Forced urbanisation: A cross-national assessment of the effects of intranational political violence on a nation’s largest cities

Abstract

Chinese

This article offers a ‘first step’ toward understanding the consequences of intranational political violence on the growth of a nation’s largest cities. Theory and research from studies on forced migration and internal displacement are used to construct several hypotheses that assess the impact that various forms of intranational political violence have on the growth of major urban areas within the developing world. Hypotheses are tested using a cross-national time-series sample of 85 developing nations from 1974 to 2005. The results provide strong empirical evidence that various forms of intranational political violence are significantly related to population growth amongst a nation’s largest cities. Specifically, attacks on government personnel are associated with an increase in population growth among a nation’s largest cities. Violence targeting civilians are associated with decelerated growth in a nation’s largest city but increased growth in major secondary cities. Finally, increases in the intensity and duration of civil wars are associated with decreases in population among secondary cities but exhibit a curvilinear growth pattern in a nation’s largest city (i.e. an increase then decrease). It is argued that the findings are at least partially explained by the ‘spatial logics’ that arise from a given form of political violence. It is concluded that more attention should be given to studying the consequences of political violence on the urbanisation process for rapidly urbanising nations.

Keywords

economic development forced migration internal displacement large cities political violence population studies urbanisation urbanisation and developing countries

Introduction

A nation’s largest cities, especially among developing nations, are often at the centre of their burgeoning economies and modern political institutions. For this reason, scholars from various disciplines have spent a great deal of time and effort assessing demographic trends associated with urban growth and the distribution of urban population within nations. In a majority of studies, scholars have either assessed the growth of population in a nation’s largest city (e.g. Bertinelli and Strobl, 2007) or the distribution of the urban population within the overall city-system (e.g. Moomaw and Alwosabi, 2004). Most, if not all the research, has been based on development models that assume urbanisation and city growth occur in politically stable environments, and mainly in response to economic changes brought about by global and localised industrialisation.

This study challenges the utility of this untested assumption by demonstrating that political violence, in all of its large-scale manifestations, is an important force that influences the size and distribution of urban population within the developing world. In short, this study provides empirical insight into the effects that internal forms of political violence have on the growth of a nation’s largest city and secondary cities.¹ Specifically, it demonstrates that different forms of political violence have different spatial logics, which when played out in the real world, have a direct and measurable impact on the distribution of a nation’s population within its largest cities. To this end, the basic research question addressed is as follows: what effect does political violence have on the growth of urban population within a nation’s city-system?

Literature review

To construct hypotheses that are appropriate for testing the research question posed above three underlying questions need to be addressed:

What determines if (and when) people flee from political violence?

To where do people flee (non-urban or urban areas)?

Are any types of political violence more common in urban or non-urban areas?

The first two questions can be answered with research on forced migration, but especially internally displaced peoples (IDPs). The third question can be answered with research dedicated to defining and measuring various forms of intranational political violence.

When will people flee?

Most of the literature on forced migration identifies or focuses on political violence as the primary cause of displacement (Schmeidl, 1997). Thus, it is biased toward assessing the underlying causes of refugee flows and offers less insight into factors related to internal displacement. A consequence of this is that most of the research on forced migration looks to nation-states as the primary (or final) destinations for displaced people.² Unfortunately, this minimises (or in some case eliminates) the need for scholars to account for whether displaced persons are migrating to rural or urban areas, or to specific cities. As a result, the empirical findings in the forced migration literature are much less relevant to the substantive concerns of this study. However, the literature contains a theory that is useful for addressing if (and when) people flee from political violence.

In his article on political conflict in Guatemala, Morrison (1993) offers a baseline model of forced migration that centralises the role of political violence. Unlike the purely economic agent-centred theory contained in the Harris-Todaro (1970) model, Morrison’s model suggests that migration decisions in conflict-ridden regions are primarily a function of people’s fear of death. This fear can impose additional migratory costs since the decision to migrate (or not migrate) is linked to the efficacy of ‘producing safety’ (Morrison, 1993: 817). As he states it, ‘some utility maximizing choices of migration may involve reductions in an individual’s expected monetary income’ if migration maximises safety (1993: 819). This possibility is not captured in the Harris-Todaro model, yet it is an important aspect of migration in the developing world, especially for people living in nations with unstable political and economic climates. Thus, Morrison’s model asserts that when political violence is present, efforts to ‘produce safety’ may be done without regard for economic consequences, but actors will seek to maximise both outcomes when and where possible (1993: 820).

To where do people flee?

Research on IDPs provides insights that can be used to answer the second question, where do people flee? Regardless of the nation, the literature suggests that both civil wars and guerrilla warfare push people to large cities, and often to a nation’s largest city. For example, in the edited volume Caught Between Borders, Birkeland and Gomes (2001) look at the cyclical effect of civil war and political instability on rural people in Angola. A large proportion of their assessment is dedicated to exploring the migratory strategies of rural IDPs who must simultaneously adapt to the ongoing political violence (i.e. produce safety) and the new economic realities of urban areas. One of the primary strategies is that IDPs either move to a city, in an attempt to integrate into the urban economy, or they move in closer proximity to a city and engage in rural economic activities on the urban outskirts. In either case, the urban and semi-urban landscapes are perceived to be safer places. This distinct migratory pattern is best conceptualised as forced urbanisation ³ since non-economic push factors related to political violence are the primary driving force behind the rural–urban migration.

Jacobsen et al. (2001) and Duffield’s (2002) research on Sudan also reveals that large cities influence migratory patterns of people who are displaced by political violence. Each study points out that the actions of the Sudanese government and NGOs, both headquartered in Khartoum (a large primate city), played a major role in influencing the urbanward movement of rural people displaced by political violence. In fact, both studies reveal that Sudan’s southern, rural IDPs were more likely to seek shelter in northern, large cities even when doing so resulted in acquiring a minority status and facing overt discrimination (Duffield, 2002: 86–88; Jacobsen et al., 2001: 78–82).

Similarly, Malthotra’s (2007) case-study of Northern India (Kashmir) suggests that the same rationalisations emerged amongst India’s rural minorities who were displaced by political violence. Moreover, like those in the Sudan and Angola, the Indian minorities exhibited a tendency to view Kashmir’s large cities as economic and political safe-places.

Finally, Fagen (2011) also offers evidence that those afflicted by political violence tend to relocate to large cities. However, her research provides additional insight since the findings suggest that large cities are not always perceived to be safe places – particularly if a group is a frequent target of terrorism. Thus, her study provides evidence that urban-based political violence can be a factor that motivates people to move out of cities, or at least discourages migration to a specific city (e.g. the primary city) if urbanised political violence creates the perception that a specific city is less safe than other cities. This insight lends further support to Morrison’s (1993) assertions and leads to the final question that needs to be addressed.

Are there ‘urban’ and ‘non-urban’ forms of political violence?

If one looks at past research and historical events, it is possible to identify distinct ‘spatial logics’ among different forms of political violence, and thus, conclude that a given form of political violence is more likely to occur in an urban or non-urban space. However, making such a connection has the potential to be problematic since different forms of political violence have manifested in rural, semi-urban, or urban areas. Still, a careful review of the literature and assessment of historical events suggests that there are distinctive forms of political violence that tend to exhibit distinctive spatial logics. In short, some forms of political violence are more effective in urban environments and others are more effective in rural environments.⁴ With this in mind, the three most common forms of intranational political violence and their likely targeted publics are discussed below.

Terrorism

Terrorism is a form of political violence in which civilians and known public spaces are purposefully targeted to maximise the symbolic value of the violent act (Crenshaw and Robison, 2009). Terrorism resembles guerrilla insurgency in that it employs a strategy of hit and run violence that uses clandestine tactics, but it does so with indiscriminate targeting. It also involves the actions of smaller groups of individuals who are unmatched by their foes in resources and personnel. However, unlike guerrilla insurgents, the targets of terrorism are much more likely to be civilians who are located within large urban areas (Clutterbuck, 1982; Le Blanc, 2013; Nedal et al., 2015; Savitch et al., 2001; Savitch, 2008).

Indeed, the reoccurring choice to target urban, rather than non-urban spaces, is largely due to the tactics, strategies and goals of terrorist organisations. The goal is to produce political theatre. This entails maximising the visibility and uncertainty of the violent act by using the least amount of effort and resources possible. Thus, acts are carried out by one or a few people, usually in densely populated areas, to maximise shock (Laqueur, 1987). Furthermore, the audiences to whom terrorists direct their messages are largely urban denizens (e.g. civilians, corporations, news agencies, NGOs, tourists, churches, schools, and so forth). Therefore, from a purely strategic and logical standpoint, urban areas provide terrorists with a set of targets that offer the greatest return. Urban areas contain a combination of people and symbolic targets, which when attacked with incivility via violence, receive the most attention and cause the most panic among the general public and governing authorities (Crenshaw, 1981; Crenshaw and Robison, 2009; Hinkkainen and Pikering, 2014; Savitch et al., 2001; Savitch, 2008). Therefore, the prevailing spatial logic of terrorism, as a modern form of political violence, is urban in nature. ⁵

Guerrilla insurgency

Generally speaking, guerrilla insurgency is a form of political violence that often (but not exclusively) exhibits a particular non-urban or semi-urban spatial logic. This is because of the nature of the tactics used, strategies employed, resources available, and goals set by political dissidents. In every case, the dissidents are facing a more powerful oppositional force (i.e. a state) (Fearon and Laitin, 2003). Given this imbalance of power, insurgents must rely on the tactical and economic advantages that less populated areas offer. For instance, they can take advantage of ‘free’ natural resources and defenses (e.g. jungles, mountains, rivers) (Schultz, 1971); they may also settle in smaller villages that provide unprotected (and unintegrated) civilians as recruits and confiscate property as a means to extract rents (e.g. farmland). In fact, from a purely strategic standpoint, guerrilla insurgents will be in a better offensive and defensive position if they establish a foothold in, or near, areas that are largely outside of major urban centres.

Yet many guerrilla insurgencies have arisen within areas that are peripheral to a state’s control, neglected by the state, or difficult to regulate (Williamson, 1965). Indeed, evidence that guerrilla insurgency is more likely to have a ‘peri-urban logic’ to it can be observed around the globe and over time (see de La Calle and Sanchez-Cuenca, 2011). Guatemala, El Salvador, Vietnam, Algeria, Angola, Chechnya, Colombia, and numerous others provide real world examples (Booth, 1980; Crenshaw and Robison, 2009; Fearon and Laitin, 2003; McCormick, 1990; Schultz, 1971; Williamson, 1965). Given the inherent logic and global historical evidence, guerrilla insurgency, as a modern form of political violence, is more likely to manifest in non-urban or semi-urban areas, as is the counter-violence that is used to discourage it.

Civil war

As a form of internal political violence, ‘civil wars’ are labelled as such because of the balance of power between opposing forces. Indeed, those studying civil wars have already observed that guerrilla insurgencies give rise to it when insurgents accumulate enough power, territory, and military strength to combat incumbents openly and consistently (Fearon and Laitin, 2003). This is important because when oppositional armies are more evenly matched rebel strategies shift when they hold larger swathes of a nation’s total land, which includes control of large cities.

Thus, when civil war breaks out rural and urban space can be used by either side for offensive or defensive purposes. This means that both spaces can be attacked, and therefore, must be defended. As a result, battle lines are more dynamic when compared with guerrilla insurgency or terrorism, and there is a more equitable distribution of political violence between town and country – especially regarding scale and magnitude. Therefore, civilians (on both sides of the conflict) will find fewer opportunities to ‘produce safety’ by crossing battle lines that might be associated with the rural/urban divide.

However, if one considers the relationship between the intensity and progression of civil wars, a probable spatial logic can be established. Specifically, short, low-intensity civil wars should be relegated to non-urban areas since most civil wars are born of lingering guerrilla insurgencies that initially gain footholds in smaller villages and unoccupied territories (Fearon and Laitin, 2003). When both sides are evenly matched, and the civil war lasts longer and fighting becomes more intense, violence is more likely to resemble international war (Johnson and Urlacher, 2012; Kalyvas, 2006). When this happens, it is likely that clashes between opposing forces will be intensified and directed toward the opposition’s stronghold – a major city (see van der Maat (2014) for a discussion of the role of elite competition in conflict). Thus, it is likely that the spatial logic tied to civil wars is not linked to the kind of violence employed, but to the changes in the intensity and duration of the conflict, or both.

Measures of interest, hypotheses and controls

Given the above, it is necessary to operationalise different forms of political violence that can be used to develop and test the hypotheses. Ideally, a cross-national longitudinal empirical assessment of the impact that political violence has on growth in the largest city would be derived from data that clearly indicate the type of political violence for each country-year. Fortunately, the Global Terrorism Database (START, 2013), or GTD, contains enough information about the targeting strategies of violent events to operationalise guerrilla insurgency and terrorism, and thus allows for an analysis of their independent effects upon urban change. Similarly, the Major Episodes of Political Violence (MEPV) Marshall (2016) data contain related observations that can be used to construct equivalent measures for civil war. In other words, the measures discussed below identify the targeting outcomes of the political violence (i.e. who is targeted and with what level of intensity), which can be used as a proxy for the ‘rural’ or ‘urban’ effect of the political violence on a nation’s largest cities. Although certainly not perfect, this approach provides consistency when aggregating observations over time and place without sacrificing large amounts of viable information (i.e. using other cross-national time-series data for important controls).

Independent variables of interest

Lethality of civilian and government attacks: Terrorism and guerrilla insurgency

One of the most important aspects of the GTD database is that it contains information on three valuable pieces of information related to small-scale acts of political violence: target type, number of casualties, and frequency of attacks. This, along with the prior assessments of spatial and targeting logics, allows for the construction of two baseline measures that proxy for the degree of lethality for a given form of political violence (i.e. terrorism and guerrilla insurgency).

Using the GTD information on target type, attacks against civilian targets were systematically separated from attacks against government targets as a way of identifying and separating likely ‘urban’ attacks from likely ‘rural’ attacks.⁶ A given attack was coded as ‘civilian’ if the target type of the attack was coded as an attack on; businesses, educational institutions/people, media, NGOs, private citizens, or religious intuitions/people. A given attack was coded as ‘government’ if the target type of the attack was coded as an attack on general government employees/resources, police or military personnel/resources, or diplomats (START, 2013). The same target types were used to code causalities related to the attacks. Once attacks and causalities were identified, the observations were summed for each nation-year and then logged.

It was reasoned that the total lethality of the given form of political violence (number of casualties per attack), as well as its aftermath, needed to be captured in the final proxies since case-studies suggest that people fleeing from political violence do not do so on a whim. In fact, the evidence suggests that people will only flee (or stay away from an area) if the perceived probability of becoming a victim of political violence is near certain (Moore and Shellman, 2006; Morrison, 1993). Therefore, more successful attacks (i.e. higher casualties per attack) should be associated with declines in perceived safety.

Given this, it was assumed that potential IDPs’ perceptions of the probability of victimisation should be a function of the lethality of the attacks, as well as the persistence of the political violence (i.e. campaigns of violence). Therefore, the primary measures of the lethality of civilian attacks (i.e. terrorism) and the lethality of government attacks (i.e. guerrilla insurgency) attempt to capture these dimensions. Both measures are calculated by dividing causalities by attacks for each country-year and then constructing a running five-year average. In the regression tables, these measures are referred to as ‘Civ. Attack Int.’ and ‘Gov. Attack Int.’ and are used as the proxies for the spatial logics associated with each type of intranational political violence. That is, the measure of civilian attack intensity is used as a proxy for “urban” targeting, and the measure of government attack intensity is used as a proxy for “rural” targeting. A five-year running average of these measures is used to test the hypotheses since information from case-studies suggests that IDP migration occurs over longer periods of time (Birkeland and Gomes, 2001; Fagen, 2011; Malthotra, 2007; Moore and Shellman, 2006).

Civil war

To construct equivalent measures of civil war, data on the intensity and duration of civil war were obtained from the MEPV dataset (Marshall, 2016). Measures are at the nation-year level and are rank-ordinal in which the severity of the conflict ranges from 1 (indicating little violence) to 10 (indicating widespread and destructive violence). To make them time-variant, and to capture the effect of the duration of the violence, the scores were multiplied by a running count of consecutive years of campaign violence (one if it was only one year). Once an episode ended (i.e. an annual pause) the count was reset to zero and resumed beginning with one if another bout of violence was observed in future years. The final measure was logged.

Dependent variables

Population growth in the largest city and secondary cities

There are two dependent variables used in the models: the first is the five-year difference of the log of the largest city’s population. The second is the five-year difference of the summed population for the next four largest cities (i.e. major secondary cities). Although growth in primary and secondary cities is certainly related to the same factors, the impact that various forms of political violence have on their growth should differ. That is, a nation may not defend its secondary cities as well as its primary one, and hence people may be more likely to flee to the primary city. Conversely, the largest city’s citizens may be more frequently targeted by terrorists, thus hindering its growth. Therefore, conducting a complementary analysis using an additional dependent variable (i.e. five-year change in secondary city population) provides a more complete tests of the hypotheses stated below.

With the above in mind, secondary cities are defined as the second, third, fourth, and fifth most populated cities within a nation for a given year. Measures of population size for secondary cities were summed and then logged before the five-year difference score was calculated. Data on secondary cities comes from Anthony’s (2014a, 2014b) urban primacy dataset.⁷

Hypotheses

With the above measures in mind, the hypotheses presented below offer the best general test of the relationship between different forms of intranational political violence (i.e. terrorism, guerrilla insurgency and civil war) and their effect on population growth in the largest city and major secondary cities. For all hypotheses and measures, lethality is operationalised as a combination of the annualised duration of political violence and number of casualties per attack as discussed above.

Civilian targets and urban displacement

Terrorism (attacks on civilians) are more effective and ‘efficient’ (i.e. lethal) when they can be carried out in densely settled areas (i.e. large cities) (Crenshaw, 1981). Moreover, terrorism as a political act seeks to maximise symbolic shock by targeting symbolic spaces and people, which means the purposeful targeting of urban landscapes that contain large numbers of people. Taking this into consideration, along with the logic of ‘producing safety’ as discussed by Morrison (1993), the following hypothesis can be established:

H_1a: The more lethal the attacks against civilians, the slower the population growth in the largest city.

However, it makes sense that the ‘fear factor’ associated with attacks against civilians should multiply the more serious the attack is in terms of casualties, and the more frequently they occur. When cities are attacked more often, and with greater success, the perception of the safety of cities will suffer more with each additional, successful attack. Therefore, there is a real possibility that the relationship is curvilinear, with low to moderate levels of political violence against civilians having little or no impact on growth, but with higher levels having a multiplicative effect. Therefore, a second hypothesis needs to be tested.

H_1b: Attacks against civilians with low to moderate levels of lethality will have little to no effect on the rate of population growth in the largest city, while high levels of lethality will have a substantial negative effect on the population growth.

Government targets and rural displacement

For guerrilla insurgencies attacks are more effective and efficient when they are carried out against government resources (military personnel or infrastructure), and in or near areas that maximise access to offensive and defensive resources – i.e. semi or non-urban areas (Fearon and Laitin, 2003). At the same time, counterinsurgencies will begin to target the same areas. Therefore, more intense, and longer instances of guerrilla insurgency should result in increased violence against government targets in rural areas which push residents away.

Since migration will be a function of maximising safety and economic opportunities, a nation’s largest city will have the perceptual advantage in a low-intensity civil war for those who seek safety in an urban area. This is because in developing nations the largest city has more resources (especially in a time of intense internal unrest), is likely to be the ‘hub’ of NGOs and foreign aid, offers the greatest degree of anonymity to minority IDPs, and has more economic opportunities (even if in the informal economy). Therefore, in the wake of high-intensity attacks on government entities by insurgents, residents in rural areas (or smaller villages) should be more likely to relocate to the nation’s largest city to maximise safety.

Additionally, if insurgencies become strong enough and last for long periods of time, incumbents will be forced to centralise personnel and resources to strengthen their resistance (e.g. Libya, Iraq, Algeria, etc.). Therefore, as governments invest a greater proportion of their resources into a single city (likely the largest, capital city), it follows that a disproportionate number of residents will perceive the nation’s largest city as the safest.

H_2a: The more lethal the attacks against government targets, the greater the population growth in the largest city.

However, just like attacks on civilians, it makes sense that the increasing fear associated with hostile insurgencies will intensify as attacks on a government intensifies. Therefore, there is a real possibility that the relationship is curvilinear.

H_2b: Low to moderate levels of attacks against the government will have little to no effect on the rate of population growth in the largest city. Moderate to high levels of attacks against the government will have a substantial positive effect on the largest city’s population growth.

Civil war and the safety of large cities

During civil war, there are several reasons to believe that people will seek out safety in large cities. First, urban areas offer anonymity because of their larger population size and increased density. Even if the likelihood that a large city will be attacked is increased during a civil war, the likelihood that a specific individual will be a victim of a future attack is decreased as population size and density increases. Second, cities offer greater defenses via technological advantages. Not only will political elites defend cities more vigorously with their resources, but the resources needed to mount an effective defence will need to be concentrated in urban areas. Given the above, it follows that large cities should play a pivotal role in rural–urban migration when a nation is experiencing a civil war, with the greatest effect observed in a nation’s largest (often primate) city.

H_3a: The greater the intensity and duration of a civil war, the greater the population growth will be in the largest city.

However, Fearon and Laitin (2003) point out that there is overlap between guerrilla insurgency and civil war. Therefore, the effect of civil war violence on the growth of the largest city might be curvilinear. That is, early in a civil war violence should be in or near non-urban or semi-urban regions, thus pushing people to larger urban areas – especially the largest city. However, as the civil war intensifies and continues for longer periods, rebel forces are likely to gain a stronger foothold in a region, control a major city (perhaps large secondary city), and thus, attack incumbents where they are strongest – the nation’s capital and/or largest city (e.g. Tunisia). When this occurs, perceptions of the largest city’s relative safety advantage is less likely to manifest. Therefore, the following hypothesis will be tested as well:

H_3b: Ephemeral, low to moderate intensity civil wars will increase population growth in the largest city, while high-intensity civil wars will decrease population growth in the largest city.

Other controls

Primacy ratio

Early on Jefferson (1939) established that urban primacy plays a major role in attracting economic, political and cultural capital because of its centrality in national affairs. Primate cities have long histories and are well known, either regionally or globally, thus further attracting investors, innovators, political elites and other migrants. In short, it is important in this study to assess and control the degree of urban primacy because primate status may effectively inflate the largest city’s growth relative to secondary cities. The data used to construct a measure of urban primacy comes from Anthony’s (2014a, 2014b) urban primacy dataset. Urban primacy is calculated by taking the population of the largest city and dividing it by the sum of the population in the next four largest cities. The measure used in the models was logged to correct for skewness.

Paved roads

As central place theory posits (Christaller, 1933; Lösch, 1940), urban growth is a function of the costs of production and distribution. Space imposes costs that improved transportation and communication networks can overcome by linking distant places to the central place (i.e. the largest city). Therefore, the more well-defined the transportation and communication networks become, the greater the potential for growth in the largest city. A measure of kilometres of paved roads (logged), taken from Anthony’s (2014a, 2014b) urban primacy data set, was included in the models.

Real gross domestic product per capita

The relationship between the level of economic development and urban development has been mostly assessed from the perspectives of modernisation theory (e.g. Berry, 1961; Williamson, 1965) as well as central place theory (Christaller, 1933; Lösch, 1940). Most of the research has either looked at the degree of urban primacy or the degree of urban concentration; both are directly related to the size of a nation’s largest city. Several studies have found that increases in the level of economic development have no relationship with concentration or urban primacy (De Cola, 1984; Sheppard, 1982). Others claim that a positive effect exists (Ades and Glaeser, 1995; Mera, 1973; Moomaw and Alwosabi, 2004). Still, others find a negative relationship (Lemelin and Polèse, 1995; Mutlu, 1989; Rosen and Resnick, 1980), and a handful concluded that an inverted U-shaped relationship exists (Davis and Henderson, 2003; El-Shakhs, 1972; Junius, 1999; Wheaton and Shishido, 1981). Given the ambiguity but importance of the control, the log of real GDP per capita (in constant prices, Chain Index) was obtained from Penn World Tables (Feenstra et al., 2013) and included in the models.

Population density

Past research has relied on central place theory to assert that larger, denser populations will require more economic centres, thus discouraging urban primacy, and perhaps growth in the largest city (Alperovich, 1992; Anthony, 2014a; Junius, 1999; Linsky, 1965).⁸ Therefore, a measure of population density was taken from the World Development Indicators (World Bank, 2009), logged to correct for skewness, and included in the models. Initially, a measure of national population size was also included, but it exhibits a high degree of collinearity with the lagged dependent variable (city size) and population density. Since its inclusion did not change the overall results (i.e. it was an unnecessary control), it was excluded from the final analysis.

Percent urban

Given the logic behind demographic transition theory as applied to urbanisation in the developing world (Davis and Golden, 1954), it follows that nations with a higher percent urban should have lower rates of urban growth (i.e. a curvilinear effect). At the same time, central place theory and modernisation theory both suggest that more developed urban networks (i.e. nations with a high level of percent urban) will have a more equitable distribution of urban population across its city system since too much concentration will impose costs on production and distribution (see, e.g. Rosen and Resnick, 1980). Therefore, there should be a curvilinear relationship between percent urban and growth in the largest city as nations with a higher percent urban should approach log-normality within in their urban system.

Fertility rates

The developing world tends to have naturally higher fertility rates than does the developed world, and any attempt to measure urban growth in these contexts should control for total fertility rate as natural increase will likely be a measurable source of urban change throughout a nation. To control for fertility, we include the total fertility rate (average births per woman) from the World Development Indicators (World Bank, 2009).

Government consumption and regime type

Research suggests that developing nations favour large cities (i.e. urban bias) as they try to jump-start development and/or consolidate political power, which often benefits a single city, usually the nation’s capital (see for example; Lipton, 1977; Pugh, 1996). Therefore, to control for urban bias a measure of government consumption, as a percent of GDP, was obtained from the World Development Indicators (World Bank, 2009).

Finally, past research among economists has found that autocracies promote urban concentration via growth in the largest city (Ades and Glaeser, 1995), with federalist democracies promoting decentralisation (Davis and Henderson, 2003). Therefore, it is important to control for the influence that political regimes have on the structure of their urban systems while keeping in mind that there might be a curvilinear effect. Following Davis and Henderson (2003), a control for political regime was constructed using The Polity IV Project’s combined polity score (Marshall et al., 2012). The measure was re-coded to range from 1 to 21, with 21 being the most democratic so that it could be squared.

Methods

The cross-national longitudinal analysis was limited to 85 developing nations (as defined by the United Nations) from 1974 to 2005. The appendix contains a list of the nations in the sample; the sample was constrained because of the availability of data for city size and the theoretical/control variables. To correct for panel heteroscedasticity and contemporaneous correlation panel-corrected standard errors (PCSE) were used (Beck and Katz, 1995, 1996, 2004). OLS with PCSE has been shown to produce less biased standard errors, a problem that can plague analyses which involve the use of large longitudinal datasets (Beck and Katz, 1995, 1996, 2004). However, PCSE does not correct other estimation problems for longitudinal analysis such as serial correlation and panel heterogeneity (for a discussion of all four issues see Kristensen and Wawro, 2003). As a result, other methodological steps had to be taken.

For serial correlation, a Lagrange multiplier test was performed for all models. The results uniformly demonstrate a failure to reject the null hypothesis of no autocorrelation (Beck and Katz, 2011). Thus, no further specifications were made to the models which already included a lagged level version of the dependent variable. As for panel heterogeneity, following the work of Davis and Henderson (2003), Moomaw and Alwosabi (2004), and the advice of Beck and Katz (2004, 2011), country fixed-effects (FE) were included in all the models. Finally, given the potential issue of multicollinearity, the variance inflation factor was assessed in post-estimation. Initially, the log of the total population was included as a control in the models for the size of the economy/society. However, the VIF returned scores well above 20 for the log of the total population and the log of the largest city. Given the interest in the log of the largest city for hypothesis testing, the log of the total population was removed. After removal, the VIF for all remaining variables was below five, thus indicating that multicollinearity was no longer an issue.

Results

The central premise of this research is that various forms of political violence have an impact on the growth of large cities. In short, intense political violence, in terms of duration and casualties, drives people away from the areas where the violence unfolds (urban or rural) and usually results in migration from rural areas to a large city. People seek shelter in large cities because those who are threatened perceive urban areas to be safer places for economic and non-economic reasons.

Given the above, an appropriate test of the hypotheses requires assessing the effects that various forms of political violence have on the growth of a nation’s largest cities, but also its largest (often primate) city. As the literature review and past research suggest, city size does matter for the displaced, and there is an overall ‘logic’ as to which cities will gain or lose displaced population. As the hypotheses assert, which cities gain displaced populations depends upon the nature of the political violence (i.e. who or what is targeted), its duration, and its effectiveness. In short, theory and case-study evidence indicate that a nation’s largest city should experience the greatest gains in population when political violence is linked to guerrilla warfare or civil war since the largest city has the most economic and non-economic advantages all else equal. However, if the political violence is directed towards civilians (most likely terrorism in the largest city), urbanites and potential rural–urban migrants will be more likely to relocate to a large secondary city as a substitute. Therefore, the empirical evidence should reveal consistency and symmetry in the hypothesised effects of political violence when observing the effects on a nation’s largest city and its secondary cities. The results below offer complementary models that provide evidence that the general premise and specific hypotheses in this study are strongly supported.

The first set of models in Table 2 look at the impact that various forms of political violence have on the growth of a nation’s largest city over a five-year period. The second set of models in Table 3 look at the same impact on the combined growth of a nation’s largest secondary cities (i.e. the combined growth of a nation’s second, third, fourth and fifth largest cities). Assessing the impact of political violence on the growth of a nation’s largest city, independent from its effect on secondary cities, provides an opportunity to observe the overall effect of various forms of political violence on urbanisation in the developing world.⁹

Test of hypotheses

The lethality of attacks against civilians

The coefficients for the lethality of civilian attacks provide strong support for H_1b; that is, there is an inverted U-shaped curvilinear relationship between the lethality of civilian attacks and a change in the largest city’s population. In Table 2, models 2 and 7, the lethality measure and its square produce significant coefficients with an increasing then decreasing effect. The threshold at which increases in lethality become associated with a decrease in population growth is ln = 1.35. This translates into a five-year average of about four causalities per attack for the duration of an episode of political violence. In model 7, where controls for all other forms of conflict are included, the estimated threshold is ln = 0.56, which is the equivalent of a five-year average of about two causalities per attack for the duration of an episode of political violence.¹⁰ All figures are clearly within the variable’s observed range (see Table 1), and suggest that only a modest level of civilian lethality is required to alter would-be migrants’ perceptions of the largest city’s safety advantage.

Table 1.

Summary statistics for measures.

Variable	N	Mean	SD	Min	Max
Anthony (2014a)
Largest city population (log)	2024	14.31	1.07	11.44	16.77
Logged five year difference	1760	0.165	0.093	−0.141	0.688
Primacy 1–4 (log)	2024	0.48	0.68	−1.14	2.4
Logged five year difference	1760	0.0028	0.0612	−0.296	0.291
Paved roads km² (log)	2024	9.05	1.6	5.11	14.29
Penn World Tables (v6.2)
RGDP/c (Chain Series, log)	2024	8.09	0.92	5.74	11.15
World Development Indicators (World Bank, 2009 )
Population density (log)	2024	3.62	1.19	0.8	7.04
Total fertility rate (log)	2024	1.48	0.414	0.077	2.16
Government consumption (% of GDP, log)	2024	2.56	0.43	1.09	4.33
World Urbanization Prospects (2007)
Percent urban (START, 2013)	2024	44.16	21.45	4.8	98.28
Civilian attacks, intensity of (5YA)	2024	0.46	0.69	0	3.71
Government attacks, intensity of (5YA)	2024	0.43	0.61	0	2.93
Polity Project IV (Marshall et al., 2012 )
Combined Polity score (1–21)	2024	10.06	6.71	1	21
Civil war, intensity of (5YA)	2024	0.35	0.98	0	4.73

The findings for the effect of civilian attacks on the growth of secondary cities are even more robust. In Table 3, when the lethality of civilian attacks is squared, the squared coefficient indicates a significant curvilinear effect for secondary city growth (p < 0.01 for models 2 and 7). This finding lends strong support to H_1b.¹¹ The lethality of civilian attacks is significantly associated with a decrease then increase in the change of secondary city population (ln = 0.67 in model 2, and ln = 0.69 in model 7). Both models in Table 3 suggest that a five-year average of about two causalities per attack for the duration of an episode of political violence is significantly associated with a decrease then increase in secondary city growth.

The lethality of government attacks

The findings in Table 2 support H_2a; the lethality of government attacks takes on a linear function that indicates a positive effect on the growth of a nation’s largest city. The range of the values for the variable span from 0 to 18.7. Thus, based on the model’s estimates a nation with the maximum value of government attacks (in this sample, Sri Lanka) can expect to have an increase of 0.032 logged units in population in the largest city over a five-year period. This represents a modest effect given that the standard deviation for the variable’s observed range is 0.093. However, one must keep in mind that the impact is likely to be cumulative, which means that the long-term effect of an ongoing conflict has the potential to significantly contribute to higher levels of growth in the largest city, especially if violence is sustained over the course of a decade or longer. Moreover, since many slum and peripheral areas are often excluded from census counts (which most urban data rely on, as is the case here), it makes sense that the push effect of political violence against government targets is likely to be much larger than the models indicate. Therefore, the fact that the results showed a significant effect, even with the inclusion of all the conventional controls used to model urban growth indicates that, this is an important finding.

Table 2.

Five-year difference in the log of the largest city’s population regressed on political violence.

	1	2	3	4	5	6	7
Largest city population (log)	−0.316***	−0.318***	−0.321***	−0.321***	−0.313***	−0.312***	−0.316***
	(0.0275)	(0.0274)	(0.0287)	(0.0287)	(0.0288)	(0.0282)	(0.0267)
RGDP/c (log)	0.121**	0.129***	0.121**	0.122**	0.124**	0.131***	0.132***
	(0.0494)	(0.0495)	(0.0486)	(0.0493)	(0.0494)	(0.0486)	(0.0483)
RGDP/c²(log)	−0.00805***	−0.00846***	−0.00792***	−0.00795***	−0.00809***	−0.00851***	−0.00846***
	(0.00306)	(0.00308)	(0.00302)	(0.00305)	(0.00307)	(0.00302)	(0.00300)
Paved roads (log)	0.268***	0.277***	0.308***	0.311***	0.267***	0.261***	0.299***
	(0.0748)	(0.0771)	(0.0783)	(0.0792)	(0.0730)	(0.0720)	(0.0788)
Paved roads² (log)	−0.0338***	−0.0345***	−0.0375***	−0.0378***	−0.0338***	−0.0335***	−0.0371***
	(0.00763)	(0.00786)	(0.00799)	(0.00808)	(0.00745)	(0.00730)	(0.00796)
Paved roads³ (log)	0.00130***	0.00132***	0.00141***	0.00141***	0.00131***	0.00130***	0.00141***
	(0.000259)	(0.000267)	(0.000269)	(0.000273)	(0.000253)	(0.000247)	(0.000267)
Population density (log)	0.296***	0.294***	0.298***	0.298***	0.296***	0.292***	0.295***
	(0.0346)	(0.0344)	(0.0341)	(0.0342)	(0.0343)	(0.0339)	(0.0343)
Percent urban	0.00766***	0.00757***	0.00769***	0.00765***	0.00744***	0.00787***	0.00766***
	(0.00126)	(0.00123)	(0.00131)	(0.00133)	(0.00132)	(0.00129)	(0.00117)
Percent urban²	−8.37e-05***	−8.31e-05***	−8.26e-05***	−8.24e-05***	−8.30e-05***	−8.53e-05***	−8.27e-05***
	(1.07e-05)	(1.05e-05)	(1.08e-05)	(1.09e-05)	(1.10e-05)	(1.08e-05)	(9.79e-06)
Total fertility rate (log)	0.00806	0.00766	0.0125	0.0123	0.00723	0.00598	0.0139
	(0.0134)	(0.0138)	(0.0138)	(0.0138)	(0.0136)	(0.0138)	(0.0148)
Combined Polity score (1–21)	0.00157***	0.00155***	0.00155***	0.00154***	0.00159***	0.00163***	0.00160***
	(0.000287)	(0.000287)	(0.000287)	(0.000287)	(0.000287)	(0.000294)	(0.000293)
Primacy 1–4 (log)	0.0328*	0.0359**	0.0332*	0.0333*	0.0300	0.0270	0.0312*
	(0.0181)	(0.0182)	(0.0177)	(0.0178)	(0.0184)	(0.0178)	(0.0170)
Government consumption (log)	0.0208***	0.0204***	0.0192***	0.0192***	0.0205***	0.0180***	0.0168***
	(0.00390)	(0.00399)	(0.00398)	(0.00400)	(0.00397)	(0.00402)	(0.00406)
Civilian attacks (5YA)	0.00255	0.0172***					0.00657
	(0.00307)	(0.00618)					(0.00835)
Civilian attacks² (5YA)		−0.00638**					−0.00586*
		(0.00250)					(0.00332)
Government attacks (5YA)			0.0107***	0.0138*			0.00437
			(0.00318)	(0.00712)			(0.00910)
Government attacks² (5YA)				−0.00166			0.00524
				(0.00416)			(0.00521)
Civil war (5YA)					0.00266*	0.0265***	0.0253***
					(0.00146)	(0.00421)	(0.00435)
Civil war² (5YA)						−0.00669***	−0.00679***
						(0.00110)	(0.00106)
Constant	2.655***	2.620***	2.577***	2.567***	2.608***	2.597***	2.499***
	(0.421)	(0.421)	(0.422)	(0.429)	(0.427)	(0.406)	(0.396)

Nation years	1760	1760	1760	1760	1760	1760	1760
R-squared	0.789	0.790	0.790	0.790	0.789	0.792	0.794
Number of nations	85	85	85	85	85	85	85

Notes: Standard errors in parentheses, ***p < 0.01, **p < 0.05, *p < 0.1.

Table 3 provides additional insight into the strength of this relationship, but it also brings up additional questions that cannot be fully answered with these data. The initial models (3 and 4) demonstrate a positive linear effect for secondary cities, the same positive linear effect that was observed for the largest city. Taken together, along with what is known about the spatial logic of government attacks and their association with guerrilla warfare, it is reasonable to conclude that this form of political violence would be a strong push factor for rural people who seek safety in any large city, and thus, government attacks appear to be associated with an overall increase in urbanisation for warring nations. However, when all the forms of political violence are observed simultaneously (model 7), the coefficients indicate a highly significant inverted U-shaped functional form. In other words, low-intensity attacks against the government are associated with increased growth in secondary cities, but if the lethality of the attacks reaches a certain threshold, the rate of growth slows (i.e. there is a ceiling effect). Still, the models support a positive linear effect for growth in the largest city, and a curvilinear effect for growth in secondary cities at high levels of intensity. Therefore, it appears that a reasonable conclusion is that increases in the lethality of attacks on government are more likely to result in disproportionate growth in a nation’s largest city (i.e. the urban primacy effect) although secondary cities will also grow as a result.

Table 3.

Five-year difference in the log of the next four largest city’s population regressed on political violence.

	1	2	3	4	5	6	7
Largest city population (log)	−0.244***	−0.242***	−0.234***	−0.234***	−0.230***	−0.230***	−0.251***
	(0.0200)	(0.0199)	(0.0201)	(0.0202)	(0.0203)	(0.0203)	(0.0206)
RGDP/c (log)	−0.0898*	−0.106**	−0.0923*	−0.0925*	−0.0978*	−0.0992*	−0.115**
	(0.0506)	(0.0498)	(0.0492)	(0.0492)	(0.0500)	(0.0507)	(0.0525)
RGDP/c²(log)	0.00708**	0.00795**	0.00747**	0.00748**	0.00764**	0.00773**	0.00831**
	(0.00321)	(0.00317)	(0.00307)	(0.00307)	(0.00312)	(0.00315)	(0.00331)
Paved roads (log)	−0.0593**	−0.0657***	−0.0741***	−0.0742***	−0.0845***	−0.0838***	−0.0549**
	(0.0237)	(0.0220)	(0.0235)	(0.0235)	(0.0231)	(0.0229)	(0.0216)
Paved roads² (log)	0.00250**	0.00282***	0.00339***	0.00339***	0.00406***	0.00403***	0.00224**
	(0.00116)	(0.00107)	(0.00118)	(0.00118)	(0.00114)	(0.00113)	(0.00107)
Population density (log)	0.181***	0.187***	0.184***	0.184***	0.185***	0.185***	0.190***
	(0.0242)	(0.0233)	(0.0245)	(0.0244)	(0.0254)	(0.0254)	(0.0243)
Percent urban	0.00763***	0.00786***	0.00676***	0.00677***	0.00649***	0.00640***	0.00801***
	(0.000984)	(0.000937)	(0.000984)	(0.000985)	(0.00101)	(0.00102)	(0.000911)
Percent urban²	−6.17e-05***	−6.32e-05***	−5.77e-05***	−5.77e-05***	−5.68e-05***	−5.63e-05***	−6.27e-05***
	(7.14e-06)	(7.19e-06)	(7.27e-06)	(7.27e-06)	(7.34e-06)	(7.42e-06)	(6.85e-06)
Total fertility rate (log)	0.0258	0.0264*	0.0313*	0.0314*	0.0301*	0.0303*	0.0300*
	(0.0167)	(0.0160)	(0.0164)	(0.0164)	(0.0156)	(0.0155)	(0.0154)
Combined Polity score (1–21)	0.000521**	0.000578**	0.000540**	0.000541**	0.000557**	0.000548**	0.000511**
	(0.000226)	(0.000233)	(0.000231)	(0.000227)	(0.000228)	(0.000231)	(0.000239)
Primacy 1–4 (log)	−0.00621	−0.00911	0.00633	0.00635	0.0153	0.0158	−0.0139
	(0.0317)	(0.0304)	(0.0323)	(0.0323)	(0.0315)	(0.0318)	(0.0305)
Government consumption (log)	−0.00625	−0.00554	−0.00454	−0.00455	−0.00186	−0.00133	−0.00327
	(0.00403)	(0.00387)	(0.00426)	(0.00428)	(0.00413)	(0.00421)	(0.00399)
Civilian attacks (5YA)	0.0128***	−0.0181***					−0.0243***
	(0.00271)	(0.00589)					(0.00749)
Civilian attacks² (5YA)		0.0135***					0.0176***
		(0.00268)					(0.00301)
Government attacks (5YA)			0.00741**	0.00664			0.0255***
			(0.00339)	(0.00737)			(0.00814)
Government attacks² (5YA)				0.000409			−0.0120***
				(0.00381)			(0.00344)
Civil war (5YA)					−0.00403**	−0.00906*	−0.0120***
					(0.00173)	(0.00483)	(0.00458)
Civil war² (5YA)						0.00141	0.000954
						(0.00132)	(0.00118)
Constant	3.480***	3.539***	3.393***	3.394***	3.396***	3.399***	3.635***
	(0.281)	(0.274)	(0.286)	(0.285)	(0.295)	(0.298)	(0.287)
Nation years	1760	1760	1760	1760	1760	1760	1760
R-squared	0.768	0.773	0.766	0.766	0.765	0.766	0.777
Number of nations	85	85	85	85	85	85	85

Notes: Standard errors in parentheses, ***p < 0.01, **p < 0.05, *p < 0.1.

The intensity and duration of civil war

Finally, Table 2 lends strong support to H_3b; civil wars appear to have a curvilinear relationship with the growth in the largest city (models 6 and 7). More intense and longer-lasting civil wars produce an inverted U-shaped curve. For growth in the largest city, this suggests that low-intensity civil wars drive people to the largest city, but high-intensity civil wars deter the displaced, or at least enough of them to slow growth as the intensity of the war increases to its highest level (i.e. people ‘freeze’ in place as movement, in general, is likely to incur greater risk). In Table 2, model 6 suggests the threshold for this emerges at ln = 0.99 (and roughly the same for model 7 with the controls). Although a bit more difficult to interpret than the previous coefficients, the average of these findings suggest that when the interaction between the MEPV intensity measure and the duration of the civil war reaches roughly 6 years of ongoing war, civil wars begin to deter migrants away from the largest city. However, this does not mean that they then choose to seek shelter in larger secondary cities as was observed for lethality against civilians or for attacks against government targets. As Table 3 shows, the effect of civil war intensity on secondary city growth is negative and likely linear as the squared coefficients never achieve significance. Model 5 suggest that increases in civil war intensity push people away from secondary cities. Given this, it is reasonable to conclude that this finding is driven by a change in the population in one of the larger secondary cities since rebel forces (now or likely equal strength) are using the city as the centre of operations, and thus being attacked directly makes the city less safe.

Conclusion

The central findings of this study are clear: Internal forms of political violence impact the growth of a nation’s largest cities via forced urbanisation. In the wake of political violence large cities are perceived to be ‘safer’ places for the displaced (often rural in origin), but which city (or cities) is perceived to be the ‘safest’ is determined by the nature of the intranational political violence. However, three major findings concerning the effects of specific forms of intranational political violence on the growth of large cities were established.

First, increases in lethality against civilians (e.g. terrorism) on average deters growth in the largest city, but secondary cities experience gains in population when terrorist attacks are at their most lethal. This provides evidence that rural people who are looking to migrate to a large city alter their choice of destination from the largest city (which is the likely target of terrorism) to a large secondary city with greater perceived safety advantage and sufficient economic opportunities.

Second, the lethality of government attacks contributes to growth in the largest city and other secondary cities (i.e. urbanisation in general). Based on the case-studies cited earlier, and the empirical evidence generated, it is reasonable to conclude that displaced rural migrants (i.e. those closest to the violence) are likely to seek temporary shelter in the nearest large city, which often turns into a permanent relocation that is likely linked to the modern amenities and economic opportunities of large cities. However, the significance of this finding, as it applies to growth in the largest city, is unclear since the models are somewhat inconsistent in Table 2 but strong and consistent in Table 3. Based on what is known and hypothesised about the concept of forced urbanisation developed in this study, it is likely that the transient nature of the migrants (and their probable relocation to peripheral urban areas and slums) means that they are underrepresented in census counts which are used to construct the dependent variables for the analysis. Thus, the true impact of internally displaced people on the growth of a nation’s largest city (via forced urbanisation) is likely to be underestimated, which would explain why the finding is weak in one set of models but strong in the other where the addition of the displaced would be more recognisable to officials (i.e. in mid-sized cities).

Finally, the impact of civil war on city growth is more nuanced because of how territory is likely to be divided between opposing forces. Civil wars appear to increase growth in the largest city, with the greatest growth occurring with low intensity and ephemeral violence. However, at high intensities and longer durations of warring, growth in the largest city slows down considerably, indicating a curvilinear effect. Conversely, in secondary cities increases in the duration and intensity of civil wars is associated with a decrease in growth regardless of the level of intensity or duration. Given the social form of civil war (i.e. two equally matched forces, one vying to retain the status quo), and the probability that the prevailing government usually assumes command of the capital and/or largest city (likely a primate city), with the rebels hold up in a major secondary city, it makes sense that as the incumbent force attacks the rebel stronghold (i.e. a major secondary city), would-be-migrants will be deterred from making it a destination while current residents might be motivated to leave as intensity and duration increase. Similarly, if rebel forces become strong enough to target the incumbent’s city (i.e. the capital and likely largest city), then at high levels of intensity and duration (when such challenges are the most likely to unfold) the same diminishing effect will be observed for the largest city. The strong and highly significant curvilinear effects found in the models support this wider assessment of the effect of civil wars on the largest city (Table 2, models 6 and 7) and secondary cities (Table 3, model 5).

In conclusion, while it may be obvious that people flee from political violence and seek out safe spaces, the major findings uncovered in this research are not so obvious. Indeed, they lend a great deal of insight into the general tendency for large cities to be perceived as safe spaces but also provide insight into the specific patterns of urban growth that have yet to be observed in any cross-national study of this size, that being forced urbanisation.

Perhaps the most important finding is the strength of the overall evidence that clearly demonstrates how and why political violence is an important force in shaping a nation’s cities, especially in terms of who moves where, how many, why, and for what reasons. For more refined assessments to emerge, scholars interested in displaced populations needed to find more accurate ways to observe the origin and destination of the displaced (i.e. urban, semi-urban, rural).

Thus, the above research offers an important first step in better understanding the relationship between intranational political violence and urbanisation within the developing world. However, the research above has limitations that should be acknowledged. Mainly, there are limits to the inferences that can be made about the origin and destination of urbanward migrants given the empirical measures used. Importantly, the GTD measures of terrorism and guerrilla war used in this study did not integrate the spatial location information available for some attacks (i.e. coordinates). While beyond the present paper’s scope, in theory a significant number of GTD events can be spatially located using GIS in order to form a more precise distinction for what constitutes ‘urban’ and ‘rural’ attacks. Moreover, some events are associated with specific cities/towns. However, a larger issue remains with these specifications; even with access to more accurate spatial locational data it is still difficult to acquire equivalent observations across time and place (i.e. cross-national time-series data), because of a lack of consistency with how administrative boundaries are conceptualised and observed.¹²

Additionally, although the regression models estimate changes in urban growth quite well, the measures used cannot isolate rural–urban, urban–rural, or urban–urban migration patterns, and thus important consequences and relationships are left unobserved. Therefore, there is a need for other methods of investigation to be applied to generate clarity on exactly who is moving where and why, and to what extent. To that end, comparative case-studies of war-torn regions, in-depth interviews of displaced rural migrants and transient urbanites, and GIS-based assessments of peripheral urbanisation in and around large cities prior to and after major episodes of internal political violence all represent interdisciplinary methods and approaches that could (and should) be applied to this topic. Yet, even with these limitations, this study provides concrete evidence that forced urbanisation is a topic that deserves serious discussion among urban scholars from various disciplines.

Footnotes

Appendix: List of nations in the sample

Algeria

Angola

Argentina

Armenia

Azerbaijan

Bangladesh

Benin

Bolivia

Brazil

Burkina Faso Cambodia

Cameroon

Central African Republic

Chad

Chile

China

Colombia

Congo, Democratic Republic of

Congo, Republic of

Costa Rica

Cuba

Dominican Republic

Ecuador

Egypt

El Salvador

Eritrea

Ethiopia

Georgia

Ghana

Guatemala

Guinea

Haiti

Honduras

India

Indonesia

Iran

Israel

Ivory Coast

Jordan

Kazakhstan

Kenya

Kuwait

Kyrgyzstan

Liberia

Libya

Madagascar

Malawi

Malaysia

Mali

Mauritania

Mexico

Mongolia

Morocco

Mozambique

Nepal

Nicaragua

Niger

Pakistan

Panama

Paraguay

Peru

Philippines

Saudi Arabia

Senegal

Somalia

South Africa

South Korea

Sri Lanka

Sudan

Syria

Tajikistan

Tanzania

Thailand

Togo

Tunisia

Turkey

Turkmenistan

Uganda

United Arab Emirates

Uruguay

Uzbekistan

Venezuela

Yemen

Zambia

Zimbabwe

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

References

Ades

Glaeser

(1995) Trade and circuses: Explaining urban giants. The Quarterly Journal of Economics 110(1): 195–227.

Alperovich

(1992) Economic development and population concentration. Economic Development and Cultural Change 41(1): 63–74.

Anthony

(2014a) Bringing up the past: Political experience and the distribution of urban populations. Cities 37: 33–46.

Anthony

(2014b) Urbanization and political change in the developing world: A cross-national analysis, 1965–2010. Urban Affairs Review 50(6): 743–780.

Beck

Katz

(1995) What to do (and not to do) with time-series cross-section data. American Political Science Review 89(3): 634–647.

Beck

Katz

(1996) Nuisance vs. substance: Specifying and estimating time-series-cross-section models. Political Analysis 6: 1–36.

Beck

Katz

(2004) Random coefficient models for time-series–cross-section data. Working Paper. California Institute of Technology, Division of the Humanities and Social Sciences.

Beck

Katz

(2011) Modeling dynamics in time-series-cross-section political economy data. American Political Science Review 14(3): 331–352.

Berry

(1961) City size distributions and economic development. Economic Development and Cultural Change 9(4): 573–588.

10.

Bertinelli

Strobl

(2007) Urbanisation, urban concentration and economic development. Urban Studies 44(13): 2499–2510.

11.

Birkeland

Gomes

(2001) Chapter 1: Angola: Deslocados in the Province of Huambo. In: Vincent

Sorensen

(eds) Caught Between Borders: Response Strategies of the Internally Displaced. London: Pluto Press, pp. 17–47.

12.

Booth

(1980) A Guatemalan nightmare: Levels of political violence, 1966–1972. Journal of Interamerican Studies and World Affairs 22(2): 195–225.

13.

Christaller

(1933 [1966]) Die Zentralen Orte in Suddeutscland. Jena: Gustave Fischer. English translation from German original by Baskin

(trans.) Central Places in Southern Germany. Englewood Cliffs, NJ: Prentice-Hall.

14.

Clutterbuck

(1982) Terrorism and urban violence. Proceedings of the Academy of Political Science 34(4): 165–175.

15.

Crenshaw

Robison

(2009) Political violence as an object of study: The need for taxonomic clarity. In: Leict

Jenkins

(eds) The Handbook of Politics. New York: Springer, pp. 235–246.

16.

Crenshaw

(1981) The causes of terrorism. Comparative Politics 13(4): 379–399.

17.

Davis

Henderson

(2003) Evidence on the political economy of the urbanization process. Journal of Urban Economics 53: 98–125.

18.

Davis

Golden

(1954) Urbanization and the development of pre-industrial areas. Economic Development and Cultural Change 3(1): 6–26.

19.

De Cola

(1984) Statistical determinants of the population of a nation’s largest city. Economic Development and Cultural Change 33(1): 71–98.

20.

de la Calle

Sanchez-Cuenca

(2011) What we talk about when we talk about terrorism. Politics & Society 39(3): 451–472.

21.

Deng

(1999) Guiding principles on internal displacement. International Migration Review 33(2): 484–493.

22.

Duffield

(2002) Aid and complicity: The case of war-displaced southerners in northern Sudan. The Journal of Modern African Studies 40(1): 83–104.

23.

El-Shakhs

(1972) The urban crisis in international perspective. American Behavioral Scientist 15(4): 581–590.

24.

Fagen

(2011) Refugees and IDPs after conflict. United States Institute of Peace, Special Report 268: 1–15.

25.

Fearon

Laitin

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

26.

Feenstra

Inklaar

Timmer

(2013) The Next Generation of the Penn World Table. Available at: https://pwt.sas.upenn.edu/ (accessed January 2014).

27.

Harris

Todaro

(1970) Migration, unemployment and development: A two sector analysis. American Economic Review 60(1): 126–142.

28.

Hinkkainen

Pikering

(2014) Land for Rebels: Forestation and Vegetation in Armed Conflicts. KUBEC IR Research Seminar. Unpublished manuscript. Available at: http://www.office.kobe-u.ac.jp/ipiep/materials/flyer20140622_en.pdf.

29.

Jacobsen

Lautze

Osman

AMK

(2001) Chapter 3: The Sudan: The unique challenges of displacement in Khartoum. In: Vincent

Sorensen

(eds) Caught Between Borders: Response Strategies of the Internally Displaced. London: Pluto Press, pp. 79–98.

30.

Jefferson

(1939) The law of the primate city. Geographic Review 29(2): 226–232.

31.

Johnson

Urlacher

(2012) Explaining the duration of counterinsurgency campaigns. Available at: http://ejournal.narotama.ac.id/files/Explaining%20the%20Duration%20of.pdf (accessed January 2014).

32.

Junius

(1999) Primacy and economic development: Bell shaped or parallel growth of cities? Journal of Economic Development 24: 1–22.

33.

Kalyvas

(2006) The Logic of Violence in Civil War. Cambridge Studies in Comparative Politics. Cambridge, UK: Cambridge University Press.

34.

Kristensen

Wawro

(2003) Lagging the dog? The robustness of panel corrected standard errors in the presence of serial correlation and observation specific effects. Paper presented at the Annual Meeting of the Society for Political Methodology. University of Minnesota, Minneapolis, 2003.

35.

Laqueur

(1987) The Age of Terrorism. New York: Little Brown and company.

36.

Le Blanc

(2013) The urban environment and its influences on insurgent campaigns. Terrorism and Political Violence 25: 798–819.

37.

Lemelin

Polèse

(1995) What about the bell-shaped relationship between primacy and development? International Regional Science Review 18: 313–330.

38.

Linsky

(1965) Some generalizations concerning primate cities. Annals of the Association of American Geographers 55(3): 506–513.

39.

Lipton

(1977) Why Poor People Stay Poor: Urban Bias in World Development. Cambridge, MA: Harvard University Press.

40.

Lösch

(1940 [1954]) Die Raeumliche Ordnung der Wirtschaft. English translation from German original by Woglom

Stolper

(trans.) The Economics of Location. New Haven, CT: Yale University Press.

41.

McCormick

(1990) The Shining Path and the Future of Peru. Santa Monica, CA: RAND National Defense Research Institute.

42.

Malhotra

(2007) Internally Displaced People from Kashmir: Some Observations. Indian Anthropologist 37(2): 71–80.

43.

Marshall

(2016) Center for Systemic Peace, Major Episodes of Political Violence, 1946-2015. Available at: http://www.systemicpeace.org/inscrdata.html.

44.

Marshall

Gurr

Jaggers

(2012) Polity IV Project: Polity Annual Time Series, 1800–2012. Available at: http://www.systemicpeace.org/polity/polity4.htm (accessed April 2013).

45.

Mera

(1973) On the urban agglomeration and economic efficiency. Economic Development and Cultural Change 21(2): 309–324.

46.

Moomaw

Alwosabi

(2004) An empirical analysis of competing explanations of urban primacy evidence from Asia and the Americas. The Annals of Regional Science 38(1): 149–171.

47.

Moore

Shellman

(2006) Refugee or internally displaced person? To where should one flee? Comparative Political Studies 39(5): 599–622.

48.

Morrison

(1993) Violence or economics: What drives internal migration in Guatemala? Economic Development and Cultural Change 41(4): 817–831.

49.

Mutlu

(1989) Urban concentration and primacy revisited: An analysis and some policy conclusions. Economic Development and Cultural Change 37(3): 611–613.

50.

National Consortium for the Study of Terrorism and Responses to Terrorism (START) (2013) Global Terrorism Database [Data file]. Available at: http://www.start.umd.edu/gtd (accessed August 2012).

51.

Nedal

Stewart

Weintraub

(2015) Urban concentration and civil war. Unpublished manuscript. Available at: https://static1.squarespace.com/static/522fc0aee4b06bf96fa60e92/t/56b289dd356fb0b4c8540d0f/1454541299205/urban+jan+10+2016_final_notanon.pdf.

52.

Pugh

(1996) Urban bias, the political economy of development and urban policies for developing countries. Urban Studies 33(7): 1045–1060.

53.

Rosen

Resnick

(1980) The size distribution of cities: An examination of the Pareto Law and primacy. Journal of Urban Economics 8(2): 165–186.

54.

Savitch

(2008) Cities in a Time of Terror. New York: M.E. Sharpe.

55.

Savitch

Ardashev

(2001) Does Terror Have an Urban Future? Urban Studies 38(13): 2515–2533.

56.

Schmeidl

(1997) Exploring the causes of forced migration: A pooled time-series analysis, 1971–1990. Social Science Quarterly 78(2): 284–308.

57.

Schultz

(1971) Rural–urban migration in Colombia. The Review of Economics and Statistics 53(2): 157–163.

58.

Sheppard

(1982) City size distributions and spatial economic change. International Regional Science Review 7(2): 127–151.

59.

Sorensen

Vincent

(2001) Caught Between Borders: Response Strategies of the Internally Displaced. London: Pluto Press.

60.

Van der Maat

(2014) A typology of mass political violence. Peace Economics 20(4): 685–695.

61.

Wheaton

Shishido

(1981) Urban concentration, agglomeration economies, and the level of economic development. Economic Development and Cultural Change 30(1): 17–30.

62.

Williamson

(1965) Regional inequality and the process of national development. Economic Development and Cultural Change 21: 309–325.

63.

World Bank (2009) World Development Indicators. CD-ROM. Washington, DC: World Bank.

64.

World Urbanization Prospects (2007) World Urbanization Prospects: The 2007 Revision. New York: World Urbanization Prospects, UN. Available at: http://www.un.org/esa/population/publications/wup2007/2007wup.htm.