Tradeoffs in Type of Agent

Colonizers can choose to delegate substantial local authority to indigenous actors, a practice known as indirect rule. Within this broad mantle, colonizers decide between two ideal types of agent: traditional elites with preestablished local legitimacy and installed agents who lack this basis. Whereas traditionally legitimate agents tend to be more effective, they are also less beholden to the colonial administration—and therefore less inherently loyal.

Traditionally legitimate elites tend to have better information about their population, which can help to uncover plots of revolt before they manifest. Such actors also tend to be better at inducing the quasi-voluntary compliance needed to effectively collect direct taxes (Levi Reference Levi1989). However, these same sources of local legitimacy can also be used against the colonizer. Local elites are better positioned to themselves rebel or otherwise conspire against the colonizer (Finer Reference Finer1997, 66–70). Even after consolidating foreign rule, they can leverage various “weapons of the weak” that undermine colonial control (Abernethy Reference Abernethy2000; Scott Reference Scott1987).

The other ideal type of indirect-rule agent is an installed indigenous agent who lacks traditional legitimacy. One example is appointing an agent based in a nearby area, a common practice in French Africa. Another example is inflating the powers of an individual who traditionally lacked such powers, such as Warrant Chiefs in Southern Nigeria. Installed agents entail the converse pros and cons of traditionally legitimate agents. On the one hand, installed agents derive their elevated status from colonial control, which increases their loyalty toward colonial officials. On the other hand, the local population is more likely to resist foreign impositions. This lessens the effectiveness of appointed agents by reducing their ability to prevent revolts or extract direct taxes.

Alternatively, colonizers can eschew indigenous agents entirely and govern the territory more directly through official metropolitan agents or unofficial settlers. The main drawback of metropolitan agents is their high expense. Governing through European settlers entailed a somewhat different tradeoff (Lee and Paine Reference Lee and Paine2024), but possibilities for European settlement were geographically circumscribed in Africa (Acemoglu, Johnson, and Robinson Reference Acemoglu, Johnson and Robinson2001; Paine Reference Paine2019) and few areas had the scope of known mineral wealth present in other historical settings like Spanish America (Engerman and Sokoloff Reference Engerman and Sokoloff2011; Garfias and Sellars Reference Garfias and Sellars2026). Given the prohibitive costs of constructing a Weberian-style bureaucratic state, we focus primarily on choosing between the types of indirect-rule agents. In Appendix D of the Supplementary Material, we discuss more direct rule in areas with either substantial European settlement or extensive corporate presence, both among colonies within our statistical sample (Kenya Highlands, Northern Rhodesia Copperbelt, and parts of Bechuanaland) and outside (South Africa and Southern Rhodesia).

Another dimension we exclude from our theory is metropolitan ideology, in particular the inherent desire of metropolitan officials to effect transformation in their colonies. This is often cited as a key difference between British and French rule in Africa (Crowder and Ikime Reference Crowder, Ikime, Crowder and Ikime1970; Müller-Crepon Reference Müller-Crepon2020). In the concluding section, we engage with this consideration while also suggesting why local realities mattered independently of metropolitan ideology.

Optimal Choice of Agent

Given the tradeoff between loyalty and effectiveness, how does a colonizer choose between traditionally legitimate and installed agents? One key factor is the degree of centralization in precolonial institutions, as studied in existing work (Gerring et al. Reference Gerring, Ziblatt, van Gorp and Arevalo2011; Müller-Crepon Reference Müller-Crepon2020). Effective local administration requires economies of scale. This leads colonial officials to prioritize elites whose scope of authority extends beyond a small locality (e.g., only a single or small cluster of villages). If the colonizer secures basic promises of allegiance from local elites with a centralized basis, aspects of traditional administration can be repurposed under foreign rule. Furthermore, colonizers often have more hierarchical political institutions at home. Therefore, they better understand how to delegate authority to elites who themselves command allegiance under hierarchical political institutions. For these reasons, we concur with much existing work by anticipating that in areas with centralized political institutions, colonizers would delegate to traditional elites. Exceptions should be confined to places that exhibited especially strong resistance to the colonial imposition; thus, the disadvantage of proven disloyalty would outweigh the advantage of greater effectiveness.

Historically decentralized areas pose greater administrative challenges. Existing work posits that colonizers necessarily install nontraditional agents like Warrant Chiefs; if no king could be found, they would make one. This implication is logical in the low-information environment of early colonialism, when traditional societal structures were illegible to the colonizers. Misunderstandings of traditional sources of authority and constraints in African societies would lead to the installation of nontraditional agents as the default choice. Even when the local population disapproved of this choice, military repression campaigns to install early administrations would discourage movements against colonially chosen agents.

However, nontraditional agents are an expedient choice under narrower conditions than implied by existing theories. Over the longer term, we anticipate that various types of pressures would reopen the choice of loyal-but-ineffective agents. Installed agents may prove hopelessly ineffective as the local population turns against invented despots who lack local accountability. Moreover, socioeconomic conditions change over time. Revolts and other forms of agitation by colonial subjects create pressure on colonizers to pivot to another type of agent.

In decentralized areas, the scope of choice over colonial agents was broader than existing work—or early colonial officials—realized. Institutions that replicate certain aspects of traditionally legitimate institutions could constitute a viable alternative to installed agents. Many historically decentralized areas had complex institutions that empower disparate members of society through village councils, age grades, purchased titles, and secret societies (McIntosh Reference McIntosh and McIntosh1999). Foreign institutions often create prohibitive transaction costs for colonizers with limited knowledge of newly conquered territory. However, after gaining experience and learning over time, the colonizers can seek to restructure earlier institutions in ways to achieve economies of scale. For example, they can scale up councils to encompass larger areas than single villages, therefore achieving aims of fiscal efficiency. The drawbacks of autocratic installed agents increase the relative attraction of delegating authority to agents with stronger local accountability.

Socioeconomic changes reinforce the desire to reform earlier local institutions. Incorporation into a larger political structure through territorial expansion or constructing an empire inevitably yields changes. For example, traditional indigenous agents lose some powers while gaining others, and they will strategically manipulate indigenous institutions to align with the new principal–agent relationship created by the ruling power. The promotion of new modes of economic output such as cash crops and cultural influence from the metropole (e.g., religious missionaries) will also create new elites (Pengl, Roessler, and Rueda Reference Pengl, Roessler and Rueda2022). These new voices create novel demands for incorporation into local institutions. Local governance through expanded councils is better able to accommodate such demands than autocratic governance through installed agents. For similar reasons that institutions of accountability are lauded for promoting better governance in Western societies, locally accountable agents can serve colonizers’ goals of minimizing revolts and collecting revenue.

In sum, we explain why colonizers will often choose to delegate governance to traditionally legitimate agents. This implication includes, but is not limited to, areas with centralized precolonial states. In historically decentralized areas, colonizers should delegate to agents with traditional sources of legitimacy (including scaled-up councils) when their advantages in effectiveness are accentuated relative to the alternative of installed agents. Scope conditions that favor traditionally legitimate agents include prior governance failures by autocratic installed agents and socioeconomic transformation, each of which would prompt colonizers to learn more about sources of authority in precolonial institutions amid heightened pressure from local actors.

Unit of Analysis: Native Treasuries and Digitized Maps

The British administrative scheme featured, in descending order of size: provinces, divisions/districts, Native Treasuries (NTs), and Native Authorities (NAs). In our main dataset, the unit of analysis is the NT. Our statistical sample consists of 458 NTs across British Africa, including 201 in Nigeria, 87 in the Gold Coast (Ghana), 52 in Tanganyika (Tanzania), 42 in Northern Rhodesia (Zambia), 25 in Kenya, 16 in Nyasaland (Malawi), 12 in Uganda, 12 in the Gambia, 9 in Bechuanaland (Botswana), and 1 in each of Lesotho and Swaziland (Eswatini).

Although we measure colonial institutions for each Native Authority, we use Native Treasuries as the unit of analysis. Fiscal authority, a key component of the decision-making powers of NAs, was exercised through NTs. NTs were thus the fiscal units to which British assessments about the scale of effective authority applied. In practice, the distinction makes little difference for our main outcome variable because NTs containing multiple NAs exhibited limited institutional variance (we later provide details).

We sourced and digitized a comprehensive set of maps of NTs for essentially our entire sample.Footnote ⁵ We compiled these maps from the colonial archives and other available sources; only for the small colony of the Gambia were we unable to uncover a disaggregated source map. In some colonies (e.g., Kenya and Uganda), almost every NT area was synonymous with a district, and thus these maps align with ones digitized in prior work (Müller-Crepon Reference Müller-Crepon2020; Ricart-Huguet Reference Ricart-Huguet2022). But in the two colonies with the largest number of NTs, Nigeria and the Gold Coast, most districts contained numerous NTs. Using varied source maps, we constructed a one-to-one mapping between NTs and polygons that we digitized using our more detailed colonial maps. For some NTs in these two colonies, the historical maps did not provide a corresponding delineated area. In such cases, we created a proxy by digitizing a small radius around a point corresponding to the location of the NT, which we identified by locating the main town associated with the NT in the source maps and cross-checking it with Google Maps. Thus, building on the pioneering efforts of Bolt and Gardner (Reference Bolt and Gardner2020a; Reference Bolt and Gardner2020b), we constructed a comprehensive set of maps of British administrative units at a finer level of detail than in existing work. The spatial data enable us to measure socioeconomic conditions for each NT and assess spatial dependence across units.

Precolonial Institutions

We constructed a new, disaggregated measure of precolonial political institutions.Footnote ⁶ This measure improves substantially upon existing ethnic-group level measures of historical institutions, as we discuss in Appendix A.2 of the Supplementary Material. To distinguish between states and decentralized areas, we extended a recent dataset of precolonial African states from Paine, Qiu, and Ricart-Huguet (Reference Paine, Qiu and Ricart-Huguet2024; Reference Paine, Qiu and Ricart-Huguet2025), yielding a list of 58 states. Our main innovation was the next step: measuring whether institutional constraints were present for each state. Based on extensive historical and anthropological sources, we coded a binary variable for whether a council imposed effective institutional constraints on the ruler around the eve of colonization. To determine this, we sought to answer the following types of questions using our sources, with affirmative answers indicating an effective council. Did the ruler regularly consult a council? Did a council regularly influence policy decisions? Was the ruler unable to regularly override the desires of the council? The scholarly literature suggests many ways in which rulers could be constrained and made accountable, but we chose this definition because it is concrete and relatively straightforward to measure. We consulted the dozens of relevant volumes of the Ethnographic Survey of Africa (edited by Daryll Forde), which we cross-checked using scores of additional books and articles about individual cases. Finally, we matched each precolonial state to an NT, which was straightforward because of the high overlap in names.

Overall, we score precolonial institutions within each NT area as either (a) state with an authoritarian ruler, (b) state with a ruler constrained by an influential council, or (c) decentralized. The latter category encompasses NTs without a corresponding precolonial state. Although we do not directly measure institutions in areas with decentralized institutions, extensive anthropological and historical evidence catalogs the prevalence of political constraints on rulers and of village governance entirely through councils (later we provide examples from specific cases). Thus, we treat decentralized institutions as synonymous with constrained leaders. On this dimension, these areas resembled states with constrained rulers.

Colonial Native Authority Institutions

We compiled numerous new variables that capture colonial-era institutions and conditions. Our primary sources are the five volumes of Lord Hailey’s Native Administration in the British African Territories (Hailey, Reference Hailey1950a; Reference Hailey1950b; Reference Hailey1951a; Reference Hailey1951b; Reference Hailey1953) and the comprehensive surveys of District Officers that Hailey used to construct these volumes, which we accessed from the U.K. National Archives in London.Footnote ⁷ The Hailey surveys constituted the first attempt to systematically characterize Native Administration in British Africa. This enabled us to develop appropriately local measures of indirect-rule institutions in the 1940s, reflecting reforms which took place over the course of the previous decade.

We classify the following institutional structures, which differ in whether a chief and/or council was legally recognized in the colonial Gazettes as part of the Native Authority.Footnote ⁸

• Solo chief (equivalently, Sole NA). Example: Emir of Kano in Northern Nigeria. He had an advisory council, but this body was not legally recognized as having powers as part of the NA.

• Chief and council. Example: Ada Manche and the State Council in the Gold Coast.

• Council only. Example: the Ndoki Clan Council in Eastern Nigeria. For this and the preceding two types, we code subnational institutions based on the structure of the Superior NA, although many encompassed lower-level Subordinate NAs that lacked the full powers of Superior NAs.

• Federal council. These consisted of a higher-level NA council that controlled the NT and multiple lower-level NAs (typically chiefs, who retained various powers delegated to authorities at a more local level). Examples: Local Native Councils (LNCs) in Kenya, federal councils of chiefs in Tanganyika and Nyasaland, and District Councils in Eastern Uganda. For our main analysis, we group federal-council NAs with council-only NAs. Such cases are coded solely based on the higher-level council because this body coincided with the NT, our unit of analysis. However, we also present a robustness check (Table B.7 in the Supplementary Material) in which we reclassify these cases as chief-and-council NAs, which reflects the institutions for the lower-level NAs as well.

For NTs that encompassed multiple NAs, we compute the fraction of NAs with each type (e.g., an NT encompassing two NAs with solo chiefs and two with chief-and-council NAs would be 0.5 on each of these variables and 0 on council only). Although many NTs contained multiple NAs, institutional variance within NTs was minimal: in 443 of 458 NTs (97%), the NT is coded as entirely solo chief, chief and council, or council only.

We additionally compiled information about the membership of councils, the creation of finance committees that enabled Africans to participate in the budgeting process, and the composition of expenditures by NTs.Footnote ⁹

Socioeconomic Variables

Socioeconomic changes play an important role in our theory, as we anticipate that colonial officials would target more developed areas when they implemented reforms to improve the fiscal efficiency of local governance units. We operationalize socioeconomic changes with the following variables, which (if not otherwise specified) are measured around mid-century.Footnote ¹⁰

Cash crops, where present, were typically the main source of economic value and export revenue (Frankema and Booth Reference Frankema and Booth2020; Pengl, Roessler, and Rueda Reference Pengl, Roessler and Rueda2022). Early British administrators sought to raise revenue and targeted long-standing cash-crop products while additionally trying to develop new ones. Thus, the value of cash crop production within each NT area directly measures an important component of the area’s development. Colonial railways were among the most significant colonial investments. They were deliberately constructed to connect areas of high economic value and, in turn, spurred significant socioeconomic change (Jedwab, Kerby, and Moradi Reference Jedwab, Kerby and Moradi2017). We capture this effect by measuring the distance between the centroid of each NT polygon and the nearest railway. We also control for distance from the capital and distance from the coast. These are, however, less direct measures of social economic development and potential because the capital was not always the only (or even main) economic center, and the relative distance from the coast was less important within inland colonies. Population density could have influenced the type of colonial institutions (Acemoglu, Johnson, and Robinson Reference Acemoglu, Johnson and Robinson2002; Engerman and Sokoloff Reference Engerman and Sokoloff2011) and the difficulty of constructing effective state institutions (Herbst Reference Herbst2000). Absolute population size, although less obviously important, could have affected prospects for constructing economies of scale.

European special interests also affected local economies and governance. The presence of Christian missions could reflect existing patterns of socioeconomic development and stimulate further change (Jedwab, Meier zu Selhausen, and Moradi Reference Jedwab, Meier zu Selhausen and Moradi2022), particularly in the case of missions founded earlier in the colonial period. We capture early missionary presence with a binary indicator for whether the NT area encompassed a mission site as of 1914. European settlers, where present, were another influential special-interest group, and we capture their presence with the percentage of European-alienated land. However, as we discuss in Appendix D of the Supplementary Material, the areas with the highest levels of alienated land were excluded from Native Administration entirely, which creates issues of sample selection for interpreting the coefficient estimates for this variable.

Panel Data on the Introduction of Councils

Our theory anticipates changes over time in the composition of indirect-rule institutions. We constructed a secondary dataset consisting of annual observations for the introduction and presence of NA councils between the 1920s and 1940s, using the somewhat more highly aggregated unit of the district ( $ N $ = 238). We used the Hailey volumes and surveys as well as numerous additional colony-specific sources to fill in dates as precisely as possible, for which we provide extensive details in Appendix C of the Supplementary Material.

Overview of Patterns

By the time of the Hailey surveys in the late 1940s, solo-chief NAs were rare—only 17% of all NAs, as shown in Table 2. The remaining NAs included councils, either chief and council (36%) or council only (47%). This observation contrasts starkly with the standard portrayal of “decentralized despotism” and unchecked chiefs everywhere, and also underscores the extreme heterogeneity in indirect-rule institutions. NAs without a council (solo-chief NAs) pervaded NT areas that encompassed authoritarian precolonial states (90%), but were rare in areas with a constrained precolonial state (9%) or historically decentralized institutions (12%). NAs without a chief (council-only NAs) were common in historically decentralized areas (58%), but completely absent from areas containing a precolonial state—all of which contained a chief as part of their NA. As a corollary of these two patterns, NT areas with constrained precolonial states usually preserved both the chief and the council elements of their precolonial institutions, as 91% had chief-and-council NAs.

Table 2.

Cross Tabulations: Indirect-Rule Institutions

Table 2. Long description

The table consists of five columns and five rows.

Column headers from left to right are: Precolonial forward slash colonial, Solo chief, Chief and council, Council only, and Totals.

Row 1: Authoritarian state. Solo chief 90 percent, Chief and council 10 percent, Council only 0 percent, Totals 7 percent.

Row 2: Constrained state. Solo chief 9 percent, Chief and council 91 percent, Council only 0 percent, Totals 12 percent.

Row 3: Decentralized. Solo chief 13 percent, Chief and council 30 percent, Council only 58 percent, Totals 81 percent.

Row 4: Totals. Solo chief 17 percent, Chief and council 36 percent, Council only 47 percent, Totals N equals 458.

The data indicates a strong correlation between authoritarian states and solo chief institutions, constrained states and chief and council institutions, and decentralized states and council only institutions.

Note: The sample is our main dataset, a cross section of Native Treasuries in the late 1940s. Tables A.2 and A.3 in the Supplementary Material provide more detailed summary statistics.

Figure 1 depicts precolonial and colonial institutions for four colonies—Nigeria, Kenya, Gold Coast, and Uganda—which, collectively, exhibited substantial variation on the main variables. These are the main cases for which we provide supporting qualitative evidence, although we present maps and discuss the remaining colonies in Appendix C.6 of the Supplementary Material.

Figure 1.

Maps of Precolonial and Colonial Institutions

Note: The units in this and the following figures with maps are Native Treasury areas (Figure C.3 in the Supplementary Material presents maps for the other colonies in our statistical sample). For all precolonial units and almost all colonial units, the institutional variable is binary. A small number of NTs, though, contained multiple NAs that varied in their institutional type. In such cases, for the colonial maps, the darkest blue indicates that more than half of the NAs were solo chiefs; and, if half or more had councils, the shade of blue indicates whether chief-and-council or council-only NAs were more prevalent.

Figure 1. Long description

A multi-panel set of geographical maps divided into four sections: a Nigeria, b Kenya, c Gold Coast, and d Uganda. Each section contains two maps: Precolonial on the left and Colonial on the right. The maps use shades of blue to represent institutional types: light blue for Decentralized or Council-only N A, medium blue for Constrained state or Chief-and-council N A, and dark blue for Authoritarian state or Chief-only N A. White indicates No Native Administration.

* Panel a Nigeria: The Precolonial map shows a high density of dark blue (Authoritarian) in the North and light blue in the South. The Colonial map shows an expansion of dark blue across the North and Central regions, with medium blue appearing in the Southwest.

* Panel b Kenya: The Precolonial map is almost entirely light blue (Decentralized). The Colonial map shows a significant shift to dark blue (Authoritarian) in a vertical band through the Central and Eastern regions, while the West remains light blue.

* Panel c Gold Coast: The Precolonial map shows a mix of medium blue in the Central-South and light blue elsewhere. The Colonial map shows a consolidation of medium blue (Constrained state) across nearly the entire territory.

* Panel d Uganda: The Precolonial map features a large medium blue area in the South-Central region with light blue in the North. The Colonial map maintains a similar distribution but with more defined medium blue boundaries in the South and Southwest.

Nigeria exhibits the greatest amount of within-colony variation. As the map shows, the authoritarian precolonial states in the North all had solo-chief NAs, the constrained states in the West all had chief-and-council NAs, and council-only NAs were nearly universal among the decentralized areas in the East. In Kenya, the predominant pattern is council-only NAs in historically decentralized areas. In the Gold Coast, chief-and-council NAs were nearly universal amid a precolonial landscape comprised of constrained states of varying sizes as well as decentralized areas. In Uganda, the southwestern areas with precolonial states preserved chiefs in their Native Administration, although formerly authoritarian rulers were recognized alongside their councils under British rule. This contrasted with council-only NAs in the decentralized East and North.

The presence of NA councils by the late 1940s reflected changes over time, as summarized in Figure 2. Early British administration in Africa was characterized by the persistence of chiefs in areas of precolonial states and the installation of chiefs elsewhere. However, revolts and socioeconomic changes made installed chiefs increasingly untenable, and the British drew from precolonial precedents when reforming governance institutions. Over time, the prevalence of councils diverged between areas that traditionally had constraints on rulers and those that did not (see also Table C.1 in the Supplementary Material).

Figure 2.

Councils in Native Authorities over Time

Note: The sample is our secondary dataset, an annual panel of colonial districts. The lines are averages for each category of precolonial institutions for every year between 1923 (one year before reforms became widespread, starting in Kenya) and 1949 (Hailey surveys). Years during WWII are shaded gray. Appendix C of the Supplementary Material details the data and Table C.1 in the Supplementary Material summarizes changes over time for individual regions and colonies.

Figure 2. Long description

The Y-axis is labeled Fraction of districts with a Native Authority council, ranging from 0 to 1. The X-axis represents years from 1923 to 1949. A gray shaded vertical region covers the years of W W I I, roughly 1939 to 1945.

* Authoritarian state: Represented by a flat black horizontal line at approximately 0.18, showing no change over the entire period.

* Constrained state: Represented by a red line. It starts at approximately 0.6 in 1923, remains flat until the mid-1930s, then rises to approximately 0.68. It remains flat through the W W I I period and then rises sharply after 1945, ending at approximately 0.9 in 1949.

* Decentralized: Represented by a blue line. It starts at the lowest point, approximately 0.05 in 1923, and shows a steady, staggered upward trend. It crosses the Authoritarian line around 1925 and continues to rise, reaching approximately 0.65 by the start of the gray shaded W W I I period. It continues a slow upward trend during the war and ends at approximately 0.82 in 1949.

Regression Models

Using our main dataset of NTs, we confirm a robust correlation between precolonial institutions and NA institutions using regression models that account for various confounders. In Table 3, the DV in Columns 1–3 is solo-chief NAs, and in Columns 4–6 is chief-and-council NAs. We do not run models using council-only NAs as the DV because, as shown in the cross-tabulations, such institutions were exclusive to decentralized areas (hence creating collinearity with the intercept term in the model). Using OLS, we estimate

$$ \begin{array}{c}{\mathrm{Native}\ \mathrm{Authority}}_i={\beta}_0+{\beta}_P\cdot {\mathrm{Precolonial}\ \mathrm{Institutions}}_i\\ {}+{\beta}_X\cdot {X}_i+{\epsilon}_i.& & \end{array} $$

Table 3.

Correlates of Native Authority Institutions

Table 3. Long description

A statistical table with two main dependent variables: Solo-chief N A (Models 1 to 3) and Chief-and-council N A (Models 4 to 6).

Key independent variables and their coefficients include:

* Decentralized: Negative coefficients across all models, ranging from -0.313 to -0.813, all significant at p < 0.01.

* Constrained state: Negative coefficients for Solo-chief N A models (-0.812, -0.718, -0.646), all significant at p < 0.01.

* Authoritarian state: Negative coefficients for Chief-and-council N A models (-0.812, -0.375, -0.534), all significant at p < 0.01.

* Population: 0.0495 for Model 3 and -0.0680 for Model 6, both significant at p < 0.01.

* Mission in 1914: -0.162 for Model 3 and 0.220 for Model 6, both significant at p < 0.01.

* Intercept: Positive values across all models, ranging from 0.351 to 1.195.

Summary statistics at the bottom:

* Native Treasuries: 458 for most models, 446 for Models 3 and 6.

* Provinces (clusters): 60 for most models, 59 for Models 3 and 6.

* R-squared: Ranges from 0.195 (Model 4) to 0.568 (Model 5).

* Colony F E: Included in Models 2 and 5.

Standard errors are provided in parentheses and Conley standard errors in brackets below the coefficients.

Note: The sample is our main dataset, a cross section of Native Treasuries in the late 1940s. All models are estimated using OLS. The precolonial variables are binary. Every socioeconomic development variable is logged except the binary indicator for any Christian mission in 1914. Province-clustered standard error estimates for all variables are reported in parentheses. Conley standard error estimates for the precolonial variables are in brackets; the distance cutoff is 1 $ {}^{{}^{\circ}} $ lat × 1 $ {}^{{}^{\circ}} $ long, or approximately 70 sq.mi. at the equator. All NTs in the Gambia are dropped in the models with covariates, and the Gambia, Swaziland, and Lesotho are dropped in the calculation of Conley standard errors. *** $ p<0.01 $ , ** $ p<0.05 $ , * $ p<0.1 $ .

In Columns 1–3, the omitted reference category for Precolonial Institutions is authoritarian state, which reflects our theoretical expectation that solo-chief NAs would be less prevalent in NT areas with either type of precolonial constraints. In Columns 4–6, the omitted reference category for Precolonial Institutions is constrained state, which reflects our theoretical expectation that chief-and-council NAs would be less prevalent elsewhere.

Columns 1 and 4 are the baseline specifications that contain the precolonial variables only. We add different covariates, $ {X}_i $ , in the subsequent models. In Columns 2 and 5, we control for colony fixed effects to account for idiosyncratic differences in the implementation of Native Administration ordinances across colonies. In Columns 3 and 6, we add the socioeconomic variables. We adopt a conservative approach in all models by estimating standard errors at the province level ( $ N $ = 60). We separately estimate Conley standard errors. Both procedures address concerns about spatial autocorrelation, which we discuss in more depth later.

The regressions confirm that the discrepancies previewed in the cross-tabulations remain large in magnitude and statistically significant when accounting for possible confounders. Columns 1–3 are the most important specifications for our thesis, as they demonstrate that areas with precolonial constraints were much less likely than areas with authoritarian states to have unconstrained solo chiefs. Thus, precolonial constraints were usually replicated with some form of empowered council under British Native Administration. Columns 4–6 demonstrate that areas with constrained precolonial states were unique, but for heterogeneous reasons: more likely than areas of authoritarian precolonial states to have councils as part of their NA, and more likely than historically decentralized areas to have chiefs as part of their NA.

Spatial Autocorrelation

As shown in the maps, precolonial institutions exhibited high spatial autocorrelation. If our theory is correct, we would therefore also anticipate high spatial autocorrelation in colonial institutions. Nonetheless, introducing a colonial council to one local governance unit could spill over to influence prospects for introducing councils in neighboring governance units. This indirect effect, as opposed to the direct effect of precolonial institutions that we attempt to estimate, creates a specific confounding concern.

Our main models partially address the nonindependence of NT units by clustering standard errors at the higher level of the province, and we also estimate Conley standard errors. Here, we directly account for nonindependent observations by estimating spatial autoregressive models. In Panels A and B of Table 4, we show that adding a spatial lag to the cross-sectional models attenuates somewhat the estimates for the main coefficients, but they remain large in magnitude and statistically significant. In Panel C, we incorporate a dynamic component into the spatial models by using our annual panel data on the introduction of councils, which enables us to include a longitudinal lag for institutions in neighboring areas. Decentralized areas and constrained states each continue to be distinguished from authoritarian states when accounting for these additional sources of confounding.

Table 4.

Spatial Autoregressive Models

Table 4. Long description

The table presents regression results for three dependent variables: Solo-chief N A (columns 1 to 3), Chief-and-council N A (columns 4 to 6), and Council in N A panel (columns 7 to 9).

Key independent variables and their coefficients include:

* Decentralized: Negative coefficients ranging from -0.284 to -0.503 in columns 1 to 6; positive coefficients from 0.392 to 0.417 in columns 7 to 9. All are significant at p < 0.01.

* Constrained state: Negative coefficients from -0.388 to -0.436 in columns 1 to 3; positive coefficients from 0.256 to 0.563 in columns 7 to 9.

* Authoritarian state: Negative coefficients from -0.351 to -0.618 in columns 4 to 6.

* Population: 0.024 in column 3; -0.067 in column 6; 0.054 in column 9.

* Population density: -0.012 in column 3; 0.004 in column 6; -0.040 in column 9.

* Value of cash crops: -0.010 in column 3; 0.006 in column 6; 0.010 in column 9.

* % alienated land: -0.012 in column 3; 0.073 in column 6; 0.040 in column 9.

* Distance from railway: 0.017 in column 3; 0.065 in column 6; -0.066 in column 9.

* Distance from capital: 0.010 in column 3; -0.059 in column 6; 0.046 in column 9.

* Distance from coast: -0.010 in column 3; 0.044 in column 6; -0.060 in column 9.

* Mission in 1914: -0.118 in column 3; 0.155 in column 6; 0.082 in column 9.

Model diagnostics at the bottom indicate the number of Native Treasuries is 446 for columns 1 to 6, while columns 7 to 9 include 6,426 district-years and 238 districts. Checkmarks indicate the inclusion of L D V, Colony F E, Covariates, Error lag, and Year F E across various models. Standard errors are provided in parentheses below coefficients.

Note: Columns 1–6 use our main dataset (cross section of Native Treasuries) and replicate the respective columns from Table 3. Here, we estimate spatial autoregressive models, with the spatial lag of the dependent variable specified using a spatial weighting matrix that assigns an identical positive weight for any NTs that shared a common border, and 0 otherwise; and the weights sum to 1. We estimate these models using maximum likelihood and present heteroskedasticity-robust standard error estimates in parentheses. Columns 7–9 use our secondary dataset, an annual panel of districts between 1923 and 1949. The DV equals 1 in any year including and after the first in which a Native Authority within the district had a legally recognized council, and 0 otherwise. These are GLS random effects models with spatial lags for the dependent variable and error term, plus year fixed effects. *** $ p<0.01 $ , ** $ p<0.05 $ , * $ p<0.1 $ .

Decentralized Areas

Case evidence from Nigeria, Kenya, the Gold Coast, and Uganda provides additional confirmation for our hypotheses. These are the four largest colonies in our sample by populationFootnote ¹¹ and, collectively, exhibited substantial variation on both precolonial and colonial institutions. The following also provides process-tracing evidence on the dynamics of institution building throughout the colonial period and the emphasis the British placed on identifying traditionally legitimate institutions while simultaneously scaling up fiscal authority.

Political organization in precolonial Southern Nigeria, east of Yorubaland, was characterized by hundreds of independent villages in densely populated areas, in which authority was concentrated in village elders who did not hold hereditary positions. The decentralized institutions among the Igbo and neighboring groups were initially illegible to the British, who granted warrants for individuals to serve on the Native Courts. Many of these Warrant Chiefs were arbitrarily selected, and even those with some traditional standing gained powers well beyond what existing customs permitted (Afigbo Reference Afigbo1972). In response to the anti-tax revolts in Warri and Aba in the late 1920s, mentioned earlier in the historical background, the British collected over three hundred Intelligence Reports to learn about traditional institutions throughout Eastern Nigeria and adjacent areas. Colonial reports noted that the Warrant Chiefs, or Native Courts, would not provide a suitable basis for constructing better NAs: “Attempts have been made to utilise in part the machinery of Native Courts already in existence, but these, being very largely alien institutions unsupported by native custom, have been found unsuitable as a foundation on which true Native Administrations can be built.”Footnote ¹²

Instead, new NAs were deliberately built up from the village level. By the mid-1930s, the scores of NTs corresponded with new clan and district councils (Hailey Reference Hailey1951a, 159–60; Ikime Reference Ikime1969, 220–63). British administrators were often surprised at the resiliency of precolonial institutions, despite decades of colonial rule. In Warri, for example, the local African response to the governance void created by the anti-tax riots was that “village councils composed of Ekpako [village elders] and titled men were ruling the towns. It thus became clear that the operation of the Native court [Warrant Chief] system had not succeeded in destroying the indigenous political institutions. …The ease with which they revived showed a resilience which is not often credited to them” (Ikime Reference Ikime1969, 227). Councils of village elders successfully collected the taxes with which they were tasked (229, 241), which met the aims of the British administration. Beyond Warri, British administrators were pleased that “[a]t the present stage the newly-organised councils are beginning to find their feet and to gain confidence, and it is a significant fact that in areas where, in the past, tax could only be collected with difficulty under direct European supervision, it has this year been collected in full by the Councils without any extraneous aid.”Footnote ¹³

Following the reforms, the main break from precolonial precedents was the much greater scale of authority than traditional village councils had wielded, as the new NA and Treasury institutions scaled up from village councils to higher-level councils for the clan or division. The British explicitly encouraged scaling up for purposes of administrative efficiency, noting the “marked tendency towards a reduction of the numbers of representatives composing these administrative and judicial bodies, and an amalgamation of the small units into larger ones which can be given a higher degree of responsibility.”Footnote ¹⁴

In Kenya, the predominant form of traditional institutions among major groups such as the Kikuyu and Kamba was age grades, which consisted of relatively small lineage groups who rotated power within the group. Headmen initially appointed by the British, who were not subject to traditional constraints, often governed in an authoritarian manner (Tignor Reference Tignor1971). But protests in the early 1920s prompted reforms. Starting in 1924, the British began supplanting the executive powers of the headmen by introducing LNCs. Scaled up to the district level, LNCs administered much larger jurisdictions than those of headmen; each district, on average, encompassed over 15 headmen. They were intended to be what Schilling (Reference Schilling1976, 221) describes as “safety valve, an acceptable framework for African political expression.” LNCs had expanded powers and responsibilities (e.g., setting local rates) and incorporated a range of voices beyond the headmen, particularly those of mission-educated Africans and local businessmen.Footnote ¹⁵

Many local leaders in Eastern Uganda were vassals to the neighboring states of Buganda and Bunyoro before the onset of British colonialism. The process of scaling up smaller traditional political institutions began shortly after the turn of the century, when agents from the neighboring state of Buganda implemented reforms in Busoga and neighboring areas. “The larger of the Soga kingdoms became saza (counties) on the Ganda model, while smaller ones were amalgamated as gombolola (sub-counties) …Following modern Ganda practice, a muruka (parish) unit with a chief was introduced between the sub-county and the village levels” (Fallers Reference Fallers and Richards1960, 84–5; see also Roberts Reference Roberts1962, 450). In 1919, a Native Administration headquarters was created for all of Busoga, which consisted of a council headed by a Soga king who served as President, a nonhereditary position (Fallers Reference Fallers and Richards1960, 84; Hailey Reference Hailey1950a, 28). During the 1930s, the colonial administration embarked on a process of reorganization aimed at amalgamating smaller chieftaincies.Footnote ¹⁶ In 1936, the new scaled-up councils gained a firmer legal footing with the introduction of a tiered system of councils building up to the county level from the lower-levels gombolola and muruka units.

Constrained States

Many areas governed by more centralized states had a tradition of constrained rulers. In the major precolonial states in Yorubaland (Western Nigeria), obas (rulers) claimed divine powers. However, in practice, each oba consulted with a council of elders on all important decisions. Early British rule broke this relationship. When Native Administration was formally introduced to Western Nigeria in 1917, the major Yoruba rulers were named as first- or second-class chiefs but their traditional councils were disregarded, at least in the British Gazettes. Footnote ¹⁷ The most notorious case was the Alafin of Oyo, whose powers were inflated well beyond traditional limits with British support (Atanda Reference Atanda1973).

By the time of the Hailey surveys, though, a council was recognized as part of the NA in every major Yoruba state. The pivot away from authoritarian chiefs began in tandem with the reforms in the 1930s to supplant the Warrant Chiefs. The shift toward councils reflected the new British emphasis that “the doctrine that the jurisdiction of any Native Authority must be based on the consent of the people over whom such authority would be exercised,” as they feared a repeat of the anti-tax revolts in Warri and Aba in the late 1920s (Atanda Reference Atanda1973, 249). “[B]ecause the people became dissatisfied with the new order, they evinced a desire to return to a system which had operated successfully under different conditions, i.e., the pre-colonial system” (Ikime Reference Ikime1968, 436–8). Moreover, the centralized Yoruba states differed from the decentralized parts of Western Nigeria, many of which had council-only NAs.Footnote ¹⁸

In the precolonial Gold Coast, a large number of Akan states consisted of a hereditary Omanhene (ruler) who controlled the state administration alongside his Council of Elders; and, for some decisions, a broader State Council (Manoukian Reference Manoukian1950, 35). The largest state was the Asante Empire, which the British originally dissolved. The subsequent restoration of the Asante Confederacy in 1935, consisting of the Asantehene (ruler) and his council, reflected the inability to find a satisfactory alternative form of Native Administration (Tordoff Reference Tordoff1965). Farther south, the British complained about the fiscal ineffectiveness of smaller Akan states (e.g., Fanti) and engaged in an extensive process of scaling up these units.Footnote ¹⁹

Authoritarian States

The structure of NAs in Northern Nigeria was more authoritarian throughout the colonial period, but this reflected precolonial precedents. The Sokoto Caliphate originated in the early nineteenth century when jihadist cavalrymen conquered long-standing Hausa states. They subsequently formed roughly two dozen emirates, each of which had substantial local autonomy but was tied to the center at Sokoto through annual tribute and claims of legitimacy. Hereditary emirs faced few constraints; councils existed, but tended to have little influence. Under British rule, the emirs of leading Sokoto emirates (including Sokoto, Gwandu, and Kano) became first-class chiefs, and every emir was recognized at least as a second-class chief. Throughout the colonial period, these solo-chief NAs had councils which were advisory rather than legally recognized.Footnote ²⁰

Even in Northern Nigeria, though, NA institutions were not uniformly authoritarian. The Sokoto Caliphate shaped British territorial claims, patterns of conquest, and the organization of provinces. Nonetheless, many parts of Northern Nigeria were decentralized. The Sokoto emirs confronted hostile enclaves populated mostly by non-Muslims concentrated in mountainous regions like the Jos Plateau. Despite persistent warfare and slave raiding, certain areas successfully resisted control by any emirate. Externally, the limits of Sokoto expansion were shaped not only by other states (e.g., Bornu in the northeast and Borguland in the southwest), but also by smaller states and decentralized peoples like the Tiv.

NAs in the historically decentralized parts of Northern Nigeria experienced substantial changes in the 1930s—like their counterparts in Southern Nigeria, but unlike the Sokoto emirs who retained their positions as solo chief NAs. In the assessment of British administrators, the emirate system worked in areas that historically had such institutions but failed in what they called the “pagan” areas. “The objective during the latter months of [1933] was therefore to endeavour to penetrate this veneer and discover the true forms of government among the numerous pagan tribes.”Footnote ²¹ For example, the Tiv were organized by small-scale units, such as descent groups and lineages. British administrators initially misunderstood traditional institutions and tried various schemes over time to impose authority and collect taxes. They eventually recognized councils throughout the roughly three dozen localities and created a scaled-up Native Authority council populated by local clan fathers (Dorward Reference Dorward1969). Following reforms throughout the historically decentralized areas of Northern Nigeria in the 1930s, two-thirds of NT areas contained at least one NA with a council (Figure 3).Footnote ²²

Figure 3.

Councils in Decentralized Parts of Northern Nigeria

Figure 3. Long description

A map of Nigeria is divided into internal administrative boundaries. A legend in the bottom-right corner defines five categories.

* Precolonial state: Represented by white or very pale off-white, this covers the vast majority of the map area.

* All N A s had councils: Light blue, appearing in small clusters in the Northwest and Central regions.

* Greater than or equal to 50 percent N A s had councils: Pale teal, located in small pockets in the Northeast and Central-South regions.

* Less than 50 percent N A s had councils: Medium blue, found in a few isolated areas in the Central and Southwest regions.

* No N A s had councils: Dark navy blue, concentrated in several distinct patches along the Northern border, the Eastern border, and a cluster in the Southwest.

The distribution shows that while most of the country is categorized as precolonial states, the presence and density of N A councils vary significantly in small, non-contiguous pockets across the Northern and Central territories.

Authoritarian states also existed in southern Uganda. Here, though, institutional reforms placed greater constraints on the precolonial ruler. The Uganda Agreement of 1900 enshrined Buganda’s position at the core of the nascent British protectorate. British administrators concluded this agreement not with the Kabaka (ruler), but instead with the county chiefs, the highest level of territorial organization beneath the Kabaka. The temporary weakness of the incumbent Kabaka—a small child who had taken the throne following the British deposition of his father in 1897—enabled the county chiefs to elevate their prerogatives by empowering the indigenous council, the Lukiiko (Low and Pratt Reference Low and Pratt1960, 73–4). This reversed trends during the nineteenth century whereby successive kabakas had amassed substantial authoritarian powers and gained the right to unilaterally appoint most county chiefs. Consequently, Buganda was an authoritarian state prior to colonization but became a chief and council under Native Administration. “[T]he Agreement quite clearly deprived the Kabaka of his single-handed legislative and judicial functions,” which he had enjoyed prior to the colonial imposition. Instead, “by the Agreement the Lukiiko became none the less quite unmistakably a ‘council’; that is, it became a legislating institution, and not just a gathering of chiefs …” (Low and Pratt Reference Low and Pratt1960, 130–1). Its members consisted primarily of important land chiefs who held their positions independent of the Kabaka (135), which weakened the Kabaka’s powers relative to the precolonial era (145).

Independence and Relevance of Councils

Local councils not only existed within NAs, but (1) exercised independent decision-making powers and (2) exhibited differential patterns of spending behavior. Collectively, this evidence supports our contention that NA councils exerted meaningful constraints on authority and promoted a form of local accountability.

First, the overwhelming majority of councils consisted primarily of traditional elites or popularly selected members. These individuals were less beholden to the colonial administration than individuals who were selected by the NA chief—who were prevalent only in areas of authoritarian precolonial states—or selected by British administrators (which was uncommon anywhere).Footnote ²³

Second, NTs with NA councils spent more on public goods and less on administration, which encompassed salaries for members of the NA. In Panel A of Table 5, the fraction of expenditures on administration is the dependent variable in the first three columns, and the fraction of total expenditures on public goods (education, medical care, and roads) is the dependent variable in the last three. The sequence of columns for each outcome mirrors those in the previous tables, although now Solo-Chief NA is the main explanatory variable. The baseline specifications show that solo-chief NAs correspond with an increase of 18.5% in the fraction of expenditures on administration, from a base level of 21.0%. Conversely, solo-chief NAs correspond with a decrease of 7.0% in the fraction of expenditures on public goods, from a base level of 34.4%.

Table 5.

Public Expenditures and Finance Committees

Table 5. Long description

The table is divided into two sections.

Panel A. Public expenditures.

This panel has six columns. Columns 1 through 3 use the dependent variable Administration percentage. Columns 4 through 6 use the dependent variable Public goods percentage.

- Solo-chief N A: In columns 1, 2, and 3, the coefficients are 0.185, 0.138, and 0.0967 respectively, all positive and statistically significant. In columns 4, 5, and 6, the coefficients are minus 0.0702, minus 0.0482, and minus 0.0593 respectively, all negative and statistically significant.

- Intercept: Values range from minus 0.0457 to 0.409.

- Native Treasuries: 309 for all columns.

- Provinces clusters: 42 for all columns.

- R-squared: Values range from 0.051 to 0.622.

- Colony F E: Checked for columns 2 and 5.

- Covariates: Checked for columns 3 and 6.

Panel B. Finance committees.

This panel has three columns using the dependent variable Finance committee.

- Solo-chief N A: The coefficients are minus 0.595, minus 0.537, and minus 0.325, all negative and statistically significant at the p less than 0.01 level.

- Intercept: Values are 0.648, 1.000, and 1.306.

- Native Treasuries: 399 for all columns.

- Provinces clusters: 55 for all columns.

- R-squared: Values are 0.211, 0.579, and 0.500.

- Colony F E: Checked for column 2.

- Covariates: Checked for column 3.

Note: The sample is our main dataset, a cross section of Native Treasuries in the late 1940s. Models are estimated using OLS with province-clustered robust standard errors. Covariates are the same as in the preceding regression tables. *** $ p<0.01 $ , ** $ p<0.05 $ , * $ p<0.1 $ .

The reforms that implemented councils across most of British Africa also created finance committees, which enabled African counselors to influence the budget process. In Panel B, we demonstrate that NA councils were strongly correlated with the existence of finance committees. As shown in column 1, the baseline prevalence of finance committees was 64.8% in NAs organized as either chief-and-council or council-only. By contrast, solo-chief NAs reduce this frequency to 5.3%.