The Politics of Anticolonial Resistance: Violence, Nonviolence, and the Erosion of the British Empire

This paper studies conflict in a hierarchical international system, the British Empire. How did the British Empire respond to violent and nonviolent resistance within its colonies? I develop a theory explaining how and why the metropole becomes involved in and grants concessions to its colonies. In contrast to more recent work, I find that violence was more effective at coercing metropolitan concessions to the colonies in the British Empire than nonviolence. This theory is supported with a wide range of data, including yearly measures of anticolonial resistance, every colonial concession made by the British Empire after 1918, daily measures of metropolitan discussions of colonial issues from cabinet archives, and web-scraped casualty data from British death records. My findings show that the effectiveness of resistance is conditional on the political structure that it is embedded in and that hierarchy matters for understanding state responses to resistance.


Introduction
The fall of the European empires in the 20th century transformed the international system. Explanations in the International Relations literature locate this change by referencing Cold War dynamics or the restructuring of the global economy. Less emphasis is placed on the role of anticolonial resistance within the colonies. This is puzzling, since anticolonial resistance was a near constant response to imperial rule, occupied an important role in metropolitan politics, and shaped colonial policy. Both violent and nonviolent forms of anticolonial resistance mobilized populations all around the globe, and the leaders of these movements at the time of independence came to rule their states for decades. Yet there is little political science research on how the European empires responded to resistance, what the relative effects of violent and nonviolent resistance were, and how resistance in imperial structures may differ from the more commonly studied context of the nation-state.
The standard approach in the IR and conflict literature is to examine resistance and conflict between states in an anarchical international system (interstate war), or within a state in a hierarchical domestic structure (intrastate war) (Waltz, 2010). Despite a burgeoning theoretical and empirical literature on international hierarchy (Mattern and Zarakol, 2016;Lake, 2009a), few works have studied how hierarchical relationships between states in the international system are contested, negotiated, and ultimately severed. Examining the decolonization process and how resistance within the colonies affected metropolitan policy can shed light on conflict dynamics in a hierarchical international structure: the European overseas empires. This paper studies the effects of violent and nonviolent anticolonial resistance on metropolitan policy in the British Empire during the 20th century. To do this, I present a model of imperial response. Empires, unlike nation-states, are political structures that are segmented into the metropole and its peripheral units. Peripheral units are directly administered by the local governors, but the metropole is in charge of large-scale changes in policy and other imperial relations.
Metropolitan involvement in the colonies varies spatially and temporally. I theorize that anticolonial resistance can induce concessions from the metropole when the resistance is better able to command metropolitan involvement. Violent resistance is more effective at coercing metropolitan concessions because it threatens the metropole's core interests and garners the attention of the metropole. While disruptive to governance, nonviolent resistance is much less likely to be debated and discussed by metropolitan policymakers. Nonviolence, because it does not threaten the core interests of the metropole, remains the province of the colonial governor.
I test this theory in the context of the largest polity in history, the British Empire (Taagepera, 1978). I collected original data on measures of violent and nonviolent anticolonial resistance in every British colony from 1918 to independence, every concession granted by the metropole to the colony, and daily measures of metropolitan discussion of each colony using British Cabinet archives.
In addition, I provide daily casualty data using web-scraped death records of British Army officials who were killed in these conflicts to validate the measures of anticolonial resistance. This paper presents a number of important findings. The main result is that violence is overwhelmingly more effective at coercing concessions from the metropole than nonviolence. In a wide range of statistical models, the effect of nonviolent resistance on a concession is null. Debates in the House of Commons were much more likely to focus on violent than nonviolent anticolonial activity, and cabinet discussions were more likely to discuss colonies experiencing violent, rather than nonviolent, resistance. These concessions were substantially quite important: metropolitan concessions are strong predictors of the suffrage level in each colony.
This paper contributes to the literature on violent and nonviolent resistance. This study has a large scope while simultaneously providing a granular analysis of the effects of resistance by using months or weeks as the unit of analysis. Doing so enables the study of the short-term effects of resistance, which can help alleviate endogeneity concerns. The finding that violence is more effective at coercing concessions is at odds with results from other works examining similar questions within the context of the nation-state (Stephan and Chenoweth, 2008). I argue that the reason for this discrepancy is that the interests of the European colonial overseas empires in general and the British Empire in particular were threatened more by violence than nonviolence. This argument implies that in order to properly assess the effects and success of any type of resistance campaign, the interests-their values and willingness to concede-of the target of a campaign must be understood.
In addition, I build on recent IR work on hierarchy in the international system by providing a theory of imperial response. A large number of works in the political economy and economics literature have used fine-grained data to examine the long-term effects of colonial investments on current developmental and political outcomes (Huillery, 2009). The general conclusions from these works are that the decisions made by the colonial powers-on where to invest, who to colonize, etc.-have long term repercussions. Yet there remains a lack of theorizing about how the empire makes decisions, how colonial subjects can shape policy, and how the unique metropole-periphery structure of imperial rule matters for understanding the dynamics of empires. By presenting a theory of imperial response, I show that the hierarchical structure of an imperial polity matters for understanding how policy within an empire is shaped, and that this may differ from the domestic context of a nation-state.
These findings are also relevant to the debate about the causes of decolonization. Decolonization was "the most important change in word politics" (Jervis, 1989: 34) during the 20th century, but the analyses of causes of this change have been understudied by IR scholars (for an exception, see (Crawford, 2002)). By showing that concessions-which increased suffrage to the colonies and facilitated self-determination-were a response to violent unrest, this work highlights the role that resistance within the colonies played in increasing the level of colonial autonomy. By studying low-level acts of resistance rather than just the maximalist campaigns of Gandhi or the Mau Mau, I am able to show that anticolonial resistance mattered more than previously thought.

Empires, Violence, and Nonviolence
Empires are a form of international hierarchy (Mattern and Zarakol, 2016;Lake, 2009a). Empires, regardless of how control is acquired, are characterized by a metropole and a periphery (Doyle, 1986;Nexon and Wright, 2007;Wimmer and Min, 2006;Lake, 2009b). This structural division matters for understanding imperial dynamics and metropolitan responses to activity in the periphery. Metropolitan involvement in the periphery varies over time and space and helps explain why some types of resistance are effective at granting concessions. Variation in metropolitan involvement only occurs in the context of an imperial structure and has no clear analogue to the nation-state.
A consequence of the core-periphery structure of an empire is that the metropole can selectively engage and disengage with its peripheral units. This tendency was higher in the modern European overseas empires than in the more traditional land-based empires like the 18th century Russian Empire (Kumar, 2019: 161). Heterogeneous contracting between the metropole and periphery means that there is variation in autonomy, obligations, and importance across colonies. The metropole can place larger demands on some colonies than others. While many historical accounts acknowledge this variation in the context of France's relationship to West Africa and Algeria (Aldrich, 1996), less is known about what drives temporal variation in metropolitan involvement and attention to its colonies. I argue that metropolitan involvement is both an important characteristic of imperial rule and also helps us understand when the metropole changes its policy towards the periphery.
Despite the centrality of the core-periphery structure as a characteristic of imperial rule, few studies have examined how hierarchical links between states in a core-periphery relationship are contested. Moreover, less is known about how formal hierarchical links break and are replaced by informal ones. One of the characteristics of international hierarchy is some form of compromised sovereignty by the subordinate state. How can the subordinate state restore some of its sovereignty from the dominant state? Why and when does the dominant state devolve power to some peripheries but not others? My key variable that explains this outcomes is violent anticolonial resistance, which disrupts the relations between the metropole and the periphery.
This work builds on existing research on the relative effects of violent and nonviolent resistance.
The majority of these works focus more on characteristics of resistance campaigns or the variety of strategies used by those doing the resistance. While government responses are an important outcome variable of these works, very often the interests of the targeted government are left unexplored.
The literature on mass violent and nonviolent campaigns, for example, only examines variation in effectiveness against democracies and autocracies (Stephan and Chenoweth, 2008;Griffiths and Wasser, 2018), as does the literature on the effectiveness of terrorism (Pape, 2006;Stanton, 2013;Fortna, 2015). In the civil war and self-determination literature, explanations for why states grant concessions are located in divisions within a movement (Cunningham, 2011) or their short-term use of violence (Thomas, 2014a). In contrast, this work focuses more on the structure of the targeted government and offers a theory designed to explain government responses.
Recent works have assessed the effectiveness of violence and nonviolence by comparing the suc-cess rate of maximalist campaigns (Stephan and Chenoweth, 2008;Griffiths and Wasser, 2018).
This approach suffers from a number of inferential problems, especially when applied to decolonization campaigns. Applying such an approach to the British Empire would not tell us much because decolonization and independence were overdetermined. Indeed, a striking feature of Griffiths and Wasser's data is that all anticolonial secessionist movements-regardless of the methods they useare coded as successful. Similarly, in Stephan and Chenoweth's dataset, all anticolonial campaigns that are active at the time of independence are coded as successful campaigns while those that petered out before independence is declared are deemed unsuccessful campaigns. Furthermore, there are huge inferential problems arising from the endogeneity between British reluctance to grant a colony independence and the likelihood that a violent campaign arises at all or that a nonviolent campaign turns violent. For some colonies, de facto independence occurred well before de jure independence, which placated more aggressive nationalists. Moreover, anticolonial resistance occurred in widely different contexts. The Mau Mau revolt in Kenya, for example, was largely a civil war among the Kikuyu ethnic group that resulted in the deaths of about two dozen settlers (Bennett, 2013;Anderson, 2005). The British spent many lives and resources to combat the Mau Mau rebellion, but in Ceylon, the British were unwilling to pay those costs. Once there was the specter of violence in Ceylon in 1948, the British withdrew (Thomas, 2014b: 110).

Theory of Imperial Response
This section presents my theory of imperial response. I specify the three key variables in my analysis: anticolonial resistance, concessions, and metropolitan attention.

Anticolonial Resistance
The first variable is anticolonial resistance. I define anticolonial resistance as contentious political actions by colonial subjects that were in any way related to the colonial state. Since my theory is target-centered, and the colonial mindset flattened nationalist and proto-nationalist resistance into criminal unrest and rebellion, I include nationalist and non-nationalist forms of anticolonial resistance in my key independent variable. 1 The variable of anticolonial resistance is also limited to non-institutional forms of contentious politics. Most colonies possessed some form of representative institutions for colonial subjects.
These institutions allowed the native elites to shape policy through conventional means: negotiations, meetings, petitions, etc. Institutional resistance was an important form of colonial politics.
However, native elites were also collaborationist. They benefited, sometimes handsomely, from the colonial regime and facilitated the colonial project. Their interactions with the British were not contentious. They were part of, and not separate from, the colonial state. For these reasons, institutional forms of resistance conducted by the native elite are not included in my variable of anticolonial resistance.
This variable includes common and visible forms of resistance such as riots, protests, strikes, boycotts, and bombings. More organized forms of resistance are included as well, such as an armed insurgency, as happened in places like Kenya, Cyprus, and Palestine. Notably, most discussions of anticolonial resistance focus on these types of mass conflicts and counterinsurgency operations, or with Ghandi or Nkrumah's use of nonviolent civil resistance.
My theory explicitly considers the relative effects of violent and nonviolent forms of resistance.
The main difference between the two variables here is whether the resistance results in any physical destruction or harm to individuals. Deaths (intentional or not) as a result of an act of resistance will almost always mean that the resistance was violent. Riots, even when no one is physically harmed, are also considered violent because they result in the destruction of property and injuries to bodies. Acts of nonviolent resistance include mass protests, organized boycotts, and strikes.
Colonial subjects commonly used strikes to coerce concessions from local industry. In some cases, these strikes were explicitly nationalist. For example, during the war in Cyprus, over 1,000 Cypriots organized a hunger strike to protest British policies (Correspondent, 1957), and in Palestine, the 1936-1939 Arab nationalists successfully orchestrated a general strike for months until it petered out and devolved into a more conventional violent insurgency (Kelly, 2017).

Concessions
The second variable, and my main dependent variable, is a concession from the metropole to a colony. I define a concession as a substantive policy change by the metropole that favored the colony. This definition excludes smaller favorable actions and larger repressive actions. The revocation of a constitution or the declaration of an emergency was a substantive policy change, but did not favor the colony. Other colonial acts did favor colonial subjects, for example the reduction of taxes collected or the distribution of investments to specific regions. These types of changes were often made at the local level by colonial or district governors and did not require metropolitan input. In my theoretical model, concessions are enacted by the metropole in order to provide some benefit to the colonies. Concretely, these benefits usually resulted in the granting of rights, increased representation, or increased suffrage. Within the colonial empires, concessions were not a tactic used only by the metropole. Colonial governors and even district governors used concessions as well. However, these were more often smaller concessions that related to the day-to-day governance of a colony, namely tax relief and colonial investments. Local level officials could accede to the demands of colonial subjects by distributing investments or by selectively choosing not to collect legally mandated taxes. This, however, required no change in colonial policy and was determined by local and colonial governors, not metropolitan politicians.

Metropolitan Involvement
In contrast to nation states, the territorial reach of an empire is much larger relative to the extent of metropolitan territorial control. Areas under control of the metropole are much more in flux as well. Territorial gains can come and go in an empire without jeopardizing the fundamental institutional core of the state itself. Britain could lose India and remain Britain, but Spain losing Catalonia would result in a fundamentally different entity. Furthermore, the federal structure of the nation-state is different from the hub-and-spoke structure of imperial systems. Federal nation-states integrate equal units and lack a core (Cooley, 2005: 27). The sheer number of formal colonies and areas with informal territorial control in an imperial system means that metropolitan involvement to each area is not constant. Quite simply, there are times where the metropole can safely ignore some areas at the expense of others. I argue that metropolitan involvement is the mechanism linking resistance in a colonial territory with any substantive change in metropolitan policy. The hierarchical structure of imperial systems provides local level governors with sufficient powers and resources to run colonies with limited metropolitan intervention and only occasional or regular oversight. Large-scale policy changes become more likely when the metropole becomes involved in colonial affairs simply because the metropole controls the resources and means in order to make these changes. When anticolonial resistance occurs in a colony, policymakers in the metropole may discuss the colony more to figure out the proper way to address the colonial unrest. Unrest without involvement does not lead to metropolitan reforms.
How does understanding metropolitan involvement matter for understanding the relative effects of violence and nonviolence on concessions? The answer is that resistance causes concessions via involvement. This happens in two ways: resistance increases the amount of involvement to a colony from the metropole, and resistance also changes the discourse about the colony with in the metropole. The empirical implication here is that when resistance occurs, the metropole will discuss the colony and will do so in different ways depending on whether the resistance is violent or nonviolent. My theory predicts that violence in a colony will lead to increased involvement to that colony relative to nonviolence, while the impact of nonviolent resistance on metropolitan discourse is lower.

Hypotheses
My theory posts that the structure of imperial rule matters for understanding the differential effects of violent and nonviolent resistance. The British Empire commonly granted concessions to its colonies. This is in contrast to federally structured nation-states where segmented states within the federal structure may seek concessions. In those cases, reputational effects and the need to preserve territorial integrity make concessions that increase autonomy to secessionists risky (Walter, 2009).
The nature of imperial structures means that concessions can more easily be granted to one colony without substantially encouraging other colonies to demand similar concessions. (The British feared that giving concessions to one African country would inspire demands in a nearby one, but most of the anticipated contagion was for geographically proximate colonies, not all of those within the empire.) In response to violence, then, concessions can be granted without jeopardizing the entirety of imperial rule.
I argue that, unlike federally structured nation-states, the distance (in terms of geographic but also in terms of information and control) between the metropole and its colonies meant that policymakers in the metropole were much less frequently concerned with the goings on in each individual colony. Metropolitan attention to colonies was at a premium. In order for subjects to do something that led to a change in policy, they needed to get the attention of the metropole.
Since the amount of attention devoted to each colony was low and the default position in most cases was to delegate authority and control to local-level officials, metropolitan policymakers were quite happy to not discuss a colony and let colonial governors administer the colony and perform other, non-imperial tasks.
Resistance had the potential to disrupt this equilibrium. Resistance could overwhelm the capacity of local colonial state to repress, contain, or defuse the resistance. Resistance was also visible to outside and domestic observers. Unrest in the colonies undermined the reputation of the British as benevolent colonial rulers who were shepherding their subjects into the next stage of development. Resistance in the colonies could also inject colonial issues into metropolitan politics. Since resistance signaled that local colonial officials were unable to maintain order in their colonies, the metropole responded by discussing the situation in the colony. For the British Empire, important discussions of colonial unrest occurred in the House of Commons and the Cabinet. Both institutions routinely conversed with local officials in the Colonial and Foreign Offices. It was in these institutions where the appropriate metropolitan response was discussed and debated.
All else equal, metropolitan governments preferred no resistance in their colonies. Resistance inhibited tax collection, the building of the colonial state, the extension of colonial investments, and perhaps most importantly, entangled empires in colonial wars. At the same time, metropoles would prefer not to devote extra-institutional metropolitan attention to colonies. The structure of imperial rule created local governors and colonial officials whose job it was to oversee and administer the colonies. As in any principal-agent relationship, the principal (the metropole) preferred that the agent (colonial officials) do what the principal delegated the agent to do with as minimal oversight and interference as possible. The default state of the metropole was thus to spend as little attention and hope for as little resistance as possible in their colonies.
If the theorized overall effect of resistance is to increase metropolitan attention, why does the difference between violence and nonviolence resistance matter? I argue that violence is effective against imperial rule for two reasons. Violence was better able to increase metropolitan attention to the colonies. This attention made it possible for the metropole to respond by offering colonial concessions. Violence is more effective at garnering metropolitan attention because violence made it very difficult to maintain colonial rule. Nonviolent resistance simply did not garner the attention of the metropole because strikes, protests, and boycotts could be easily dealt with using traditional colonial methods of concessions made by the local colonial government. Increasing wages, for one, was a tried and true tactic of officials in colonial Africa. It is not that nonviolent resistance was futile or ineffective-indeed, nonviolent resistance by colonial subjects likely had a serious impact on the development of the colonial state and patterns of taxation and investments. The fiscal state of a colony was, of course, important and affected the lives of colonial subjects. However, the metropole controlled the most important political features of the colonial state: its constitution, level of suffrage, political representation, etc. Nonviolence may grant a wage increase but could not bring about a reformed constitution or self-government. I use the above theory to generate testable hypotheses. The fundamental difference between violence and nonviolence in my theory, and in most other definitions, is that violence is destructive, either by harming or killing people, or by destroying infrastructure. Nonviolence, however, makes it difficult for the empire to rule its colonies, engage in resource extraction, and maintain international legitimacy. If the metropole is more concerned with the destructive effects of activity by subjects, then we would expect the metropolitan response to this activity to respond with concessions in an attempt to stave off future unrest. If protests, strikes, and boycotts threaten imperial rule, then the metropole should respond to these actions with concessions. This generates Hypotheses 1 and 2, respectively.
Hypothesis 1 (H1). Violent anticolonial resistance within a colony should lead to concessions by the metropole to the colonial state.
Hypothesis 2 (H2). Nonviolent anticolonial resistance within a colony should lead to concessions by the metropole to the colonial state.
Before the metropole responds to unrest with concessions metropolitan policymakers must dis-cuss the colony in order to formulate a policy. The effectiveness of a type of resistance can in part be assessed by its ability to garner metropolitan involvement. My theory predicts that violence will have a positive effect on metropolitan involvement because violence directly threatens the core interest of the metropole, whereas nonviolent activity can safely be handled by the local colonial government. This theory generates hypotheses 3 and 4.
Hypothesis 3 (H3). Violent anticolonial resistance should increase metropolitan involvement to a colony.

Resistance & Concessions in the British Empire
I test this theory in the context of the British Empire. Not all colonies were equally important to those in London. For example, India was the "Jewel in the Crown" of the British Empire, while much smaller island colonies rarely received any involvement from London. Geostrategic dynamics altered the value of colonial possessions over time. The postwar anxiety over declining sterling reserves raised the importance of some colonies (Krozewski, 2001). British interests in Malaya grew rapidly after the loss of India due to Britain's need to convert sterling into dollars and rising great power rivalry in the region (Stockwell, 1995: liv). Maintaining rule over Malaya became essential during the sterling crises since the large volume of rubber exports from Malaya enabled the easy exchange of dollars into sterling (Hinds, 1999: 107). The Cold War also changed the value of colonies. Foreign Secretary Ernest Bevin argued in the immediate postwar period that Britain must maintain control of all their Middle Eastern territories in order to prevent Soviet encroachment (Heinlein, 2013: 15-16).
Some colonies were viewed as strategically essential for the defense of the empire. The British fought to maintain control of Cyprus largely because the loss of India and Palestine reduced their strategic foothold in the Middle East. Withdrawal from Cyprus would result in immediate unification with Greece and the decline in British influence in the region (Heinlein, 2013: 57). Similarly, Palestine's importance was primarily strategic since it provided easy access to British military bases in Egypt and India (Fieldhouse et al., 2006: 196,218). Aden was viewed by The Committee of Imperial Defense as a colony of "general strategic importance to the empire" because of its wireless telegraph and fueling station (Hankey, 1996: 47).
Throughout the 20th century the British Empire routinely granted concessions to the colonies.
The main reason these concessions occurred was due to unrest within the colonies (Smith, 1978).
Concessions were either a response to past unrest or an attempt to stave off future unrest (Rathbone, 1992: xxxvii). In the era of anticolonial nationalism, colonial officials engaged in a policy of "making reasonable concessions [to the colonies] without conceding any points which are genuinely essential" (Heinlein, 2013: 24). The goal was the gradual advance to self-government and the empowerment of native elites that ensured favorable relations with Britain in the post-independence period. For example, in 1946 the Governor of Burma quickly conceded demands to the Anti-Fascist PFL in order to defuse the situation after a series of strikes threatened to paralyze the colony (Heinlein, 2013: 43). During the late 1940s, the strategy of granting concessions to quell unrest was also used in Malaya alongside conventional counterinsurgency tactics (Heinlein, 2013: 52).
London granted concessions to the colonies mostly by reforming the constitution or expanding the legislative council. Constitutional reforms often expanded suffrage, increased internal selfgovernment, and established institutions with native control. Expanding the legislative council (the primary legislative body within a colony) was an easy concession to make that increased local representation without requiring the process of a constitutional amendment. In the vast majority of cases, the seats added to the legislative council were seats earmarked to be held by native, and not European, elites. This helped dilute the power of the European seats. In theory and in practice, expanding the legislative council increased the representation of the non-European population.
While the Colonial Office was mostly in charge of implementing concessions, they did so under metropolitan pressure. For example, Howard Johnson argues that the Colonial Office focused their involvement on the implementation of labor legislation reforms in the wake of the West Indian disturbances of 1937-1938 due to criticism from the Labour Party in Parliament (Brown and Louis, 1999: 608). Lord Moyne, then the Secretary of State for the Colonies, in 1941 proposed a constitutional reform for Jamaica and an increase in suffrage to the Jamaican Governor (Wallace, 1977: 56). For the reform of the Gold Coast constitution in 1954, the Colonial Office drafted recommendations which were then formally accepted by those in Whitehall (Rathbone, 1992: lx).
In 1950 the Secretary of State for the Colonies appointed a board to review the constitutions of British colonies in the Caribbean (Wallace, 1977: 75). Ultimately the decision to grant independence was a metropolitan one. Before Ceylon became fully self-governing, the House of Commons passed the Ceylon Independence Bill in November of 1947 (Heinlein, 2013: 50).
Indeed, selective metropolitan involvement was baked into British rule. Colonial governors ruled over district administrators, and London was quite happy to let the agent in the principle-agent relationship rule the colony without unnecessary interference. In some sense, metropolitan involvement to a colony was a failure of local colonial governing. There were regularly scheduled avenues for metropolitan oversight of the colonies. Territories administered by the Colonial Office were subject to Question Time in Parliament. The production of annual reports-collected for posterity in the appropriate nicknamed 'Blue Books'-occupied substantial resources of colonial officials and the numbers contained in these reports captured enormous detail about colonial happenings. For example, if the metropole cared to know about how many pairs of socks Iraq exported to Palestine in 1937-including whether the socks were made out of plain cotton yarn mixed with shoddy of wool and cotton or plain cotton yard mixed with artificial silk-they could consult the Blue Books.
The relevant point here is that metropolitan involvement to each one of of its colonies by the cabinet and parliament was outside of regularly scheduled channels.

Research Design
To test my theory, I run a number of statistical models to analyze the relationship between variables.
Since concessions were relatively infrequent events that were not immediate reactions to anticolonial resistance, I use colony-years as the unit of analysis when the dependent variable is whether a colony receives a concession from the metropole. However, metropolitan attention was more frequent and also responded quicker to events in the colonies. For this reason, and to limit the threat of unobserved heterogeneity that may bias my inferences, I use a colony-month as the unit of analysis when estimating the effect of unrest on metropolitan attention. Since the choice to use violence or nonviolence by imperial subjects was a strategic one, it is unlikely that there exists a scenario where there is as-if random variation in violence and nonviolence, and it is even more unlikely that such a design examining this hypothetical scenario would have any meaningful external validity. Using a granular unit of analysis such as the colony-month (while not equivalent to complete randomization of the independent variables) aids inference by ensuring that similar units of analysis are being compared. Furthermore, the private nature of metropolitan discussion as measured by cabinet archives helps limit reverse causality as one potential threat to inference.
While endogeneity is present in nearly all observational studies-and especially those involving the dissolution of centuries-old international phenomena involving dozens of countries around the globe-this approach helps increase the likelihood that the effects uncovered in the statistical models reflect a causal relationship.
Temporally, my study is limited to the so-called 'Second' British Empire, which began after World War I (Parsons, 2014). I chose this cutoff for three reasons. One is that there is better quantitative and qualitative data available after World War I. The other is that, after the fall of the Ottomans, the British acquired some of its former possessions, including Palestine. These territories were the last important colonies acquired by the British Empire, which simplifies the analysis by not having units enter into the sample. Colonies are removed from the sample after they are formally granted independence. The third, and perhaps most important reason for this temporal cutoff, is that this time period is commonly believe to be one where the British Empire started declining (Hyam, 2007). This is important for understanding the scope conditions of my theory. Concessions must be a reasonable policy option; in the era of imperial expansion, the usual response to unrest was the consolidation of imperial rule, not retrenchment (Gopal, 2019). I include territories formally administered by the Colonial Office and the India Office, and exclude the Dominions, since there was almost no variation in either the independent or dependent variable. 2 2 The obvious exception to a Dominion that experienced resistance is Ireland. Given Ireland's proximity and geographic contiguity with the United Kingdom, it is excluded from the sample. The colonies in the study are: Aden, Barbados, Bechuanaland, British Guiana, Burma, Ceylon, Cyprus, Gambia, Gold Coast, India, Jamaica, Kenya, Malaya, Malta, Nigeria, Northern Rhodesia, Nyasaland, Palestine, Sudan, Southern Rhodesia, Tanganyika, Trinidad, and Uganda.
Instead of relying on hand-coded measures of violence and nonviolence, I perform a textual search of the entire Times of London newspaper corpus. This results in a more granular independence variable that also includes smaller-scale acts of resistance. I construct this data by executing a query on the Gale Historical Newspapers Database. First, I select all news articles where the colony name is in the subject or keyword of the article in the Times. Then, I filter the articles to ones that include words from a selected list that refer to violent and nonviolent activity. 3 I then sum the number of articles containing each word for every year. This results in two colony-year variables measuring violent and nonviolent activity in the colonies. The data are presented in Figure 1.  1920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 1970 (Thurston, 1998). For each document, I note what colonies were discussed and the date of the discussion. 4 Since the data are right-skewed, I take the log of the total discussions. These data are presented in Figure 2.  1920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 19701920 1930 1940 1950 1960 1970

Model, Estimation and Results
To estimate the effects of my independent variables on my dependent variables, I begin by estimating a number of models with the following baseline form: Where X is a matrix of control variables and ζ is a corresponding vector of coefficients. The main parameters of interest are β 1 and β 2 , which represent the marginal effect of an increase in violence and nonviolence in the previous year on a concession, respectively. Violence and nonviolence are lagged one year because concessions required metropolitan deliberation and the formation of policy, and were rarely granted immediately after resistance occurred. Standard errors are clustered at the colony.
I begin by estimating four models: a logit model, a normal linear model with colony fixed effects where the dependent variable is a binary indicator for whether any concession occurred, a normal linear model with an ordinal dependent variable measuring the size of a concession, and an ordinal logit model with the same dependent variable. Later, I estimate a Bayesian hierarchical model with observations grouped by the colony to account for dependence between observations within a colony and to explore heterogeneous effects. Whenever the dependent variable is a concession, cubic time splines are included to control for temporal dependence (Carter and Signorino, 2010).
The unit of analysis for all models is the colony-year. Additional specifications are provided in the appendix.
These results are presented in Table 1. Across models, specifications, and measures of the dependent variable, violence is a statistically significant positive predictor of a concession, while nonviolence is not. To understand the size of the effect of violence on a concession, Figure 3 presents the predicted probabilities for Model 1 in Table 1. My specified theoretical mechanism linking resistance to concessions is metropolitan involvement. The theorized causal process is that resistance in the colonies leads the metropole to discuss events in those colonies, and when the metropole is more likely to discuss colonial issues, they are more likely to respond to this issues with concessions. This also enables the analysis of data that is more fine-grained than the colony-year. Table 2 shows the results of models estimating the effect of violence and nonviolence on the amount of discussion of that colony by the cabinet. Both the independent and dependent variables are logged to account for skewness. The unit of analysis is the colony-month. The results show that violence in the previous month predicts the level of cabinet discussion in the subsequent month, while no such effect is present for nonviolence. This pattern holds when the variable measuring levels of violence and nonviolence are lagged by two months. Year fixed effects ensure that the comparisons are made within the same year.

Cabinet Discussion
(1) Note: * p<0.1; * * p<0.05; * * * p<0.01 Table 2: This table shows the results of models where the independent variables are measures of violent and nonviolent activity within a colony and the dependent variable is the level of discussion in the British Cabinet about the colony. The unit of analysis is the colony-month. Measures of activity are lagged by one (t-1) and two (t-2) months. Both models include colony and year fixed effects with standard errors clustered at the colony.
To provide evidence for the link between metropolitan involvement to a concession, I run models where the main independent variable is the level of cabinet discussion about a colony and the dependent variable is whether there is a concession in the following year. Table 3 presents the results. Models 1 and 2 use the binary measure of concessions as the dependent variable, while models 3 and 4 use the ordered measure. The results show that increased discussion by the cabinet leads to a higher probability of concessions.
Additional results are presented in the appendix. In Section A3, I explore how the effect of resistance changes over time. Those results show that the effect of violent resistance increases steadily over time. To ensure the robustness of the results, I also estimate models using alternate measures of violent and nonviolent resistance using the House of Commons debates, and find similar effects of resistance on concessions. Those results are presented in Section A1.

Dependent variable:
Concession Note: * p<0.1; * * p<0.05; * * * p<0.01 Table 3: This table presents the results of linear models where the dependent variable captures a concession and the independent variable is the level of cabinet discussion of a colony in the British metropole. The goal here is to assess how increasing metropolitan involvement in colonial politics leads to a increased probability of a concession. Standard errors are clustered at the colony.
Britain's relationship to its colonies varied. To explore the heterogeneity in the effect of violence on a concession, I estimate a Bayesian hierarchical logit model where the independent variables are allowed to vary by colony. This also accounts for within-colony dependence between observations, rather than treat each observation independently. The main parameters of interest is the general and colony-specific effects of violence on a concession. These parameters are jointly estimated, with the population parameter being an mixture of all the colony-specific parameters. The results of this model are presented in Figure 4.  Table 1.
These results show that there was important heterogeneity in the effects of violence on concessions, especially for colonies that experienced a sustained violent conflict, like Cyprus and Palestine.
These findings are consistent with the historiographical literature detailing Britain's hasty abandonment of Palestine in the face of a violent insurgency (Hoffman, 2016) and Britain's unwillingness to grant concessions in the face of violence in Cyprus due to the island's perceived strategic importance (Rappas, 2014).

Measurement Error Models
The data used to construct the measures of anticolonial resistance rely on newspaper reporting.
This creates a number of biases that may affect the inferences made from the statistical models.
This is a common problem. More contemporary event data using newspaper reports often suffers from undercounting bias (Hendrix and Salehyan, 2015). In other words, there are events that occur that do not get reported on, and thus never enter into the datasets used by researchers. Not all events that do not get reported are the same. A mass demonstration with tens of thousands of participants will likely get some coverage, but more routine and smaller events will get ignored.
One of the advantages of using newspaper data is that the biases are generally well known (Weidmann, 2015). This enables the use of statistical models that account for these different biases and to see how the main effect behaves as these biases change. Standard regression models assume that there is no measurement error in the independent variable. This is almost always a naive assumption. In this section, I present the results of Bayesian measurement error models that relax this assumption in ways that account for the likely biases of newspaper reporting about the British Empire.
Bayesian measurement error models replace variables where no data exists with probability distributions. The most basic measurement error model replaces the mismeasured independent variable with a normally distributed variable centered around the true value with some added noise. Equation 2 presents this formally: This simple model can be extended to account for more complex bias patterns. For the current application, research on reporting bias sheds light on the type of bias present in the models. The most probable source of bias is differential reporting of violent and nonviolent resistance. This bias occurs when nonviolent resistance is more likely to be underreport and ignored by the Times of London than violent resistance.
I create a Bayesian measurement error model that accounts for underreporting and estimate it using the Stan language (Stan Development Team, 2018). The code for the model is included in Section A6. I first examine how the general tendency to underreport nonviolence more than violence affects affects parameter estimates in the baseline logit model. 5 To do this, I introduce an additional parameter into the measurement error model in Equation 2, γ, which represents the degree of undercounting the true distribution of resistance.
When γ = 1, there is no underreporting, and when γ = 0, underreporting is so severe that no events get recorded at all. When γ = 0.5, half the true events are reported and entered into the dataset. The parameter σ 2 , is the standard deviation of the measurement model. This can more intuitively be interpreted as "random noise". I set this at 0.125. I then vary the value of γ to see how the parameter estimates change for different levels of undercounting.
The estimated effect of violent and nonviolent resistance at different values of γ are presented in Figure 5. The x-axis is the , or γ. When γ is one, there is no undercounting. An interesting exercise is to examine the relative parameter values across different values of γ for each independent variable. For example, the effect of violent resistance on a concession is about 1.125 when γ = 0.50, which is still higher than the effect of nonviolent resistance when there is no undercounting at all (γ = 1.0). This provides evidence that different rates of undercounting the different independent variables are not driving the main results.  Figure 5: The plot shows how the coefficient for violent and nonviolent resistance changes as a function of the undercounting rate parameter γ. The shaded area indicates the 95% posterior density interval. When γ is 1, there is no undercounting and when γ is 0 no acts of resistance get reported at all. The effect of both violence and nonviolence on a concession attenuates as undercounting increases. Violence has a positive effect on a concession even if half of the violent events that occur enter into the dataset.
Bias driven by newspaper reporting likely varies across colonies and time. In times when there is more interest in the events of a colony, measurement error should decrease. To account for this, I create a colony-year measurement of newspaper coverage using the Times of London database. This is created by performing a search of all newspaper articles with the colony name listed as a keyword and then extracting the number of news articles with that keyword per year. I then scale this measure to make it range from 0 to 1. This data is then multiplied by the discount rate, γ.
In effect this means that γ varies by the colony-year, and undercounting is higher when there is lower newspaper reporting. I then estimate this model with varying values for σ 2 to see how the parameter values change when γ is driven by colony-year variation in newspaper reporting and to see how increased random measurement error affects the estimate of the effect of violence on a concession.
The results are presented in Figure 6. As σ 2 increases, the uncertainty bands increase as well along with the effect size. This is as expected, since measurement error induces attenuation bias.
The main takeaway is that even if random measurement error is quite high, and even if the tendency to undercount violence varies from year-to-year in each colony as newspaper coverage increases, the effect of violence on a concession remains positive and statistically distinguishable from zero.

Verification with Casualty Data
To further support my claim that the newspaper data accurately reflects the level of violence and nonviolence in a colony, I validate these measures with administrative data on British Army deaths.
I webscraped the Gro Army Death Indices from https://www.findmypast.co.uk/. The death indices include the year, name, and location of death for every individual British Army member who died from 1881 to 1955. I aggregated each death to the year level to create a dataset on the number of casualties in each year for each colony. The data are presented in Figure ?? in the Appendix.
The casualty data is not affected by reporting bias at all since it is collected from administrative records. This enables us to validate the measures of violence and nonviolence driven by newspaper reporting. If the casualty data are correlated with the measures of violence, and not correlated with the measures of nonviolence, then this should increase confidence in the data and resulting model estimates, with a caveat: the data do not measure the same underlying concept. Violent anticolonial resistance includes acts that result in property damage and deaths of settlers, British colonial officials who are not members of the military, and deaths of other native colonial officials.
The British casualty data includes all army servicemen who die while serving abroad-including those who die of disease or accidentally (automobile accidents were quite common, especially in Palestine). The relevant point here is that if both data sources are completely free of error, we should not expect a perfect correlation between the two.
Regressing violence and nonviolence on the log of the number of casualties shows that violence and casualties rates are positively correlated, while nonviolence and casualty rates are not. The result of these models are presented in Table 4.  on the probability of a metropolitan concession to a colony. The mechanism linking resistance to concessions is metropolitan discussion and involvement, which reacted more to violent resistance than nonviolent resistance.
My theoretical framework attempts to make sense of how the British Empire actually functioned, and how activity in the colonies led to metropolitan involvement. Understanding metropolitan involvement is important because the metropole was in charge of concessions that reformed the constitution or increased native representation. Smaller-scale concessions, such as wage increases or tax relief, remained the province of the colonial governor. Overseas empires have a common structure where local officials manage the day-to-day operations of the colonies and metropolitan involvement in local governance varies spatially and temporally. This matters for understanding how and when resistance by colonial subjects shapes policy.
What are the scope conditions of this theory? To answer this question it is important to decouple the dynamics of imperial expansion and contraction (Abernethy, 2000). Briefly, this theory only applies to an era of imperial decline. That is, when imperial retrenchment and withdrawal are realistic policy options and the goal of metropolitan policymakers is to reduce their colonial commitments. When empires are seeking to expand, anticolonial resistance can be used a pretext to more aggressive repression and the formalization of control over colonial subjects. This pattern occurred in the British Empire during the 19th century after the Morant Bay uprising in Jamaica and the Indian Mutiny. In both cases, these rebellions resulted in significant metropolitan discussion (Gopal, 2019). Metropolitan policy, however, did not result in concessions, but resulted in making Jamaica a Crown Colony and assuming control from the British East India company.
While the focus has been on the British Empire, similar dynamics apply to other imperial projects. In the French Empire, nonviolence was often futile. When the Algerians practiced nonviolence, the French responded with repression (Clayton, 2014). Violence both threatened the core interests of the French in Algeria and catapulted the FLN onto the world stage (Connelly, 2002).
France's interest in sub-Saharan Africa was much lower and they were more willing to grant autonomy to these colonies than in North Africa (Chafer, 2002). Not all empires shared a clear division between the metropole and periphery as there was in the modern European overseas empires. The land-based Russian Empire, for example, lacked any clear metropole and the various nationalities in the periphery made up the ruling class (Kumar, 2019).
This study has a number of limitations that can be addressed in future work. The data on violent and nonviolent resistance is culled from British newspapers. One advantage of this approach is that it ensures that the metropole is aware, or at least could be aware of unrest within the colonies. In addition, data on unrest from colonial archives would suffer from other reporting biases since many reports containing information on resistance were compiled by subordinates and sent up the chain of command. Too much reported unrest suggests a poor performing colonial official. Of course, not every act of resistance was covered in the press. Many of the foreign correspondents were located in the colony's capital or other large cities. Presumably, acts of resistance in the hinterlands were less likely to be reported by these correspondents, thus making it less likely for these acts to be discussed by the metropole and to have a meaningful effect on policy. A full model of what types of unrest were reported by the press and by colonial officials could significantly advance our understanding of when resistance is effective and the inner workings of colonial empires.
The finding that violence was more effective than nonviolence at coercing concessions raises important implications for understanding the fall of the European empires during the 20th century.
Explanations abound for the global transition from empire to the nation (Emerson, 1962). Much of this work has focused on variation in the timing of independence for colonies within and across empires (Spruyt, 2005). This work shows that the transition from colony to nation was not a rupture in hierarchical relations between states, rather it was a process where autonomy was slowly granted in response to violence. For many British colonies, independence was a formality, although this applies less to colonies engaged in a sustained campaign against colonial rule. The results here suggest that decolonization spanned the 20th century, accelerated after the end of World War II, and that violent anticolonial resistance played an important role in the decolonization process. The effectiveness of violence was not driven by the more commonly studied large-scale colonial conflicts, rather, concessions were also a response to low-level riots and unrest.
What can this study tell us about contemporary hierarchical relations between states? With few exceptions, however, formal imperial ties are nonexistent in contemporary international politics.
Nevertheless there are similarities between a metropole's selective involvement in its formal colonies and great power intervention in its informal client states after the stability of local rule in the client state becomes threatened. In addition, this study shows that Britain successfully managed its imperial decline and avoided large-scale costly conflicts when it made piecemeal concessions to colonial subjects, which has important implications for the literature on great power retrenchment (MacDonald and Parent, 2018).