with it. The PSPM encodes the explanatory factors of partitioning p _i as functions that contribute to its energy ϵ _i and estimates associated effect parameters.

I proceed in two steps. I first follow Müller-Crepon et al. (2025) and use edge-level predictors to test how far ethnic boundaries explain administrative borders as an overall measure of ethnic preservation. Second, I introduce new macro-level predictors to distinguish between ethnic dismemberment and suffocation. The following introduces both approaches.

Modeling Preservation as Edge-Level Effects of Ethnic Boundaries

Following Müller-Crepon et al. (2025), I first model a partitioning’s energy ϵ _i as depending on the realization of attractive and repulsive forces ϵ_j,k on the edges between all nodes j, k ∈ L:

$\begin{array}{c}\hfill {ϵ}_i=\sum _{j,k\in L}{\mathbb{1}}_{j,k}(u){ϵ}_{j,k},\end{array}$

(2)

$\begin{array}{c}\hfill {ϵ}_{j,k}={\beta }_0+{\beta }_1{\mathsf{e}\mathsf{t}\mathsf{h}\mathsf{n}\mathsf{i}\mathsf{c}\mathsf{b}\mathsf{o}\mathsf{u}\mathsf{n}\mathsf{d}\mathsf{a}\mathsf{r}\mathsf{y}}_{j,k}+\delta {\mathbf{x}}_{j,k},\end{array}$

(3)

As denoted by the indicator ${\mathbb{1}}_{j,k}(u)$ in equation (2), edges’ energy ϵ_j,k is only realized if vertices j and k are part of the same administrative unit u and 0 otherwise. If realized, ϵ_j,k is determined by the constant β₀ which accounts for the average size and compactness of units, the effect of ethnic boundaries ${\beta }_1$ , as well as the effects of covariates x_j,k captured by the parameter vector δ.

The coefficients denote repulsive forces if positive and attractive forces if negative. They are estimated similarly to typical regression coefficients. I expect a positive coefficient for ethnic boundaries in British colonies which is larger than that obtained from the French sample. Importantly, larger repulsive effects of ethnic boundaries suggest greater overall preservation: the absence of ethnic dismemberment and suffocation. Ethnic boundaries have a maximal effect if all ethnic boundaries align with a district border and no district border cuts through an ethnic group. In the baseline specification, covariates x_j,k capture potential joint determinants of administrative borders and ethnic geography that relate to alternative origins of administrative borders. In particular, I include edges’ length (km, logged), the size of the largest river and watershed they cross, their average elevation, as well as the average population density (Goldewijk et al., 2010) and distance to the coast of the two vertices they connect (both logged). The baseline specification pools across all three periods. For ease of estimation and implementation, I estimate the models separately for the French and British samples throughout.

The main inferential risk of the baseline analysis consists in reverse causality. In particular, in the apt words of Lawrence and Sajid (2025, p. 968), ideological commitment to colonial policies of (in)direct rule might well have “necessitated that local conditions be interpreted and represented in ways that made colonial administrators’ policy prescriptions seem legitimate and feasible.” In other words, the creation of administrative units likely had a direct effect on ethnographic maps which might simply go on to pretend that administrative borders align with ethnic geography (see also Müller-Crepon, 2025; Posner, 2005).

A lagged dependent variable (LDV) specification addresses this risk of reverse causality and omitted variable bias by adding a lagged dependent variable that captures administrative boundaries in t − 1. In addition, for the LDV model, I adjust the ethnic boundary measure such that it is only based on ethnic maps drawn in years prior to the observation of administrative borders in t − 1. Because of the scarcity of early ethnographic maps, the LDV model can only be estimated for the late colonial period, with administrative borders mapped in the mid-colonial period as the lagged dependent variable. Early ethnographic maps are available for West Africa in 1924, French Equatorial Africa in 1914, ²¹ and British East and Central Africa in 1943. ²²

Through the parameter coefficients, the PSPM estimates the contribution of each variable to the overall potential energy of edges to maximize the likelihood of observing the realized partitioning of colonies into administrative units. I use the model and estimator developed by Müller-Crepon et al. (2025), which estimates parameters via a maximum composite likelihood approach and derives standard errors from a parametric bootstrap. ²³

Ethnic dismemberment and suffocation as macro-level predictors

The edge-level characteristics in the main specification cannot distinguish whether ethnic alignment of administrative borders is driven by the absence of ethnic dismemberment, suffocation, or both. Such an analysis requires predictors situated at higher levels of analysis. I therefore extend the parametrization of partitionings “energy” ϵ _i (Equation (2)) with functions that operate across sets of vertices. ²⁴ In the present case, this is the extent to which a partitioning p _i dismembers groups g ∈ G so that they are internally fragmented by the borders of units u ∈ U _i and suffocates groups in ethnically heterogenous units:

${ϵ}_i={\gamma }_1\mathsf{d}\mathsf{i}\mathsf{s}\mathsf{m}\mathsf{e}\mathsf{m}\mathsf{b}\mathsf{e}\mathsf{r}\mathsf{m}\mathsf{e}\mathsf{n}\mathsf{t}(G,U_i)+{\gamma }_2\mathsf{s}\mathsf{u}\mathsf{ff}\mathsf{o}\mathsf{c}\mathsf{a}\mathsf{t}\mathsf{i}\mathsf{o}\mathsf{n}(G,U_i)+\sum _{j,k\in L}{\mathbb{1}}_{j,k}(u){ϵ}_{j,k}$

(4)

where the degree of suffocation and dismemberment of partitioning p _i is computed as fractionalization indices aggregated across groups g ∈ G and districts u ∈ U _i weighted according to the number of nodes they comprise (w _g and w _u ). ²⁵

$\mathsf{d}\mathsf{i}\mathsf{s}\mathsf{m}\mathsf{e}\mathsf{m}\mathsf{b}\mathsf{e}\mathsf{r}\mathsf{m}\mathsf{e}\mathsf{n}\mathsf{t}(G,U_i)=\sum _{g\in G}\left(w_g\left(1-\sum _{u\in U_i}{\left(\frac{g_u}{g}\right)}^2\right)\right).$

(5)

$\mathsf{d}\mathsf{i}\mathsf{s}\mathsf{m}\mathsf{e}\mathsf{m}\mathsf{b}\mathsf{e}\mathsf{r}\mathsf{m}\mathsf{e}\mathsf{n}\mathsf{t}(G,U_i)$ is computed based on the fractionalization of groups g spread across districts u in which they reside with proportions g_u/g. Dismemberment is 0 if each group is fully contained within one district (irrespective of its ethnic diversity). It increases as groups are split into ever more parts by administrative units. A negative γ₁ in equation (4) then suggests that districts are drawn to explicitly dismember groups, since the respective decrease in the overall energy ϵ of a partitioning increases its chance of realization. In turn, a positive γ₁-estimate would suggest that district designs leave groups less divided than expected.

Suffocation is similarly defined as

$\mathsf{s}\mathsf{u}\mathsf{ff}\mathsf{o}\mathsf{c}\mathsf{a}\mathsf{t}\mathsf{i}\mathsf{o}\mathsf{n}(G,U_i)=\sum _{u\in U_i}\left(w_u\left(1-\sum _{g\in G}{\left(\frac{u_g}{u}\right)}^2\right)\right),$

(6)

$\mathsf{s}\mathsf{u}\mathsf{ff}\mathsf{o}\mathsf{c}\mathsf{a}\mathsf{t}\mathsf{i}\mathsf{o}\mathsf{n}(G,U_i)$ is 0 if every district u is fully ethnically homogeneous. This is the case if the proportion of a district that belongs to any one group u _g /u is either 0 or 1. Suffocation increases as districts become more ethnically diverse. ²⁶ Accordingly, a negative γ₂ in equation (4) would signal that units are designed to suffocate groups, since the respective decrease in the overall energy ϵ increases a partitioning’s chance of realization. In turn, a positive coefficient γ₂ would suggest that administrative units are ethnically more homogeneous than expected.

Note that a coefficient of zero for the dismemberment and suffocation variables would entail no statistical association of ethnic geography with administrative partitioning. In effect, such a drawing of administrative borders without reference to ethnic geography would entail substantive dismemberment and suffocation. Moving below zero would entail district design with more dismemberment and suffocation than expected from (conditionally) random unit designs. Because 0 might thus be a relatively low bar for both indicators, I also test whether estimates for the British and French empires differ from each other. I expect dismemberment and suffocation to be less likely in British colonies, which should yield larger positive γ parameters than the French colonies.

Macro-level predictors such as those developed here are powerful in the extent to which they can be tailored to a theoretical argument. Yet, this precision comes with difficulties in modeling potential omitted variables, which could be modeled in an essentially limitless set of supra-edge predictors. Avoiding potentially arbitrary choices and complex computational procedures, I here simply control for the edge-level covariates used in the baseline analysis. These geographic factors likely have a mostly local effect. The lagged-dependent variable for the LDV specification furthermore accounts for the effects of further omitted variables, including time-invariant effects of stable supra-edge predictors.

Since the dismemberment and suffocation predictors are based on a categorical measurement of ethnic geography, ²⁷ I draw on the earliest available ethnographic maps for Central (1914), West (1924), and East (1943) Africa to estimate the γ parameters in equation (4). I continue to rely on the same composite maximum likelihood estimator with parametrically bootstrapped standard errors as Müller-Crepon et al. (2025). ²⁸

The Varying Effects of Ethnic Geography on Administrative Borders

The results from estimating the PSPM indicate that district borders were designed in relative congruence with ethnic geography in the British colonies. In turn, the borders of French cercles do not correlate systematically with ethnic boundaries. This difference is driven by greater levels of ethnic dismemberment in French compared to British colonies, with no significant difference in ethnic suffocation. Additional results show that colonial legacies in the estimated effects of ethnic boundaries on administrative borders persist until today. Results of robustness checks are available in the Appendix and discussed alongside the main results.

Preservation: The Effect of Ethnic Boundaries

Looking first descriptively at the raw data, Table 1 shows the (joint) incidence of ethnic boundaries and administrative borders by colonial empire. Overall, 44% of edges in British colonies cross ethnic boundaries, compared to 50% in French colonies. Similarly balanced, edges cross district borders at a rate of 35% in British, as compared to 32% in French colonies. Despite these similarities, the British sample features a much higher coincidence of ethnic and district boundaries. In British colonies, edges that cross ethnic boundaries are 28 percentage points more likely to also cross a district border (22.7 vs. 50.5%). This difference amounts to only 9 percentage points in French colonies (27.6 vs 36.5%).

Table 1.

Ethnic Boundaries and Administrative Borders at the Edge-Level, in %

	District border
Ethnic	British			French
boundary	All	0	1	All	0	1
All	100	65.0	35.0	100	68.0	32.0
0	56.0	77.3	22.7	50.4	72.4	27.6
1	44.0	49.5	50.5	49.6	63.5	36.5

Note.Displays average district border status by rounded ethnic boundary value.

Table 2 shows the main PSPM estimates across all three periods using the baseline and LDV specification. Looking first at the British colonies, we see a consistent and precisely estimated effect of ethnic boundaries on the probability that two vertices are separated by district borders. For fully independent, so-called “bridge” edges which can change whether they cross a district border or not irrespective of all other edges, ethnic boundaries are associated with a hazard ratio of 1.60 [1.49, 1.70] in the baseline model and 1.46 [1.29, 1.71] in the LDV specification. Setting all covariates to their median values, ethnic boundaries in British colonies are associated with an increase of the chance of a bridge edge being crossed by a border by 10 percentage points from 24.6 [23.9, 26.1] to 34.3 [33.7, 36.4] percent in the baseline specification. In the LDV model with a median lagged dependent variable value of 0, the effect of ethnic boundaries amounts to an increase of 6 percentage points from 19.1 [16.7, 21.6] to 25.6 [22.7, 29.0] percent. Note that interdependence between non-bridge edges tends to increase effect sizes as ethnic boundaries cross strings of connected edges. Most of the effect of ethnic boundaries in the LDV model is driven by the emergence of new administrative borders along ethnic boundaries and less so by greater stability of already existing ones (see Appendix Table A2). Patterns in the set of French colonies look different with a much smaller and unstable effect associated with ethnic boundaries. While the ethnic boundaries show an association with cercle borders in the baseline specification, this finding is not robust to the inclusion of the lagged dependent variable. In the former, the hazard ratio for bridge-edges associated with ethnic boundaries amounts to 1.15 [1.08, 1.23] but decreases to a statistically and substantively insignificant 1.02 [0.80, 1.28] in the LDV specification.

Table 2.

Ethnic Boundaries and Administrative Borders in British and French Colonies

	British		French
	(1)	(2)	(3)	(4)
Constant	−9.94*	−9.12*	−10.07*	−6.48*
	[−10.70; −9.24]	[−10.49; −6.99]	[−10.81; −9.10]	[−8.05; −2.87]
Ethnic boundary	0.47*	0.38*	0.14*	0.02
	[0.40; 0.53]	[0.25; 0.54]	[0.07; 0.21]	[−0.22; 0.25]
Lagged dep. var.		0.81*		1.06*
		[0.73; 0.97]		[0.89; 1.24]
Controls	Yes	Yes	Yes	Yes
No. of vertices	11,664	1,662	10,084	1,010
No. of edges	31,576	4,493	27,760	2,584
No. of units	1,466	247	876	117

Notes. 95% confidence intervals from parametric bootstrap in parenthesis. ^* Statistically significant at the 95% level.

Figure 7 disaggregates the results from baseline Models 1 and 3 in Table 2 into the three measurement periods of colonial administrative borders. ²⁹ The results show that the association between ethnic boundaries and administrative borders becomes stronger for the British and French colonies over time, yet at differing levels. While the association is substantive, increasing, and statistically significant for British colonies throughout, it only becomes statistically significant in the French parts of Africa in the late colonial period. The small and insignificant results from the LDV model (4) in Table 2 suggest that this increase might be due to bias from reverse causality by which administrative borders have affected the drawing of later ethnic maps. The increase in the cross-sectional effects of ethnic boundaries also suggests that the results are not driven by information-poor border designs using ethnicity as a heuristic shortcut to gauge political loyalties. If that were the case, we would expect the effect to dissipate over time as knowledge grew.

Figure 7.

Baseline effect of ethnic boundaries on colonial administrative borders by measurement period. Note. Results from the baseline specification for the pooled and period-wise samples. 95% Confidence intervals and estimate distributions result from a parametric bootstrap with 160 iterations

Estimated effects of the control variables in models that exclude measures of ethnic geography conform with qualitative evidence on the influence of geographic features on the design of administrative units in the colonial period (see Appendix A.7). In particular, rivers (but not watersheds) are frequent causes of district borders. Average population density along edges is positively associated with the likelihood of them being crossed by administrative borders, indicating smaller units in more densely populated areas.

Additional analyses show robustness to dropping and extending the vector of control variables (Appendix A.7). Importantly, estimating a linear model in Appendix A.11 allows for including vertex-level fixed effects. The stable results show that the main effect of ethnic boundaries is not explained by any local vertex-level characteristics such as, for example, the identity of an ethnic group, the presence of resources, or other non-relational geographic or climatic variables. Further evidence against reverse causality from endogenous identity formation beyond the LDV comes from the finding of similarly sized effects for small and large linguistic distances across ethnic boundaries (Appendix A.4).

An important confounder of ethnic boundaries might consist in the centralization of precolonial political institutions, in particular since stronger precolonial states might have successfully assimilated local ethnic minorities into a common identity (Chlouba, 2026). I capture this dynamic with a precolonial stateness index based on data on precolonial states from Paine (2019) and Wilfahrt (2025). Consistent with previous evidence on (in)direct rule (Müller-Crepon, 2020), precolonial statehood is significantly associated with larger districts in British but not French colonies. However, this pattern does not reduce the estimated effects of ethnic boundaries.

Disruption Through Dismemberment or Suffocation?

Analyzing the effects of dismemberment and suffocation of ethnic groups separately allows for gaining additional insights into the mode of disruption in cases where administrative borders do not align closely with ethnic boundaries. Descriptively, Figure 8 shows two substantive patterns in the incidence of dismemberment and suffocation. First, in (a), it appears that French colonies feature a 50% higher average level of ethnic dismemberment than British colonies (mean of 0.72 vs 0.47) but a similar level of suffocation (mean of 0.44 vs 0.47). Second, panel (b) plots the joint distribution of dismemberment and suffocation. In line with the idea of them being strategic substitutes for reasons of practicability, they tend to be negatively correlated: at least empirically, dismemberment and suffocation are most often applied to different areas and groups. However, French colonies feature many more particularly disruptive instances of combined dismemberment and suffocation than British colonies.

Figure 8.

Dismemberment and suffocation at the vertex-level across periods. Note. Red straight (dotted) lines result from a linear (LOESS) model

The results from the extended partition model in Table 3 underline significant differences between the degree to which administrative units dismember ethnic groups in the British and French empires. For the British sample, I find a large positive estimate of γ₁ in equation (4), supporting the argument that dismemberment decreases a partitioning’s realization probability by increasing its energy ϵ _i . Featuring low levels of dismemberment, British administrative units thus left ethnic groups less divided than one would expect conditionally on edge-level covariates. More specifically, decreasing the average dismemberment of ethnic groups by .25 in a toy example of 10 vertices would more than double its odds. This effect slightly increases when adding the lagged dependent variable in Model 2. The French sample, in contrast, shows a small cross-sectional estimate for dismemberment in Model 3 which turns negative (but statistically insignificant) when adding the lagged dependent variable in Model 4. These results suggest that ethnic groups are as dismembered as one would expect from administrative unit designs influenced solely by the covariates.

Table 3.

Ethnic Dismemberment or Suffocation?

​	British		French
	(1)	(2)	(3)	(4)
Constant	−9.84*	−8.94*	−10.74*	−6.54*
	[−10.52; −9.11]	[−10.19; −6.54]	[−11.29; −9.57]	[−7.83; −2.45]
Dismemberment	0.32*	0.44*	0.04	−0.15
	[0.20; 0.38]	[0.18; 0.68]	[−0.10; 0.14]	[−0.55; 0.29]
Suffocation	0.44*	0.27*	0.31*	0.27
	[0.38; 0.56]	[0.09; 0.50]	[0.24; 0.43]	[−0.03; 0.58]
Lagged dep. var.	​	0.81*	​	1.06*
	​	[0.68; 0.97]	​	[0.92; 1.29]
Controls	Yes	Yes	Yes	Yes
No. of vertices	10,418	1,662	8,060	1,010
No. of edges	28,084	4,493	21,132	2,584
No. of units	1,362	247	856	117

Notes. 95% confidence intervals from parametric bootstrap in parenthesis. ^* Statistically significant at the 95% level.

Second, I find only small and statistically insignificant differences between the suffocation of groups in the British and French samples. In both empires, administrative units were designed with low levels of suffocation conditional on edge-level covariates. The baseline estimate of suffocation is positive and only 25% smaller in the French than in the British sample. Adding the lagged dependent variable slightly decreases both estimates. In substantive terms, the estimate in the British Model 1 suggests that lowering suffocation by increasing the ethnic homogeneity of districts that cover a set of ten nodes by .25 increases a partitioning’s odds of realization by a factor of three.

These results are robust to dropping all or including additional covariates as well as when accounting for precolonial statehood (Appendix B). When only modeling either dismemberment or suffocation in separate models, estimates roughly double in the British (but not French) sample (Appendix B.2). This suggests a relatively strong negative correlation between both parameters. The same negative correlation also surfaces when shifting the exact location of the spatial graph to assess robustness to the design of the graph (Appendix B.6). While the British results are robust to such shifts, the exercise shows that, for French colonies, Table 3 tends to overestimate the extent of dismemberment and underestimate the degree of ethnic suffocation. Averaging across 100 shifts, the results show not only more ethnic dismemberment but also more suffocation in French than British colonies.

In sum, the empirical results suggest that the alignment of administrative borders with ethnic boundaries in British colonies is driven by districts featuring low levels of ethnic dismemberment and suffocation. In turn, French colonial cercles feature significantly more ethnic dismemberment while estimates of ethnic suffocation are sensitive to the precise location of the spatial graph.

Post-Colonial Effect Persistence and Change

An additional analysis that employs the full panel of post-colonial regional borders (Müller-Crepon, 2023) across former British and French colonies shows long-lasting persistence of the patterns of colonial administrative designs described above. Figure 9 shows the results of the baseline and LDV specifications ³⁰ estimated for bi-decadal time periods as well as the pooled post-colonial data.

Figure 9.

Effect of ethnic boundaries on the partitioning of post-colonial states into administrative regions. Note. Results from the baseline and LDV specification for pooled and approximately bi-decadal samples. The LDV model includes an indicator for the presence of colonial borders in addition to the lagged dependent variable. 95% Confidence intervals and estimate distributions result from a parametric bootstrap with 160 iterations

The results from the baseline model clearly show the differing levels of ethnic alignment with which former British and French colonies gained independence. With coefficients similar in size to those in the main analysis, administrative borders are significantly more in line with ethnic boundaries in countries gaining independence from the British as compared to the French empire. Over the years however, this difference in colonial legacies does not change much in size. This is because administrative border changes in former British and French colonies tend to, on average, follow ethnic boundaries. This is evidenced by the positive and statistically significant effect of ethnic boundaries in the LDV specification for the pooled samples which does not differ between countries with a French and British colonial past. Yet, when breaking down the effect into the different time periods, we observe that such ethnic alignments occur with significant temporal variation.

Discussion and Conclusion

The partitioning of states’ territories into administrative units is a crucial feature of territorial states with important outcomes on, among others, ethnic politics, economic development, and political stability. Going beyond previous literature that focuses on unit change, this paper has investigated the strategies underlying initial designs of administrative geographies. Administrative units can be drawn to align with ethnic groups, their social networks and political institutions, thereby preserving groups’ internal functioning. Or, administrative borders can be aimed at disrupting societies, dismembering groups through administrative division and/or suffocating them by creating diverse units. While the first strategy is cheap to implement, it leaves political power fragmented and decentralized. The second approach, in turn, is more costly in the short run but centralizes power at the expense of local society and its elites.

I have tested this argument by analyzing administrative unit designs in British and French colonies in Sub-Saharan Africa. While the British had a comparatively decentralized style of indirect rule, the French, for ideological and material reasons, used more direct rule to centralize power. My results show that the design of administrative units differed accordingly. Using newly collected data on administrative borders and employing ethnic geography, the results show that British colonial governments tended to design administrative units along ethnic boundaries, preserving ethnic groups by avoiding both dismemberment and suffocation. In turn, administrative borders in French colonies do not robustly correlate with ethnic boundaries, with the main difference to British designs consisting in comparatively high levels of administrative dismemberment of ethnic groups.

Taken together with findings on the effects of administrative units on ethnic groups’ geography (Müller-Crepon, 2025), ³¹ these findings point towards a dynamic co-development of administrative units and ethnic groups. The mindset of particularly British colonial rulers distributed power roughly along initially fuzzy ethnic lines and indirectly ruled through prevailing or newly invented ‘traditional’ institutions. This led to a crystallization of identities along borders drawn, as well as demands to change unit designs along sharpened ethnic lines (Grossman & Lewis, 2014), in turn likely strengthening ethnic identities. This dynamic alignment of administrative geographies and ethnic identities is fundamentally driven by ethnicized governance through neatly bounded and non-overlapping territorial divisions.

Yet, beyond one-sided evidence, this dynamic alignment between administrative units and ethnic groups remains empirically poorly understood. In addition to this article’s highlighting of general effects of ethnic geography on unit borders, mechanisms that could explain variation within empires should receive more attention. Further research should also examine alternative drivers of unit designs, such as colonial resource extraction or post-colonial state building. Lastly, future studies could study implications of endogenous unit designs for post-colonial politics and the importance of regional and ethnic cleavages (e.g., Boone, 2024; Huber, 2012; Müller-Crepon & Bormann, 2026).

More generally, my findings show that comparative research should take seriously the historical endogeneity of administrative units and the caveats it produces for comparative analyses. As more and more analyses are carried out at the subnational level (Pepinsky, 2019), administrative units gain increasing prominence as units of analysis, often for the simple reason that they act as ‘measurement containers’ for the data produced by states and other actors. My findings on the colonial and ethnic origins of administrative units show that they are endogenous to political processes in comparable ways as other attributes of states and their capacity (Suryanarayan, 2024). Going beyond the commonly referenced Modifiable Areal Unit Problem (e.g., Lee, Rogers, & Soifer, 2025), this introduces the risk of endogeneity bias. To avoid such biases, analysts should account for the impact of units’ origins on the social processes they study.

Supplemental Material

Supplemental Material - Ethnic Geography and the Colonial Design of Administrative Units in Sub-Saharan Africa

Supplemental Material for Ethnic Geography and the Colonial Design of Administrative Units in Sub-Saharan Africa by Carl Müller-Crepon in Comparative Political Studies