Economic Backwardness and Social Tension�

Christa N. Brunnschweilery

University of East Anglia, UKPäivi Lujalaz

Norwegian University of Science and Technology

This version: March 2016

Abstract

We propose that relative economic backwardness contributes to the build-upof social tension and the incidence of both nonviolent and violent forms of oppo-sition to the current political regime. We take inspiration from Gerschenkron�s(1962) essay on economic backwardness and more recent �ndings on internationalcomparisons and status-seeking to develop a testable hypothesis. We use informa-tion on a large number of countries and years from a new dataset on episodes oforganized mass movements, and data on armed civil con�ict, to �nd that greatereconomic backwardness is indeed consistently linked to a higher probability of see-ing both violent and nonviolent forms of civil unrest. IV estimations using threenew instruments, including distance to either London or Washington, D.C., andmailing speeds and telegram charges around 1900, suggest that this relationshipis causal.

Keywords economic backwardness, economic development, con�ict, social tension, IVestimation

JEL codes O10, C23, F50�The authors thank Indra de Soysa, Bjarne Strøm, Krishna Chaitanya Vadlamannati, Simone Va-

lente, and seminar and conference participants at NTNU and UEA for their helpful comments.yChrista Brunnschweiler, School of Economics, University of East Anglia, Norwich NR4 7TJ, United

Kingdom. E-mail: c.brunnschweiler@uea.ac.uk.zPäivi Lujala, Department of Geography, Norwegian University of Science and Technology (NTNU),

7491 Trondheim, Norway. E-mail: paivi.lujala@svt.ntnu.no.

1

"[...] great delays in industrialization tend to allow time for social tensionsto develop and to assume sinister proportions."

(Gerschenkron 1962, p. 28)

1 Introduction

Human history is marked by upheaval and violence, and though violence in moderntimes has thankfully been on the decrease (Pinker 2011), we are still confronted withdescriptions of mass demonstrations, political violence and civil con�ict in the dailynews. Why is it that some countries have witnessed increasing political stability, eco-nomic prosperity, and no large-scale civil violence for half a century or longer, whileothers seem caught in a vicious circle of poverty, social unrest, political upheaval, andeven civil war?In this article, we add to the understanding of the origins of social tension and orga-

nized political violence. We develop a hypothesis on how economic backwardness a¤ectssocial unrest, which is inspired by the classic contribution by Alexander Gerschenkron(1962),1 and test its implications empirically. We argue that economic backwardnesscan bear serious consequences for a country�s peace and stability.Gerschenkron studied the history of industrialization in Europe up to the mid-20th

century and pointed out that Russia�s "delayed industrial revolution" was surely toblame for the violent revolution of 1917 and the subsequent establishment of the dicta-torial Soviet government (Gerschenkron 1962: 28). Had Russian serfdom been abolishedearlier than it was, he hypothesized, the discontent among the peasantry that was thedriving force behind the Russian Revolution would not have built up as it did, andeconomic development would have come about more gradually. In sum, "[i]f the Sovietexperience teaches anything, it is that it demonstrates ad oculos the formidable dan-gers inherent in our time in the existence of economic backwardness" (ibid., 29). Helater generalized this observation to state that delayed industrialization would lead to"mounting tension between the prevailing economic conditions and the promise o¤eredby rapid industrial developments" (ibid., 362).Gerschenkron�s work was published at a time when income gaps and economic back-

wardness between countries around the world were on the rise.2 The end of the colonialera across Africa and many parts of Asia brought a large number of newly independent� and very poor �nations onto the global stage. Many of these countries have stillnot seen economic take-o¤ and lag ever-further behind the most highly developed na-tions. We contend that Gerschenkron�s insights into how economic backwardness can

1We refer to the collection of essays titled "Economic backwardness in historical perspective", �rstpublished in 1962, which includes the title essay (�rst published in 1951) and other related ones.

2Bairoch (1981) shows that before the industrial revolution, income disparities between most coun-tries in the world were very limited, with the ratio between richest and poorest country being in therange of 1.0 to 1.6.

2

contribute to the emergence of social tension and large-scale violence are still usefultoday, particularly when combined with the concept of international comparison andstatus-seeking behavior.In the spirit of Gerschenkron (1962), we interpret economic backwardness in terms

of a country�s distance from the world development frontier: it is a measure of rela-tive economic (under-) development that includes technological know-how, welfare, andconsumption possibilities.3 Particularly in an era of the globalization of information, abackward country�s poor or underprivileged population compares its situation not onlywith that of its better-o¤ co-nationals, but also with the situation of peers in neigh-boring countries and places farther a�eld.4 An unfavorable comparison, coupled with apolitical regime�s inability or unwillingness to respond to growing popular discontent,can then lead to a dangerous build-up of social tension. We propose that the greater acountry�s economic backwardness with respect to the development leader, the higher itsprobability of witnessing organized forms of social tension such as mass demonstrationsfor political regime change or even armed civil con�ict.5

We empirically test the hypothesized link between economic backwardness and so-cial tension by using extensive new data on violent and nonviolent mass movements atthe country-year level for the post-war period (Chenoweth and Lewis 2013), as well asestablished data on armed civil con�ict from the UCDP/PRIO Armed Con�ict Dataset(Gleditsch et al. 2002; Themnér and Wallensteen 2012). Our indicator of backward-ness is a simple measure of a country�s distance to the economic (and technological)world leader. In Gerschenkron�s context of pre-WWII industrialization, the develop-ment leader was England. During the last century, however, the United States havebeen at the forefront of global economic development. We therefore take the ratio of acountry�s per-capita income relative to that in the United States as our main measureof economic backwardness.Across a large series of pooled OLS and logit estimations, we �nd that economic

backwardness is an important and hitherto neglected factor in explaining mass move-ment, and to a lesser degree armed con�ict. Greater backwardness particularly increasesa country�s likelihood of seeing nonviolent mass demonstrations for regime change or se-cession. In a second step, we take potential endogeneity issues seriously by instrumentingour backwardness measure together with income per capita. We use the minimum phys-ical distance to either London or Washington, D.C., and mailing speed and telegramcharges around 1900 as exogenous instruments for economic backwardness. To the bestof our knowledge, the latter two are entirely novel instruments for economic develop-ment. Linear two-stage estimations reinforce our �ndings of a positive link between

3Our concept of economic backwardness is similar to distance to the technological frontier found inthe recent growth literature (e.g., Acemoglu et al. 2006, Madsen et al. 2010).

4See James (1987) for an early contribution on international comparison and emulation e¤ects, andValente (2009) for a recent formalization in a growth model.

5We are not attempting to explain the incidence of strikes or other forms of popular dissent thatdon�t aim at achieving maximalist political outcomes.

3

backwardness and the probability of witnessing both violent and nonviolent mass move-ments, and suggest moreover that the relationship is causal. Instrumental variables (IV)results for armed con�ict onset remain weaker, but still show a clear positive relationshipbetween backwardness and civil con�ict. In addition, the impact of backwardness hasbeen on the increase in recent decades, consistent with the idea that status comparisonshave become easier with globalization.The backwardness indicator generally has a strong positive e¤ect on the social ten-

sion outcomes. However, this does not necessarily mean that the indicators that havebeen proven to be related to con�ict in other studies �particularly income per capita,population size, institutional measures, and resource wealth �are no longer relevant.Instead, we contend that backwardness is a new and complementary factor that canhelp us explain the incidence of violent and nonviolent social unrest across countries.Our paper contributes to the growing empirical literature on nonviolent con�ict.

Svensson and Lindgren (2011) showed that since 1979, unarmed uprisings had becomemuch more frequent than armed ones in East Asia, yet di¤ered substantially in their aimsand outcomes. The seminal contribution by Chenoweth and Stephan (2011) studied theparticularities of nonviolent civil resistance and found that nonviolent campaigns havewider participation than violent insurgencies, are more successful, and also lead to thecreation of more durable and peaceful democracies. There is as yet little consensus onwhat factors are most robustly linked to the onset of nonviolent con�icts; and to ourknowledge, no contribution attempts to establish causality. Two other articles use thesame dataset we employ to measure nonviolent con�ict: Gleditsch and Rivera (2015)show that there is signi�cant di¤usion of nonviolent campaigns between neighboringstates. Butcher and Svensson (2014) show that modernization �de�ned in terms ofabsolute development and measured by the share of manufacturing value-added in GDP�is positively related to the onset of nonviolent campaigns. Using a di¤erent dataset,Chenoweth and Ulfelder (2015) �nd that youth bulges, regional contagion and (thelack of) civil liberties present the strongest relation to nonviolent uprisings, but thatmeasures of modernization �including the GDP share of manufacturing (and services)�are not strongly linked to nonviolent uprisings.The paper is also closely related to the vast armed con�ict literature, which has

established that per capita income levels are one of the key factors a¤ecting the like-lihood of an armed civil con�ict (see, among others, Collier and Hoe­ er 2004; Fearonand Laitin 2003; Hegre and Sambanis 2006). This research has linked increased con-�ict likelihood in poor countries to frustration and grievances, low opportunity cost inrecruiting rebels, and lack of military capabilities and state capacity to prevent and sup-press armed con�icts. Several recent contributions have tried to disentangle the causallinks between economic factors and con�ict outcomes by using instrumental variablesestimations (e.g., Miguel et al. 2004; Brückner and Ciccone 2010; Bergholt and Lujala2012). In our analysis, higher per capita income does not reduce the likelihood of vi-olent forms of social tension once we include backwardness, suggesting that these may

4

be more an outcome of relative backwardness rather than the absolute level of income.Particularly in our two-stage estimations, we even �nd that richer countries in absoluteterms are more likely to experience forms of social tension that fall short of outright civilwar, especially nonviolent mass movements, which may be explained by the strong em-pirical relation between income levels and democracy, and the institutionalized channelsfor (peacefully) expressing social and political dissent that democracies o¤er.The present paper is also linked to classical contributions in sociology and politi-

cal science that explore the origins of revolutions and social unrest. Davies (1962: 6)proposed an upside down J-curve to illustrate how revolutions are more likely to occurafter a sharp reversal of fortune following a prolonged period of "objective economic andsocial development". People compare their current situation with their own situationin the immediate past, and the greater the mismatch between what people want andwhat they get, the greater the "revolutionary mood". However, the size of this gap isindependent of income levels, i.e., poor satis�ed people will not revolt (and the trulyimpoverished will be too preoccupied with survival to spare much thought for revolu-tion), while rich dissatis�ed people may well do so. The J-curve is what Gurr (1970) inhis seminal contribution referred to as "progressive deprivation", one of three potentialsources of "relative deprivation" that can lead to political violence. Another sourceis "aspirational deprivation": expectations, e.g., for welfare and political freedom, risewhile capabilities (i.e., what you have) remain the same. Gurr�s theory reinforces ourargument that people compare their own status with that of others in other nations,but Gurr remained vague about the mechanism behind the comparison.In order to shed light on this comparison mechanism, our contribution draws on

the large literature on status-seeking and catching-up with the Joneses. The idea of acomparison of one�s own social status with peers was �rst advanced by Veblen (1899).Status-seeking behavior and its economic spillovers has since been the subject of a largenumber of articles (see, e.g., Duesenberry 1949; Carroll et al. (1997); Frank 1997;Alvarez-Cuadrado et al. 2004). Most of these papers, however, focus on status com-parison and catching-up behavior within a country. James (1987) �rst posited thatstatus-seeking behavior is important at the international level, too; it may be a conse-quence of the modernization process and have important welfare e¤ects. The idea ofinternational comparison and status-seeking has been formalized more recently by Va-lente (2009) in an endogenous growth model; and by Aronsson and Johansson-Stenman(2015) in a two-country model of international tax coordination.Finally, our paper contributes to the broader context of the recent political economy

literature that looks at historical origins of development. Acemoglu et al. (2006) andAcemoglu and Robinson (2012) re�ne and extend many of the arguments for why somecountries industrialize earlier than others that were advanced by Gerschenkron half acentury before. Their focus, however, is on economic development as the �nal outcomerather than on social tension and unrest.The rest of the paper is organized as follows. Section 2 develops the theory and

5

testable hypothesis; Section 3 describes the data and methodology; Section 4 discussesthe results of the empirical analysis; and Section 5 concludes.

2 On economic backwardness and social tension

We argue that economic backwardness comes at a cost: it not only entails widespreadpoverty in the backward country, but also increases the risk of falling prey to politicalinstability characterized by mass demonstrations for regime change or secession, andpossibly even civil war. As mentioned above, we draw inspiration from Gerschenkron�s(1962) seminal contribution on economic backwardness in pre-WWII Europe. In hisessay, Gerschenkron describes and compares three countries� industrialization experi-ences. In chronological order of industrialization � and therefore also in increasingorder of economic backwardness �he discusses the experiences of France, Germany, andRussia. Although Gerschenkron does not explicitly de�ne the meaning of "economicbackwardness", the main determining factor is the relative level of industrialization, thebenchmark for comparison for the period being England.Gerschenkron�s main focus is on analyzing the patterns of industrialization in these

countries, pointing out commonalities and di¤erences, and developing a general theoryon the pattern of and mechanisms behind industrialization. His main insight was thatthere is no �xed sequence to the industrialization steps that each country must takeon the road to economic development: instead, knowledge can be adapted from earlierindustrializers, opening the possibility for faster catching-up for lagging countries. He�nds three aspects that have proven to be important for industrialization and the levelof economic backwardness. The �rst is the institutional set-up found in the later in-dustrializers, starting with the prerequisite of eliminating institutional obstacles suchas serfdom, the absence of political uni�cation and trade barriers; and proceeding tothe development of the � industrial or universal �banking sector, or � in the case ofthe most backward �the need for heavy state intervention in �nancing the early stagesof industrialization. The widespread corruption and "general distrust of the public"in Imperial Russia is also mentioned as an obstacle to business-driven industrialization(ibid.: 19). These points have clearly found an echo in the recent literature on institu-tions and development (e.g., Acemoglu et al. 2006; Acemoglu and Robinson 2012). Thesecond aspect concerns country-speci�c characteristics such as the presence of naturalresources, which favors industrialization. The third and �nal aspect is the intellectualclimate: what today would be considered part of the national culture. This is mainlyde�ned by political ideology, in particular varying �avours of socialist ideology. We takethese aspects into account in our empirical approach below when choosing the controlvariables.An important observation is that the delay in industrialization, and consequently in

economic progress and potential social change it entails, could be a deliberate policy pur-sued by the political leaders. A few years after the Russian Revolution, Edwards (1927:

6

3) stated that all revolutions stem from the "repression of [people�s] elemental wishes",their ideas and ambitions, and that the greater the repression, the greater the violenceduring revolutions. Such delaying tactics by the ruling élites were employed for examplein Austria-Hungary, Bulgaria and Russia (Gerschenkron 1962). They were also partof the French colonial policies, which sought to avoid competition for the French econ-omy by dominated territories� industries; belated attempts at industrialization of thecolonies by the Vichy regime proved only half-hearted (Coquery-Vidrovitch 1981). Suchobstructionist policies are however not con�ned to the pre-WWII or colonial period. Forexample, Ekundare (1981) shows how post-colonial Nigerian development policies werehostage to socio-political constraints, particularly regional and ethnic politics, which �deliberately or not �delayed industrialization and economic development. This politicalperspective of economic backwardness has been formalized by Acemoglu and Robinson(2006). In their model, political élites can have an incentive to block technological andinstitutional progress because these threaten the status quo and increase the likelihoodof the élite�s replacement.Unfortunately, the missed opportunity for development is likely to come at the price

of mounting social tension, eventually proving the delaying tactics of the political éliteto be short-sighted, as change may be forced on it by a discontented population. Ger-schenkron used the metaphor of missing "the bus that is supposed to take a countryacross its great spurt of industrialization"; the wait for the next bus could be a long one,and it might not be "as convenient or as fast as its predecessor" (Gerschenkron 1962:363). During this time of forced delay, the tensions created by the suppressed forces ofeconomic development could erupt into outright violent con�ict, as witnessed in Russia.In order to distill these ideas into a theory of economic backwardness and social ten-

sion, we are missing one crucial element: the comparison with the frontier. The notionthat there is self-awareness on the part of a (large section of a) country�s population ofthe position on the ladder of relative backwardness is implicit in Gerschenkron�s writings.However, we can theoretically question the desire for catching-up (and even for freedomfrom suppression) in the under-developed country. Why should there be mounting dis-content among a suppressed entrepreneurship and labor force in a backward country?Why should they seek development in the �rst place and not be satis�ed with the statusquo, even as the economic development frontier moves farther and farther away?Gurr (1970) discussed the gap between a collectivity�s "value expectations" �"the

average value positions to which its members believe they are justi�ably entitled" �and "value capabilities" �"the average value positions its members perceive themselvescapable of attaining" (Gurr 1970: 27). This gap between aspirations and capabilities,termed "relative deprivation" (in later contributions "grievances" or "sense of injus-tice"), determines a group�s propensity to political violence. As mentioned above, onesource of relative deprivation is "aspirational deprivation", where aspirations rise butcapabilities remain the same. Rising aspirations are born from exposure to new modesof life through modernization, especially mass communications media, and the spread

7

of literacy and Western-style education. However, in a slightly convoluted argument,men are only likely to be mobilized by such demonstration e¤ects if they are already ina situation of deprivation, and especially if their situation is getting worse (Gurr 1970:ch. 4).6

Gurr�s theory brings us a step closer to the mechanism behind the comparison withthe frontier, which together with political repression or ineptitude acts as a catalyst forsocial tension in backward countries. But why would people harbor these aspirationsin the �rst place, particularly if their own situation had not objectively deteriorated?We believe the answer lies in a psychological trait that is closely related to economicbehavior: in human beings�tendency for interpersonal comparison and their desire forstatus (Frank 1997). Commonly referred to as "catching-up with the Joneses", thisbehavior has been shown to apply not only to individuals within a country, but alsoat the international level. Individuals in underdeveloped, backward countries comparetheir situation with that of peers in advanced countries and seek to catch up withthem. Although the awareness of relative status and development has plausibly beenaccentuated by the rapid pace of globalization and spread of the internet during the mostrecent decades, James (1987) argues that this comparison has been present for longer.He cites education, work in foreign �rms�a¢ liates, advertising, and historical contactthrough colonialism as potential mechanisms for "positional taste transfer" (ibid.: 455).In his view, this international taste transfer then leads to positional consumption by thepoor, with important welfare implications.7 More formally, Valente (2009) shows howthe incorporation of preferences for international status-seeking into a growth modelcan a¤ect convergence in growth rates and income levels. The model�s predictions areconsistent with the extraordinary growth performance of East Asian countries. Aronssonand Johansson-Stenman (2014, 2015) incorporate international comparisons into modelsof public-good provision and international tax coordination. There is as yet much lessliterature on international comparisons and catching-up behavior than on the within-country counterpart, but the contributions point in a clear direction: individuals inpoorer countries not only look to better-o¤ countries for their role models, but activelyseek to emulate these models.Goods consumption lies at the heart of the theory of "catching up with the Joneses",

but the human tendency for international comparison and imitation goes beyond thedesire to buy new goods. Recent events in the Arab Spring, and arguably also in the

6Gurr (1970) also di¤ers in his focus of what a¤ects the sense of backwardness or "relative depri-vation": although often vague in the details, he includes not only economic factors, but also politicalfreedom as part of "values". The combination of political and socio-economic iniquities gives rise toso-called grievance-based theories of con�ict (see also e.g., Cederman et al. 2011, Tilly 2003). Ourmain focus is on economic freedom, indicated by our choice of income per capita gaps as the measureof backwardness, though we control for political and social factors.

7In a European context, Becchetti et al. (2010) �nd that the comparison with better-o¤ neighboringcountries over the past thirty years has also had an impact on personal well-being and happinessperceptions, with the e¤ect being stronger the greater the media exposure.

8

Euro-Maidan demonstrations in the Ukraine, show that the frustrated desire to catch upwith the frontier can extend to the political sphere, particularly in repressive regimes.China�s considerable e¤orts to censor the information available on web-based media andsocial network sites demonstrate the totalitarian regime�s fear of how its populationmay be in�uenced by outside impressions and critical comparisons with the rest ofworld. The Turkish government�s recent (failed) attempt to deny access to social mediawebsites believed to have disseminated information on high-level corruption and to haveincentivized mass demonstrations for political opposition also show how modern massmedia can impact not only consumption but political preferences. In the past, evenin a country such as the Soviet Union, whose communist ideology was heavily biasedtowards holding up its own model as the global frontier to be aspired to, many peoplelooked to the West �particularly Britain and the United States �for their cultural andpolitical ideals.8

In general, it is plausible to a¢ rm that a country�s relative economic position withrespect to the most advanced nations matters for those excluded from the developmentprocess. If the desire to catch up with the development leaders is being �actively orinadvertently �suppressed by the political élites, this sense of exclusion can result inorganized resistance against the political regime; resistance which may take the form ofpeaceful demonstrations for change, or violent campaigns against the regime.In sum, we propose that the main mechanism at work is a combination of inter-

national comparison, and therefore the awareness of relative backwardness, and theinability to emerge from economic backwardness because of either political ineptitudeor the outright suppression of any entrepreneurial activity, which is perceived by the éliteas a potential threat to the status quo. Economic backwardness may lead to a dangerousbuild-up of social tension; once the lid is o¤, it can erupt into nonviolent mass protestsagainst the current regime, or even revolution. Our theoretical explanation for socialtension and con�ict is schematically summarized in Figure 1. Based on these arguments,we propose the following testable hypothesis: Greater economic backwardness will, allelse equal, lead to a higher probability of experiencing episodes of mass civil unrest.We do not contend of course that economic backwardness is the main culprit for

mass demonstrations or even civil wars. Instead, we believe that it complements othertheories on the origins of con�ict. In particular, economic backwardness is related tobut distinct from income per capita (i.e., income levels), which has proven to be oneof the most robust explanatory factors in cross-country armed con�ict studies (see e.g.,Hegre and Sambanis 2006). The former measures relative economic development, whilethe latter measures absolute levels of development. Poor countries have consistentlybeen found to be more prone to armed con�ict, be it because of low opportunity costsof enganging in combat (Collier and Hoe­ er 2004), or because per capita GDP can

8There is ample anecdotal evidence of how the younger Soviet generations in the Cold War erasought to own Western status symbols such as branded jeansware and listened to Beatles music. Thelatter was o¢ cially banned in the Soviet Union because of the band�s perceived politically corruptivee¤ects.

9

be interpreted as a measure of state capacity (Fearon and Laitin 2003). Modernizationtheory also looks at absolute levels of development, but focuses on nonviolent con�ict anddemocratization (Butcher and Svensson 2014, Chenoweth and Ulfelder 2015, Inglehartand Welzel 2005, Lipset 1959). The prediction is that economic development (i.e.,modernization, industrialization) leads to greater urbanization and the emergence ofa middle class willing and able to mobilize to demand more political and civil rights.We will consider both the relative and the absolute e¤ect of economic developmentsimultaneously, as there is little reason to believe that one excludes the other.

3 Empirical approach and data description

3.1 Methodology

In order to test our hypothesis, we use a two-pronged approach. First, we use linear esti-mations to establish the basic relationship. We concentrate on pooled OLS estimations,but also show our baseline speci�cations using pooled logit, which is the most commonapproach used in the con�ict literature. We estimate the following basic model:

social tensionit = a+ b+ �1 � backwardnessit + �2 � incomepcit + �3 �Xit + �it: (1)

We have a range of zero-one dummies as our dependent variable social tension inyear t in country i. Our interest is mainly focused on the backwardness indicator and itscoe¢ cient �1. However, absolute development is also important: income per capita hasbeen found to be the most robust explanatory variable in the armed con�ict literature. Itis also a proxy for modernization, which is an explanation for the emergence of nonviolentcon�ict. We will therefore also pay particular attention to the coe¢ cient �2. The vectorX includes other common suspects from the violent and nonviolent con�ict literature, aswell as variables derived from the theory above; all are described in more detail below.a is the constant term, b are decade dummies, and � the error term. We report robuststandard errors clustered at the country level in all tables. For logical reasons and as a�rst step towards addressing endogeneity issues, we lag most time-varying variables byone year (exceptions are the Cold War and socialist country dummies, and time sinceindependence and since the last con�ict).However, this approach still leaves some open questions regarding the endogene-

ity not only of our main explanatory variable, backwardness, but also of one of themost important other explanatory variable in the con�ict literature, namely income percapita. In particular, we have to consider the possibility of reverse causality. Economicbackwardness is a slow-changing variable, and a one-year lag cannot exclude potentialreverse-causality issues: social tension can build up over many years and �are up sev-eral times if the underlying problems are not solved, a pattern which in turn coulda¤ect backwardness, as con�ict becomes a setback for development. Similarly, income

10

per capita is also likely to be a¤ected by reverse causality, as (the threat of) con�ict �whether violent or not �is often quickly mirrored by the economy, for example, due tothe pull-out of investment in uncertain times. In both instances, the bias is likely toaugment the e¤ect on social tension and drive results in our favor. In addition, while weadd a wide variety of control variables, it is impossible to exclude completely the possi-bility of omitted variable bias, which would likely attenuate the e¤ect of backwardnessand income per capita found in OLS estimations. Which type of bias is stronger is anempirical question.Our second strategy explicitly deals with endogeneity by instrumenting economic

backwardness and per capita income in a series of pooled two-stage least squares (2SLS)estimations. In addition to a second stage similar to equation (1) above, we add a�rst-stage estimation of economic backwardness and per capita income levels:

backwardnessit = c+ b+ �1 � Ii + �2 �Xit + "it: (2)

incomepcit = d+ b+ �3 � Ii + �4 �Xit + "it: (3)

We have a total of three di¤erent exogenous instruments I at our disposal, which allowsus to achieve a strong �rst-stage identi�cation and to test for overidentifying restrictions.We next describe our data in more detail.

3.2 Data description

Our panel dataset covers the years 1946-2011 and includes up to 163 independent statesand over 7,800 country-years. The dataset contains countries for which we have economicdata and Polity IV data for regime type, and that had a population larger than 500�000in 2012. Summary statistics are presented in the Appendix (Table 6).Dependent variables. We use three measures for social tension, our dependent

variable, to test our hypothesis. The variables come from two separate datasets withdi¤erent de�nitions of con�ict, providing a good robustness test for our hypothesis.The �rst two measures cover both violent and nonviolent forms of social unrest and

are taken from a new panel dataset on nonviolent and violent campaign onsets �theNAVCO 2.0 dataset (Chenoweth and Lewis 2013). NAVCO 2.0 data is available for theperiod 1946-2006 and includes all sustained mass movements or "campaigns" that have aclear maximalist (political) objective (such as expelling a foreign occupier, secession, orchanging a regime), at least 1�000 participants, and recognizable leadership. Short-livedmovements, as well as most coups d�état, are not included. This de�nition well accordswith our theory on widespread social tension stemming from economic backwardness.The dataset includes campaigns that are observable in the sense that tactics used bythe participants are overt and documented. NAVCO is unique because it includes bothviolent and nonviolent mass movements in a detailed cross-country time-series dataset.To be coded as a nonviolent campaign, the mass movement must rely primarily onnonviolent tactics such as boycotts and civil disobedience, and it must not seek to

11

threaten or harm the opponent. Conversely, violent campaigns primarily rely on violenttactics with at least campaign-related 1,000 deaths. This means that some violence canoccur in a nonviolent campaign, but it cannot be the campaign�s main means to theend. An important note to keep in mind is that the de�nition of violent and nonviolentcampaigns focuses on the strategy of the mass movement, not on the answering strategyof the government, which may be violent in all cases. In total, the NAVCO dataset has251 campaign onsets, of which 142 are violent. Our dataset includes 216 NAVCO onsets(we do not consider independence campaigns in colonial countries), of which 90 arenonviolent onsets. We concentrate on the NAVCO nonviolent episodes and all (violentplus nonviolent) episodes.9

We also use information on the onset of armed civil con�ict from the annually up-dated UCDP/PRIO Armed Con�ict Dataset, which covers the period 1946�2011 (Gled-itsch et al. 2002; Themnér and Wallensteen 2012; version v.4-2012). We include allinternal and internationalized internal con�icts using the lower threshold of 25 annualbattle-related deaths. An onset is coded one if there is a new con�ict, or if there hasbeen a reactivation of a con�ict that has been inactive for more than two calendar years.In total our dataset has 290 con�ict onsets. The correlation between violent campaignsrecorded in the NAVCO dataset and armed civil con�icts in the UCDP/PRIO datasetis 0.28. In total, there are 58 country-years that are recorded in both datasets. Four ofthem are country-years for which NAVCO records a non-violent onset and UCDP/PRIOan onset of armed con�ict.10

A country with an ongoing con�ict of any type could experience a new con�ictoutbreak: we therefore do not drop years with ongoing con�ict, keeping our full samplesize. Each new con�ict onset is coded as one, and subsequent years with ongoing con�ictand no new outbreak are coded as zero. To control for the possibility that a countrythat is already experiencing con�ict (or that recently endured one) may be more likelyto experience another con�ict, we adopt a two-pronged approach (see also Lujala 2010).First, we include a variable called Peace years that counts the years since the last yearof con�ict (in logs), as suggested by Beck, Katz & Tucker (1998).11 Second, we add adummy variable for a con�ict in the previous year that will capture ongoing con�icts.Economic backwardness. Our main independent variable of interest is the proxy

for economic backwardness (Backwardness). Gerschenkron himself proposed "the size ofper capita income" as a natural measure of comparison of backwardness (Gerschenkron1970: 99). We construct a yearly backwardness measure based on the ratio of a country�sGDP per capita to the GDP per capita of the technology leader �the U.S. in the post-WWII period.12 The ratio is constructed using Maddison data (Bolt and van Zanden

9Results using only NAVCO violent campaigns as the dependent variable were weaker but otherwiseconsistent with those employing all episodes.10These are: Soviet Union 1990, Mali 1990, Niger 1991 and Indonesia 1997.11For the NAVCO dataset, we include the years of peace since the end of the last campaign of any

type in all speci�cations.12Note that this measure of backwardness is very similar to the proxy for "proximity to the technology

12

2013), with GDP per capita measured in 1990 international Geary-Khamis (Purchasing-Power-Parity-adjusted) dollar terms. The data is available on a yearly basis from 1800-2010 for 163 countries. The ratio can take values larger than 1 because a handful ofcountries have higher GDP per capita than the U.S. in some years. To ease interpretationof the results, the ratio is multiplied by -1, so that higher values correspond to greaterbackwardness. This means that we expect backwardness to be positively linked to socialtension. The countries with the largest ratios (that is, the most backward countries) inthe �rst decade of the 21st century are the Democratic Republic of the Congo (DRC),Sierra Leone, Burundi, Niger, Chad and the Central African Republic.Other covariates. There are many common explanatory variables used in the

armed and nonviolent con�ict literature, as well as some di¤ering ones. Common fac-tors include the (natural logarithm of) real GDP per capita (Income pc), the singlemost robust covariate from the empirical con�ict literature and arguably also a proxyfor modernization, which is an explanation for nonviolent con�ict. Yearly per-capita realGDP growth (Growth) and the (natural log of) population size (Population) are also incommon; data for all three measures come from the Maddison dataset.13 Political insti-tutions play a salient role in explaining all forms of con�ict: we use the revised "polity2"variable (Polity) from the Polity IV dataset (Marshall et al. 2013), which varies from -10(most autocratic) to 10 (most democratic). To measure non-linear e¤ects of Polity, wealso construct regime-type dummies by assigning the country to be an Autocracy if theoriginal "polity" score is smaller than -5; a Democracy if the "polity" score is larger than5; and an Anocracy for all the other "polity" scores, including those coded as missing.Other variables include data on ethnic, religious and linguistic fractionalization (Ethnic,Religious, Linguistic fractionalization, respectively) from Alesina et al. (2003) to controlfor the grievance e¤ects of ethnic divides. Decade dummies and a post-Cold War dummycontrol for time e¤ects. Data on Oil rents from the World Bank Development Indica-tors (World Bank 2013, available only after 1970) tests the resource-curse hypothesisin violent con�ict (see, e.g., Lujala 2010) and takes into account the importance thatGerschenkron (1962) gave to the presence of natural resources for industrialization andeconomic backwardness. Measures for Civil liberties from Freedom House and Youthbulges were both obtained from Chenoweth and Ulfelder (2015), who found them tobe particularly important in explaining nonviolent social con�ict. The index of Civilliberties varies from 1 (most liberal) to 7 (least liberal), and Youth bulges is measuredby the proportion of the population aged 15-24-years. A measure of Contagion has alsobeen found to be important for explaining nonviolent con�icts onset (Chenoweth andUlfelder 2015; Gleditsch and Rivera 2015). We construct a dummy variable denoting

frontier" for an industry employed by Acemoglu et al. (2006).13We completed missing population data in the Maddison dataset (Germany 1991-2011, Yugoslavia

1992, Montenegro 2007-2011, Ethiopia 1951-1992, Yemen 1951-1989, Vietnam 1955-1975) with datafrom the World Development Indicators (World Bank 2013) and the Penn World Tables (PWT 8.0,Feenstra et al. 2013). Maddison dataset includes population data only up until 2008. The years2009-2011 were extrapolated using the Maddison data.

13

the occurrence in a given year of another campaign or armed civil con�ict (respectivelyfor NAVCO and UCDP/PRIO datasets) in the same UN region.14

We also consider several control variables that are directly linked to our theory. Ger-schenkron (1962) mentions corruption as an obstacle to industrialization. Widespreadcorruption may also provide a further reason for mass discontent. With these considera-tions in mind, corruption could potentially a¤ect both backwardness and social tension,and not taking it into account could lead to omitted variable bias. We use the politicalcompetition (Polcomp) variable from the Polity IV dataset (Marshall and Jaggers 2013)as a proxy for corruption. Our reasoning is that greater political competition leadsto greater transparency at all levels of government, and from there also greater trans-parency and therefore less corruption within the �non-elected �bureaucracy. We usethis proxy as, to our knowledge, no other, more direct measure of corruption is availablefor the entire period since the end of WWII. Political competition is a component vari-able of the composite "polity" measure, but it is coded on a scale of 0 to 10 (ten beingthe most competitive system) and captures a speci�c aspect of a political regime. Werevise the political competition variable provided by the Polity IV dataset in a similarmanner to that used to construct the "polity2" measure: cases of foreign �interruption�(-66) are treated as missing; cases of �interregnum� or anarchy (-77) are assigned ascore of zero; and cases of �transition�(-88) are prorated on a case-by-case basis acrossthe span of the transition. The correlation coe¢ cient between the Polity and Polcompvariables is 0.23, so they are not measuring the same aspects of political institutions.Gerschenkron (1962) also mentioned country-speci�c culture as an important factor foreconomic backwardness; in particular, having a socialist system may (negatively) in�u-ence the development pattern. We construct a Socialist dummy based on Kornai (1992)and completed with information from the CIA Factbook for recent years. Finally, wealso control for time since 1945 or from the year of independence, if this is later than1945 (Independence), since according to the theory, national uni�cation is a necessaryprerequisite for the industrialization process to take o¤. We expect social tension dueto delayed development to increase with time since independence.Instrumental variables. We use three di¤erent exogenous instruments for eco-

nomic backwardness and income per capita. Our �rst instrument is a simple measure ofgeographical distance of a country�s present-day capital from either London or Wash-ington, D.C., whichever is shorter (Distance, in log kilometers).15 Gerschenkron (1962)pointed out that the temporal progression of industrialization from England across thechannel to France, Germany and then Russia was no mere geographical coincidence, butre�ected the greater time for di¤usion of new technologies the further away from thecenter of industrialization. This conviction is echoed by Bairoch (1988: 259), who states

14The UN regions are: Eastern Africa, Middle Africa, North Africa, Southern Africa, Western Africa,Caribbean, Cantral America, South America, Northern America, Central Asia, Eeastern Asia, SouthernAsia, South-Eastern Asia, Western Asia, Eastern Europe, Northern Europe, Southern Europe, WesternEurope, Australia and New Zealand, and Melanesia.15This data is provided by Kristian Gleditsch at http://privatewww.essex.ac.uk/~ksg/data-5.html.

14

that "... there is a de�nite correlation between the time when the process of change gotunder way and the distance from England: the countries nearest to England were as arule the soonest a¤ected by the Industrial Revolution." Since our focus is on the periodafter WWII, when the United States �rmly rose to the forefront as the world technologyleader, we add the U.S. capital as the second "pole of attraction", next to London. Thecloser a country is to either of these two technology and development leaders, the fasterwe expect the di¤usion process to have been, and the less backward the country shouldbe. For similar reasons, countries closer to London or Washington, D.C. should alsohave higher incomes per capita.We argue that this instrument is unlikely to a¤ect the social tension outcomes other

than through backwardness or income. One possible objection could invoke militarypolicy: an intervention by either the U.S. or the UK in a country experiencing episodesof social tension, and particularly armed civil con�ict, is more likely the greater thegeographical proximity. However, aspects of Realpolitik have historically played a greaterrole in such situations than physical distance, for example in the U.S. intervention inVietnam, or NATO�s intervention in Libya and its non-intervention in nearby Syria.Political "spheres of interest" reach beyond geography, and geographical distance isunlikely to systematically prevent (or encourage) the eruption of social tension.Our second instrument is based on mailing times around 1900 from either London

or Washington, D.C. This is, to our knowledge, the �rst time this information has beenused. We calculate mailing speed from either London or Washington, D.C �whicheveris faster � as miles covered per "mailing day" in 1903 (the �rst year after 1900 forwhich we found documentation for both the United Kingdom and the United States),using data on mailing times for regular correspondence (i.e., not packages) in daysfrom Post O¢ ce Department (1903) and Post O¢ ce (1903).16 17 Where mailing timeinformation for letters was missing, we used mailing times for parcels from the Post O¢ ceGuide (1903), subtracting three days (the average additional time for processing bulkiermail). Where this was also missing, we used extensive additional sources on travel timesand transportation routes at the time to approximate the mailing time between theclosest country we had data for and the entry point or the capital for the country inquestion. In the remaining cases, we assigned one of three values depending on whetherthe country could be reached roughly within 1-2 months, 3-6 months, or longer.18 Wedivided the mailing days by the (approximate) miles covered by the correspondence atthe time. This distance was given for all countries listed in Post O¢ ce Department

16Sometimes we had several cities for one country. In these cases, we chose the city with the shortestcorrespondence time, adjusting our distance calculations accordingly.17We are grateful to Jenny Lynch from the United States Postal Service and to employees of The

Royal Mail Archive for their generous help in �nding the data.18These remaining countries are Afghanistan, Bhutan, Burkina Faso, Chad, Kazakhstan, Kyrgyzstan,

Mali, Mongolia, Nepal, Niger, Rwanda, Swaziland, and Tajikistan. A dummy variable for countrieswhere we had to use own estimates was insigni�cant. The codebook for this and all other variables willbe made available by the authors.

15

(1903), and supplemented by own calculations based on mailing route information forthe remaining countries, using either the capital or main entry point for each country(the choice was determined by the information on mailing times).19 We then took thenatural logarithm to construct our �nal measure, Mailingspeed, which is expected to benegatively (positively) linked to backwardness (income).We argue that mailing times are directly related to economic development and back-

wardness. Not only did it take longer for correspondence to reach the more remoteparts of the world; but at equal distances, letters reached a more developed and better-connected country before its more "backwater" counterpart. For example, a letter postedin Washington, D.C. could be read by the recipient in Moscow, Russia a mere ten dayslater; while a correspondent in Bolivia would have to wait 37 days to receive writtennews from the U.S. �even though Bolivia�s capital is, at around 5100 miles, physicallycloser to Washington, D.C. than Moscow (over 5500 miles). We do not expect mailingtimes to have a direct e¤ect on social tension or violent civil con�ict. Former coloniesmight have been more developed on average than their non-colonial neighbors, and (for-mer) colonies are arguably more likely to have experienced con�ict of some form in thepost-WWII period. However, this link turns out to be tenuous at best: the correlationof our mailing speed measure with a British colony dummy is 0.05, and -0.38 for Frenchcolonies. Nevertheless, we will address this point in our robustness analysis below.The third instrument is also novel: we use telegram charges around 1900. In the 19th

century, the telegraph had launched "the greatest revolution in communications sincethe development of the printing press" (Standage 1998: 2), and not having access to thetelegraph network soon became a disadvantage for business. There were signi�cant dif-ferences in the way the telegraph system was used in America and Europe. In America,80-90 percent of cables were business-related and telegraph managers were more con-cerned "about dispatch than low tari¤s... [In Europe] the telegraph is used principallyfor social correspondence�(Gardiner Hubbard, quoted in Standage 1998: 158). Therewere also di¤erences in the organization of the telegraph systems on either side of theAtlantic: while largely seen as a public utility in Europe and consequently run by publicmonopolists (e.g., by the Post O¢ ce in the United Kingdom), in America the telegraphnetworks were privately owned, though by the start of the 20th century Western Unionhad a virtual monopoly in the United States.Telegram pricing principles were however similar across the globe. The cost of send-

ing a telegram depended on distance and the number of words in the message. Includedin the charge was also the labor cost, right down to the messenger boy (or the telegrapho¢ cer himself in rural parts) who delivered the telegram from the telegraph o¢ ce to thedoor of the receiver (Downey 2002). The pricing basis was usually for ten words, andthen per word for each additional word (Ross 1928, Downey 2002). By the end of the19th century, the cost of sending a telegram had been greatly reduced thanks to a com-

19For sea and navigable rivers we used Google Earth to calculate the distances and for many inlanddestinations we used the length of railway travel.

16

bination of widely expanded cable networks and growing competition with telephones.In the 1870s, charges for international telegrams had still been very high: a transat-lantic telegram cost £ 20 or around $100 at the time (Standage 1998). Around 1900, thecharge of an international telegram sent, for example, from London to Bathurst (today�sBanjul) in The Gambia (42 pence per word) would have taken into account not onlythe distance covered (around 2700 miles), but also the relatively poor infrastructure andrelated extra costs of getting the cable to its recipient in that colonial outpost. Sendinga telegram several thousand miles further to more developed Bathurst in New SouthWales, Australia would have cost 8 pence less (34 pence per word) than sending it tothe West African Coast.The data for telegram prices is based on "Charges for foreign and colonial telegrams"

in pence per word as listed in Post O¢ ce (1903). The historical tables are remarkablycomplete: only for 11 countries do we have no available information at all on the chargingprinciple applied. For these, we combine information from historical maps of telegramnetworks with information on the communication routes at the time compiled to con-struct our mailing speed variable (see above) to assign charges based on remoteness.20

We then take the natural logarithm to construct our instrument Telegram. The corre-lation with economic development is evident from the criteria for telegram pricing: weexpect telegram charges to be positively (negatively) linked to backwardness (income).Our arguments in favor of exogeneity of this instrument are similar to the ones for themailing speed variable given above. Any potential direct e¤ect on social tension in thepost-WWII period should be additionally mitigated by the fact that the use of telegramswent on a sharp decline at the start of the 20th century as telephone systems expandedrapidly.

4 Results

4.1 OLS and logit estimations

We will discuss in turn the results using NAVCO and UCDP/PRIO data for our de-pendent variable, social tension. Table 1 shows the main OLS results for the NAVCOdata. Panel A shows the �ndings for nonviolent campaigns only, while Panel B showthe �ndings for all campaigns. Column 1 gives a parsimonious baseline speci�cationincluding only some of the most robust variables from the con�ict literature. In column8, we show pooled logit results using our baseline speci�cation for easier comparisonwith the existing literature. Columns 2-7 add di¤erent control variables. First, we notethat backwardness is positive with both dependent variables and that the coe¢ cientsalso have similar magnitudes across all speci�cations. The results are highly signi�cantfor the nonviolent campaigns, but less so when all campaigns are included. These initial

20A dummy variable for our own estimates was not signi�cant.

17

�ndings suggest that economic backwardness indeed has a positive relation with violentand nonviolent social tension.When we look at the other explanatory variables, we �nd one striking di¤erence with

respect to the conventional armed con�ict literature in the coe¢ cient for income percapita. What is arguably the strongest and most consistent (violent) con�ict-reducingfactor is now positive and signi�cant when considering nonviolent campaigns (Panel A),and insigni�cant when it comes to all types of campaigns (Panel B). This suggests twothings: �rst, social tension that falls short of outright armed civil con�ict may have moreto do with relative economic backwardness and the comparison with the frontier thanabsolute income and the in�uence of opportunity costs. Second, more a­ uent countriestend, on average, to engage more in nonviolent movements for political change than inviolent ones. A positive link between income per capita and nonviolent campaign onsetwas also found by Chenoweth and Lewis (2013), who posit that nonviolent resistanceemerges where resistance is supposedly more di¢ cult.21 we note that high income percapita is strongly correlated with high democracy levels, and that strong democratic sys-tems o¤er more institutionalized opportunities for the (nonviolent) expression of populardissent than autocracies or weak democracies.22

The results for violent civil con�ict, using the UCDP/PRIO data, are shown in Table2. Speci�cations are analogous to those in Table 1. Backwardness is positive in mostmodel speci�cations, but only signi�cant in our logit models (column 8). GDP percapita is negative and mostly signi�cant.The remaining covariates show similar results for all measures of social tension.

Polity mostly has a negative and signi�cant sign for NAVCO campaigns, suggestingthat (stronger) democracies see fewer campaigns on average. It is insigni�cant forUCDP/PRIO armed con�ict onsets. In columns 2, the break-down into regime types�Autocracy, Democracy and Anocracy (the latter is the omitted variable) � to takenon-linearity into account shows that democracies have a clear tendency to experiencefewer episodes of nonviolent and violent campaigns, with highly signi�cant, negative co-e¢ cients in both panels. Autocracies also see fewer campaigns in general (Panel B), butthey are neither more nor less likely to see nonviolent forms of mass protest than otherpolitical systems. Both democracies and autocracies show a weak tendency to experi-ence fewer armed con�icts. These �ndings are consistent with the non-linear relationshipbetween political regimes and social unrest that has been found in the armed con�ictliterature, where weak regime types (i.e., anocracies) are most prone to experiencingcivil war.Larger countries, measured by their population size, consistently see more social ten-

21Income per capita was positive but insigni�cant in Chenoweth and Lewis (2013), who used datafor 1946-1999 in their estimations.22Exploratory estimations using an interaction term between democracy and backwardness showed

that democratic backward countries were much more likely to see both nonviolent and any form ofmass campaign than non-democratic countries. However, analysing this point in detail is left forfurther research.

18

sion on average, which again con�rms the �ndings from the armed con�ict literature.The strong decreasing trend in the frequency of both violent and nonviolent campaignssince the end of the Cold War from that literature is also con�rmed. Ethnic fractional-ization (columns 3) has no e¤ect on nonviolent or violent campaigns, but increases theincidence of armed civil con�ict. In additional estimations shown in the Appendix (Ta-bles 9-10), language and religious fractionalization also have no signi�cant relation withNAVCO campaigns, though language franctionalization is positively linked to armedcon�ict. Economic growth (columns 4) shows no link with any form of social tension.Oil-rich countries �measured by their resource rents �have a weak negative link withnonviolent campaign onset (column 7), which runs counter to the conventional resourcecurse hypothesis, but is in line with Gerschenkron�s thoughts about natural resourcesbeing important for industrialization.Turning to our other theory-speci�c covariates, we note that our proxy for corruption

�political competition �shows some contradictory e¤ects in: it doesn�t appear to a¤ectthe likelihood of nonviolent campaigns or armed con�ict at all, while it has a strongcampaign-reducing e¤ect when all campaigns are considered. This �nding is driven bythe violent campaigns, and it supports our expectation that less corruption (i.e., morepolitical competition) is linked to less social unrest. Finally, socialist countries show anincreased frequency of nonviolent mass campaigns and a decreased frequency of armedcon�ict. This is interesting, as to our knowledge it is the �rst time that a "socialisme¤ect" has been considered in the con�ict and social tension context. It is possible thatthis e¤ect is driven by the largely peaceful demonstrations against the Soviet regimein Central and Eastern Europe during �and especially at the end of �the Cold War.However, robustness tests limiting the period of analysis to post-1970 diminish insteadof increase the socialism e¤ect (see Appendix, Table 7-8).In further robustness tests (see the Appendix, Table 7-8), we �rst restrict the sam-

ple period to after 1970. Rising globalization has made the comparison with othercountries easier and more immediate and it is thus possible that the importance of eco-nomic backwardness as an explanatory factor for social tension has been on the increasein more recent decades. The results show this to be case particularly for nonviolentcampaigns: the magnitude of the coe¢ cients is larger compared with the results fromthe full time period. Results for all types of campaigns and for armed con�ict do notchange substiantially. Another noteworthy di¤erence with the restricted time period isthat autocracies have no longer been signi�cantly less likely to see any type (violent ornonviolent) of campaign in recent decades.We also include more covariates from the violent and nonviolent con�ict literature

(Appendix, Table 9-10), none of which much a¤ects our results, particularly for nonvi-olent con�ict. We note that civil liberties and nonviolent campaigns have an invertedU-shape relation as in Chenoweth and Ulfelder (2015): the countries in which peoplehave the least and most civil rights are less likely to experience an campaign. Youthbulges, although not signi�cant themselves, generally weaken our results.

19

4.2 Instrumental variables estimations

Although suggestive, the results of the pooled OLS and logit estimations discussed aboveleave some open questions regarding the exogeneity of our main explanatory variable,economic backwardness, as well as income per capita. In particular, we noted thepossibility of reverse causality, as well as the ever-present chance of omitted variablebias. Can we further strengthen our conclusions by drawing a credible causal linkbetween greater economic backwardness and increased probability of social tension?This is what we attempt in our pooled 2SLS estimations that use distance and mailingspeed from London orWashington, D.C., and telegram charges as exogenous instrumentsfor backwardness and GDP per capita. The main results are presented in Tables 3-4for NAVCO campaigns and in Tables 5 for UCDP/PRIO armed civil con�icts, withspeci�cations echoing those of the OLS estimations presented above.The 2SLS results generally strengthen the �ndings from the OLS estimations for

backwardness: it has a strong, positive e¤ect on the likelihood of a country witnessingnonviolent and any type of campaign included in NAVCO. Although not signi�cant atconventional levels, backwardness positively a¤ects the onset of armed con�ict, too.23

Compared with the OLS results, the magnitude of the coe¢ cients for backwardness isapproximately tripled for nonviolent campaigns, increases around �ve-fold for all typesof campaigns, and roughly doubles for armed con�ict. We also con�rm the strongpositive e¤ect of income per capita on nonviolent campaigns and � interestingly � asimiliar positive e¤ect on all forms of mass campaign. These �ndings seem to contradictthe armed-con�ict literature, where the richer countries usually show less tendency todescend into civil war. Our results suggest that episodes of social unrest that fall shortof civil war, and especially nonviolent mass movements, are more likely to happen inricher countries. Income per capita shows no e¤ect on armed con�ict: the inclusion ofrelative development has made the impact of absolute development disappear.24

In terms of impact, the models indicate that backwardness has a substantial e¤ect onsocial tension. This can be illustrated by using a couple of examples from our NAVCOresults, which are the most signi�cant. From 1987 onwards for over a decade, Mexicoexperienced a series of protests aimed at the government that the NAVCO dataset codesas a nonviolent campaign. In 1986, Mexico�s score for backwardness was -0.27, whichincidentally is also the mean score for backwardness in our dataset. Results from ourbaseline model for nonviolent campaigns (Tables 3, column 1) suggest that had Mexicobeen one standard deviation (0.27) less backward �i.e., with a similar score to Israel

23Signi�cance levels on the backwardness coe¢ cient are between 10-20%.24Note that the sizeable increase in the magnitude of both the backwardness and income per capita

coe¢ cients suggests that the upward bias in OLS estimations due to reverse causality is much smallerthan the probable downward bias caused by omitted variables. Miguel et al. (2004) similarly foundthat reverse causality bias from con�ict to economic growth was not as great as expected. In their case,they argued that it was outweighed by the bias introduced by measurement error in economic data inAfrican countries.

20

(-0.57) and Singapore (-0.51) � the risk of a nonviolent campaign would have beenreduced by 2.1 percentage points. The contrast is even starker when comparing Mexicoto its northern neighbor: being at a par with the United States, the development leader,would have decreased Mexico�s probability of a nonviolent campaign by an additional3.5 percentage points, or 5.6 percentage points in total. As an illustration for violentcampaigns, we can consider the South African rebellion against apartheid that emergedin 1984 with the introduction of the new constitutions. The movement used both violentand nonviolent methods and is coded as a violent campaign in the NAVCO dataset.In 1983, South Africa�s score for backwardness was -0.22, placing it among the mostdeveloped countries in Africa. One standard deviation decrease in that score, bringingit to the same level as Spain (-0.50), would have lowered the likelihood of con�ict by3.5 percentage points (using the estimate from Table 4, column 1). These are largeand substantial impacts considering that the average rates for nonviolent and violentcampaign onsets in our dataset are 0.012 and 0.027, respectively.The control variables have largely similar e¤ects on the chance of witnessing social

tension and armed con�ict as found in the OLS and logit estimations. Di¤erencesinclude ethnic fractionalization, which has a general campaign-increasing as well asarmed con�ict-increasing e¤ect; economic growth, which has a negative e¤ect on thelikelihood of witnessing all types of NAVCO campaigns; and the dummy for socialistcountries, which is only signi�cantly decreasing armed con�ict onset.In additional estimations (see the Appendix, Tables 11-8), we again restrict the

sample period to post-1970 to see whether the e¤ect of backwardness has been on theincrease. We see a notable increase in the magnitude of the coe¢ cients for backward-ness for all measures of social tension, con�rming that the e¤ect of backwardness onsocial tension has increased in parallel with the increase in globalization. Additionalsensitivity tests with more covariates (see the Appendix, Tables 14-16) also support ourmain results, with the notable exception of the speci�cations with youth bulges, whoseinclusion �though never signi�cant itself �weakens the main results for backwardnessand income per capita.Instrument validity. Our causal interpretation of the e¤ects of economic back-

wardness and income per capita rests on the strength of our instruments. In order totest this, in addition to our theoretical arguments given above in Section 3.2, we alsoconsider a range of more formal tests. All 2SLS tables provide �rst-stage exogenousinstrument t-statistics, excluded instrument F-statistics, and Hansen J test p-values atthe bottom. The Hansen J test for over-identifying restrictions can never reject thejoint null hypothesis that our three instruments are valid, i.e., uncorrelated with theerror term, and that they are correctly excluded from the second-stage equation. Theinstrument t-statistics show that our instruments are strong and a¤ect backwardnessand income in the expected manner. The distance variable proves to be a slightly betterinstrument for backwardness than income per capita, while mailing speeds around 1900show a stronger link with income per capita. Telegram charges around 1900 are a good

21

instrument for both endogenous RHS variables across nearly all our speci�cations. Thepartial r-squareds for both �rst-stage estimations are comfortably high, generally rang-ing between 0.20-0.27, showing that our combination of instruments is able to capture agood part of the variation in backwardness and income per capita. The excluded instru-ment F-statistics also give reassurance that the inference is robust to the possibility ofweak instruments: in our main tables, they are always above 10 (the generally accepted"rule of thumb" value).25

5 Conclusion

In this paper we extend and investigate the idea �rst put forward by Alexander Ger-schenkron (1962) that economic backwardness can increase the emergence of social ten-sion. Our paper has three novel points: �rst, we develop a theory that combines insightsfrom economic history and political economy with recent �ndings on individuals�ten-dency for international comparison and status-seeking. Economic backwardness, de�nedas relative economic under-development or distance from the technology and develop-ment frontier, increases the likelihood of witnessing outbursts of social tension among apopulation whose desire for catching-up with the development leader is being frustrated.Second, we test our hypothesis on the link between economic backwardness and so-

cial tension empirically, using new data on nonviolent and violent forms of mass move-ments, and established data on armed civil con�ict. Third, our approach includes notonly pooled OLS estimations and, for comparison with the empirical con�ict litera-ture, pooled logit estimations, but also two-stage least squares estimations to determinecausality. The latter address the potential endogeneity of backwardness and income percapita by using three exogenous instruments, two of which are entirely novel.The results strongly suggest that economic backwardness contributes to the emer-

gence of social tension in the form of mass (violent and nonviolent) movements, andto a lesser degree also armed con�ict. This e¤ect is not only causal, but also appearsto be on the increase in more recent decades, in parallel with rising globalization andthe di¤usion of rapid channels of communication. This supports our theoretical linkbetween economic backwardness and international comparison with better-o¤ peers asa mechanism that contributes to the eruption of social tension.Of course, we do not propose economic backwardness as the main explanation for

mass demonstrations, much less armed civil con�ict. Instead, we believe that it com-plements other theories on the origins of political violence and con�ict. In particular, it25Note that we also kept track of the Stock-Yogo weak identi�cation test statistics, based on the

Kleibergen-Paap rk Wald F statistic and using the critical values for Cragg-Donald F statistics andi.i.d. errors. The results suggest that we can generally not reject the null hypothesis that the truesigni�cance level of hypothesis tests based on 2SLS is below 20%. However, given our use of clustereds.e., the exact critical values that apply are in fact unknown, which is why we prefer to use the excludedinstrument F-statistics as our test of weak identi�cation (see Baum et al. 2007 for a formal discussionand recommendation of this point).

22

is related to the measures of absolute levels of development based on income per capitathat have proven to be one of the most robust explanatory factors in cross-country con-�ict studies. In our �ndings, income levels often do not only become insigni�cant whenincluded together with economic backwardness, but actually show that higher incomelevels increase the likelihood of mass movements that fall short of civil war. This holdsalso when we instrument income levels. These results open up interesting new avenuesfor future research.

References

Acemoglu, Daron, Philippe Aghion, and Fabrizio Zilibotti (2006). Distance to frontier,selection, and economic growth. Journal of the European Economic Association4(1): 37-74.

Acemoglu, Daron and James Robinson (2006). Economic backwardness in politicalperspective. American Political Science Review 100 (1): 115-131.

Acemoglu, Daron and James Robinson (2012). Why nations fail. New York: Crown.

Alesina, Alberto, Arnoud Devleeschauwer, William Easterly, Sergio Kurlat and RomainWacziarg (2003). Fractionalization. Journal of Economic Growth 8(2): 155�194.

Alesina, Alberto, Sule Ozler, Nouriel Roubini, and Phillip Swagel (1996). PoliticalInstability and Economic Growth. Journal of Economic Growth 1: 189-211.

Alvarez-Cuadrado,Francisco, Goncalo Monteiro, and Stephen J.Turnovsky (2004). Habitformation, catching-up with the Joneses, and economic growth, Journal of Eco-nomic Growth 9: 47�80.

Aronsson, Thomas and Olof Johansson-Stenman (2014). When Samuelson met Veblenabroad: national and global public good provision when social comparisons matter.Economica 81: 224-243.

Aronsson, Thomas and Olof Johansson-Stenman (2015). Keeping up with the Jone-ses, the Smiths and the Tanakas: On international tax coordination and socialcomparisons. Journal of Public Economics 131: 71-86.

Bairoch, Paul (1981). The main trends in national economic disparities since the Indus-trial Revolution. In: Bairoch, Paul and Maurice Lévy-Leboyer (eds.), Disparitiesin economic development since the Industrial Revolution. New York: St. Martin�sPress.

Bairoch, Paul (1988). Cities and economic development: From the dawn of history tothe present. Chicago: University of Chicago Press.

23

Baum, Christopher F., Mark E. Scha¤er, and Steven Stillman (2007). Enhanced rou-tines for instrumental variables/generalized method of moments estimation andtesting. The Stata Journal 7 (4):465-506.

Becchetti, Leonardo, Stefano Castriota, Luisa Corrado, Elena Giachin Ricca (2010).Beyond the Joneses: Inter-country income comparisons and happiness. The Jour-nal of Socio-Economics 45: 187�195.

Beck, Nathaniel, Jonathan N. Katz, and Richard Tucker (1998). Taking time seriously:Time-series�cross-section analysis with a binary dependent variable. AmericanJournal of Political Science 42(4): 1260�1288.

Bergholt, Drago and Päivi Lujala (2012). Climate-related natural disasters, economicgrowth, and armed civil con�ict. Journal of Peace Research 49(1): 147�162.

Bolt, J. and J. L. van Zanden (2013). The First Update of the Maddison Project;Re-Estimating Growth Before 1820. Maddison Project Working Paper 4. Avail-able at http://www.ggdc.net/maddison/maddison-project/publications/wp4.pdf.(Accessed 19 September, 2013).

Brückner, Markus and Antonio Ciccone (2010). International commodity prices, growthand the outbreak of civil war in sub-saharan Africa. Economic Journal 120: 519�534.

Brunnschweiler, Christa N. and Erwin H. Bulte (2009). Natural Resources and ViolentCon�ict: Resource Abundance, Dependence and the Onset of Civil Wars. OxfordEconomic Papers 61 (4): 651-674.

Butcher, Charles and Isak Svensson (2014). Manufacturing dissent: Modernization andthe onset of major nonviolent resistance campaigns. Journal of Con�ict ResolutionDOI:10.1177/0022002714541843

Carroll, Christopher D., Jody Overland, and David N. Weil (1997). Comparison utilityin a growth model, Journal of Economic Growth 2: 339�67.

Cederman, Lars-Erik, Nils B. Weidmann, and Kristian S. Gleditsch (2011). Horizon-tal inequalities and ethnonationalist civil war: a global comparison. AmericanPolitical Science Review 105 (3): 478-95.

Chenoweth, Erica and Orion A. Lewis (2013). Unpacking nonviolent campaigns: In-troducing the NAVCO 2.0 dataset. Journal of Peace Research 50(3): 415�423.

Chenoweth, Erica and Maria J. Stephan (2011). Why civil resistance works: Thestrategic logic of nonviolent con�ict. New York: Columbia University Press.

24

Chenoweth, Erica and Jay Ulfelder (2015). Can structural conditions explain the onsetof nonviolent uprisings? Journal of Con�ict Resolution DOI:10.1177/0022002715576574

Collier, Paul and Anke Hoe­ er (2004). Greed and grievance in civil war. OxfordEconomic Papers 56(4): 563�596.

Coquery-Vidrovitch, Catherine (1981). Industry and empire: the beginnings of Frenchindustrial politics in the colonies under the Vichy regime. In: Bairoch, Pauland Maurice Lévy-Leboyer (eds.), Disparities in economic development since theIndustrial Revolution. New York: St. Martin�s Press.

Davies, James C. (1962). Toward a Theory of Revolution. American SociologicalReview, 27(1): 5�19.

Downey, Gregory J. (2002). Telegraph Messenger Boys: Labor, Communication andTechnology, 1850-1950. London: Routledge.

Duesenberry, James S. (1949). Income, Savings and the Theory of Consumer Behavior.Cambridge, MA: Harvard University Press.

Edwards, Lyford P. (1927). The natural history of revolution. Chicago: University ofChicago Press.

Ekundare, Richard O. (1981). Constraints to economic development in Africa: somedeterminants of economic disparities. In: Bairoch, Paul andMaurice Lévy-Leboyer(eds.), Disparities in economic development since the Industrial Revolution. NewYork: St. Martin�s Press.

Fearon, James D. and David D. Laitin (2003). Ethnicity, insurgency, and civil war.American Political Science Review 97(1): 75�90.

Feenstra, Robert C., Robert Inklaar and Marcel P. Timmer (2013). The Next Gener-ation of the Penn World Table, available for download at www.ggdc.net/pwt

Frank, Robert H. (1997). The frame of reference as a public good. Economic Journal107: 1832�47.

Gerschenkron, Alexander (1962). Economic backwardness in historical perspective.Cambridge: Harvard University Press.

Gerschenkron, Alexander (1970). Europe in the Russian mirror: Four lectures in eco-nomic history. London: Cambridge University Press.

Gleditsch, Kristian S., and Mauricio Rivera (2015). The di¤usion of nonviolent cam-paigns. Journal of Con�ict Resolution DOI: 10.1177/0022002715603101.

25

Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg andHåvard Strand (2002). Armed con�ict 1946�2001: A new dataset. Journal ofPeace Research 39 (5): 615�637.

Gurr, Ted R. (1970). Why men rebel. Princeton: Princeton University Press.

Hegre, Håvard and Nicholas Sambanis (2006). Sensitivity analysis of the empiricalliterature on civil war onset. Journal of Con�ict Resolution 50(4): 508�535.

James, Je¤rey (1987). Positional goods, conspicuous consumption and the interna-tional demonstration e¤ect reconsidered. World Development 15(4): 449-462.

Kornai, Janos (1992). The Socialist System: The Political Economy of Communism.Princeton: Princeton University Press and Oxford: Oxford University Press.

Lujala, Päivi (2010). The Spoils of Nature: Armed Civil Con�ict and Rebel Access toNatural Resources. Journal of Peace Research 47(1): 15�28.

Madsen, Jakob, Rabiul Islam, and James Ang (2010). Catching up to the technologyfrontier: The dichotomy between innovation and imitation. Canadian Journal ofEconomics 43 (4): 1389-1411.

Marshall, Monty G., Ted Robert Gurr and Keith Jaggers (2013). Polity IV Project:Dataset Users�Manual. Center for Systemic Peace. http://www.systemicpeace.org/inscr/p4manualv2012.pdf(Accessed 19 September, 2003).

Miguel, Edward, Shanker Satyanath and Ernest Sergenti (2004). Economic Shocks andCivil Con�ict: An Instrumental Variables Approach. Journal of Political Economy112 (4): 725-53.

Pinker, Steven (2011). The better angels of our nature. London: Allen Lane.

Post O¢ ce (1903). Post O¢ ce Guide n.187. London: Post O¢ ce (Guidance & Ser-vices).

Post O¢ ce Department (1903). United States O¢ cial Postal Guide, January 1903.Philadelphia: United States of America Post O¢ ce Department.

Ross, Nelson E. (1928). How to write telegrams properly. Little Blue Book n. 459.Girard (USA): Haldeman-Julius Publications.

Themnér, Lotta and Peter Wallensteen (2012). Armed Con�ict, 1946-2011. Journal ofPeace Research 49(4): 565-575.

Tilly, Charles (2003). The politics of collective violence. Cambridge (UK): CambridgeUniversity Press.

26

Valente, Simone (2009). International status seeking, trade, and growth leadership.Canadian Journal of Economics 42 (2): 554-589.

Svensson, Isak and Mathilda Lindgren (2011). From bombs to banners? The declineof wars and the rise of unarmed uprisings in East Asia. Security Dialogue 42(3):219-237.

Veblen, Thorstein (1899). The Theory of the Leisure Class: An Economic Study ofInstitutions. London: Allen and Unwin.

World Bank (2013) World Development Indicators. Washington, DC: World Bank(http://data.worldbank.org/data-catalog).

27

Figures and tables

Figure 1: Schematic representation of the theory

28

Table 1: Backwardness and social tension: NAVCO campaigns

(1) (2) (3) (4) (5) (6) (7) (8)OLS OLS OLS OLS OLS OLS OLS logit

Panel A : non-violentBackwardnesst�1 0.023** 0.023** 0.023** 0.023** 0.023** 0.021** 0.027** 3.05**

(2.27) (2 .28) (2 .24) (2 .24) (2 .28) (2 .16) (2 .40) (2 .06)Incom e p.c.t�1 0.0062** 0.0070** 0.0064* 0.0062* 0.0062* 0.0057* 0.0079** 0.68**

(2.01) (2 .21) (1 .94) (1 .95) (1 .96) (1 .96) (2 .29) (2 .17)Populationt�1 0.0065*** 0.0064*** 0.0067*** 0.0066*** 0.0065*** 0.0060*** 0.0076*** 0.48***

(6.05) (6 .04) (5 .88) (6 .04) (6 .05) (6 .30) (6 .62) (7 .82)Post Cold War -0 .024** -0 .025** -0 .021** -0.024** -0 .024** -0 .023** -0 .023** -1 .70***

(-2 .30) (-2 .37) (-2 .07) (-2 .31) (-2 .29) (-2 .22) (-2 .23) (-3 .44)Polityt�1 -0 .00097*** -0 .00098*** -0 .00098*** -0 .00092* -0 .00079*** -0 .0014*** -0 .079***

(-4 .25) (-4 .07) (-4 .22) (-1 .95) (-3 .95) (-4 .71) (-3 .81)Auto cracyt�1 -0 .0028

(-0 .74)Democracyt�1 -0 .019***

(-4 .30)Ethnic frac. 0 .0010

(0.15)G rowtht�1 0.012

(0.79)Polcompt�1 -0 .00012

(-0 .11)So cia list 0 .013*

(1.73)O il rentst�1 -0 .00064*

(-1 .78)R2 0.012 0.013 0.012 0.012 0.012 0.013 0.014Panel B : allBackwardnesst�1 0.020* 0.018 0.024* 0.022* 0.020* 0.019 0.018 2.94***

(1.71) (1 .54) (1 .88) (1 .70) (1 .67) (1 .57) (1 .37) (3 .24)Incom e p.c.t�1 -0 .0029 -0 .0017 -0 .0011 -0 .0025 -0 .0022 -0 .0034 -0 .0027 0.16

(-0 .80) (-0 .46) (-0 .28) (-0 .66) (-0 .61) (-0 .94) (-0 .66) (0 .97)Populationt�1 0.0095*** 0.0097*** 0.010*** 0.0095*** 0.0094*** 0.0091*** 0.010*** 0.33***

(6.74) (6 .62) (6 .79) (6 .66) (6 .54) (6 .36) (7 .36) (6 .80)Post Cold War -0 .033** -0 .034*** -0 .031** -0 .035*** -0 .032** -0 .032** -0 .032** -1 .02***

(-2 .52) (-2 .66) (-2 .36) (-2 .65) (-2 .48) (-2 .46) (-2 .46) (-3 .45)Polityt�1 -0 .00083*** -0 .00085** -0 .00086*** 0.00089 -0 .00069** -0 .0013*** -0 .027*

(-2 .62) (-2 .57) (-2 .63) (1 .09) (-2 .11) (-3 .38) (-1 .92)Auto cracyt�1 -0 .013**

(-2 .28)Democracyt�1 -0 .029***

(-4 .80)Ethnic frac. 0 .0097

(1.09)G rowtht�1 -0 .038

(-1 .31)Polcompt�1 -0 .0039**

(-2 .28)So cia list 0 .011

(1.06)O il rentst�1 -0 .00043

(-0 .95)R2 0.013 0.016 0.014 0.014 0.014 0.014 0.016Observations 7,115 7,186 6,914 6,994 7,115 7,115 4,968 7,115Countries 160 160 154 159 160 160 160 160

Notes: The dependent variable in panel A is the onset of a non-violent campaign, and in Panel B the

onset of any type of campaign. All speci�cations control for an ongoing episode in the previous year

and peace years and include decade dummies and a constant term (not shown). S.e. are clustered at

the country level. Robust t-statistics in parentheses. *** p < 0:01, ** p < 0:05, * p < 0:1

29

Table 2: Backwardness and social tension: UCDP/PRIO armed con�icts

(1) (2) (3) (4) (5) (6) (7) (8)OLS OLS OLS OLS OLS OLS OLS logit

Backwardnesst�1 0.0066 0.0036 0.013 0.0030 0.0066 0.0093 -0.0035 2.26**(0.39) (0.21) (0.72) (0.17) (0.39) (0.53) (-0.21) (2.57)

Income p.c.t�1 -0.011** -0.010* -0.0070 -0.012** -0.011** -0.011* -0.015** -0.096(-1.99) (-1.78) (-1.14) (-2.11) (-2.01) (-1.90) (-2.55) (-0.55)

Populationt�1 0.011*** 0.011*** 0.011*** 0.010*** 0.011*** 0.011*** 0.011*** 0.30***(3.23) (3.27) (3.36) (3.16) (3.23) (3.44) (2.94) (4.67)

Populationt�1 -0.043*** -0.043*** -0.042*** -0.046*** -0.043*** -0.044*** -0.043*** -0.86***(-3.00) (-3.08) (-2.96) (-3.20) (-3.00) (-3.06) (-3.05) (-3.90)

Peace years -0.012*** -0.011*** -0.011*** -0.012*** -0.012*** -0.011*** -0.012*** -0.34***(-4.53) (-4.30) (-4.23) (-4.46) (-4.48) (-4.20) (-3.89) (-4.97)

Polityt�1 0.00018 0.00023 0.00015 0.00022 -0.000079 0.00025 0.0081(0.39) (0.50) (0.33) (0.22) (-0.16) (0.35) (0.58)

Autocracyt�1 -0.012(-1.37)

Democracyt�1 -0.012(-1.26)

Ethnic frac. 0.026**(2.37)

Growtht�1 0.0076(0.18)

Polcompt�1 -0.000099(-0.055)

Socialist -0.020**(-2.28)

Oil rentst�1 0.0011(1.33)

R2 0.024 0.025 0.026 0.024 0.024 0.025 0.029Observations 7,745 7,822 7,529 7,631 7,745 7,745 5,645 7,745Countries 162 162 155 161 162 162 162 162

Notes: The dependent variable is the onset of an armed con�ict. All speci�cations control for an

ongoing con�ict in the previous year and peace years and include decade dummies and a constant term

(not shown). S.e. are clustered at the country level. Robust t-statistics in parentheses. *** p < 0:01,

** p < 0:05, * p < 0:1

30

Table 3: IV estimations: NAVCO nonviolent campaigns

(1) (2) (3) (4) (5) (6) (7)Backwardnesst�1 0.083*** 0.080*** 0.10** 0.084*** 0.081*** 0.072** 0.099***

(2.84) (2.68) (2.52) (2.77) (2.80) (2.53) (3.13)Income p.c.t�1 0.022*** 0.022*** 0.029** 0.022*** 0.021*** 0.018** 0.024***

(2.75) (2.82) (2.28) (2.66) (2.81) (2.50) (3.08)Populationt�1 0.0064*** 0.0064*** 0.0071*** 0.0066*** 0.0064*** 0.0061*** 0.0077***

(5.95) (5.88) (5.72) (5.97) (5.95) (6.21) (6.27)Post Cold War -0.025** -0.026** -0.023** -0.025** -0.025** -0.024** -0.024**

(-2.42) (-2.49) (-2.23) (-2.43) (-2.41) (-2.35) (-2.37)Polityt�1 -0.00095*** -0.00099*** -0.00097*** -0.00082* -0.00073*** -0.0013***

(-2.87) (-2.73) (-2.83) (-1.77) (-2.64) (-3.44)Autocracyt�1 -0.0030

(-0.76)Democracyt�1 -0.019***

(-3.36)Ethnic frac. 0.013

(1.10)Growtht�1 -0.0025

(-0.14)Polcompt�1 -0.00030

(-0.26)Socialist 0.012

(1.53)Oil rentst�1 -0.00088*

(-1.95)First stage backwardnessDistance 3.10 3.01 3.22 3.05 2.91 3.03 2.89Telegram 2.53 2.44 1.67 2.53 2.47 2.67 2.96Mailingspeed -2.09 -2.17 -1.68 -2.08 -2.21 -2.28 -2.22Excl. instr. F-stat. 17.4 17.6 10.5 17.1 17.4 17.4 18.6Partial R2 0.21 0.20 0.18 0.21 0.20 0.22 0.22First stage income pcDistance -1.72 -1.57 -2.08 -1.70 -1.46 -1.64 -1Telegram -4.69 -4.74 -2.77 -4.63 -4.65 -4.83 -5.12Mailingspeed 3.29 3.38 2.76 3.30 3.49 3.44 4.19Excl. instr. F-stat. 26 26.6 13.9 25.4 26.6 26.3 32.7Partial R2 0.25 0.24 0.18 0.25 0.25 0.26 0.30Hansen J stat. p-val 0.82 0.70 0.69 0.79 0.81 0.72 0.75Observations 7,115 7,186 6,914 6,994 7,115 7,115 4,968Countries 160 160 154 159 160 160 160

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of a non-violent campaign

in the NAVCO dataset. All speci�cations control for an ongoing episode in the previous year and peace

years and include decade dummies and a constant term (not shown). First stage information includes

exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and Hansen

J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses. ***

p < 0:01, ** p < 0:05, * p < 0:1

31

Table 4: IV estimations: NAVCO all campaigns

(1) (2) (3) (4) (5) (6) (7)Backwardnesst�1 0.13*** 0.12*** 0.17*** 0.13*** 0.11*** 0.12*** 0.13***

(3.00) (2.84) (2.95) (3.07) (2.63) (2.73) (2.78)Income p.c.t�1 0.020* 0.023** 0.035** 0.023* 0.018 0.018 0.021*

(1.82) (2.07) (2.11) (1.94) (1.62) (1.52) (1.70)Populationt�1 0.0095*** 0.0097*** 0.011*** 0.0097*** 0.0094*** 0.0093*** 0.011***

(6.26) (6.13) (6.39) (6.25) (6.24) (6.08) (6.34)Post Cold War -0.035*** -0.036*** -0.033*** -0.037*** -0.034*** -0.035*** -0.034***

(-2.72) (-2.84) (-2.61) (-2.84) (-2.65) (-2.65) (-2.64)Polityt�1 -0.00054 -0.00065 -0.00056 0.0011 -0.00039 -0.0011**

(-1.24) (-1.35) (-1.25) (1.38) (-0.89) (-2.04)Autocracyt�1 -0.014**

(-2.32)Democracyt�1 -0.027***

(-3.67)Ethnic frac. 0.026*

(1.88)Growtht�1 -0.061**

(-1.98)Polcompt�1 -0.0038**

(-2.12)Socialist 0.0078

(0.73)Oil rentst�1 -0.00068

(-1.12)First stage backwardnessDistance 2.98 2.89 3.09 2.94 2.82 2.92 2.60Telegram 2.57 2.50 1.80 2.58 2.51 2.71 3.07Mailingspeed -2.14 -2.24 -1.77 -2.14 -2.26 -2.35 -2.38Excl. instr. F-stat. 17.6 18.4 10.9 17.5 17.6 17.6 19.4Partial R2 0.20 0.20 0.17 0.20 0.20 0.21 0.21First stage income pcDistance -1.49 -1.32 -1.87 -1.47 -1.25 -1.41 -0.41Telegram -4.78 -4.86 -2.94 -4.72 -4.72 -4.92 -5.20Mailingspeed 3.37 3.48 2.87 3.38 3.56 3.53 4.41Excl. instr. F-stat. 26.3 27.2 14.7 25.7 26.6 26.6 34Partial R2 0.24 0.27 0.18 0.24 0.24 0.25 0.30Hansen J stat. p-val 0.81 0.96 0.92 0.72 0.90 0.86 0.74Observations 7,115 7,186 6,914 6,994 7,115 7,115 4,968Countries 160 160 154 159 160 160 160

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of any type of campaign

in the NAVCO dataset. All speci�cations control for an ongoing episode in the previous year and peace

years and include decade dummies and a constant term (not shown). First stage information includes

exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and Hansen

J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses. ***

p < 0:01, ** p < 0:05, * p < 0:1

32

Table 5: IV estimations: UCDP/PRIO armed con�ict

(1) (2) (3) (4) (5) (6) (7)Backwardnesst�1 0.11 0.10 0.15 0.099 0.11 0.13 0.10

(1.44) (1.37) (1.62) (1.35) (1.49) (1.61) (1.35)Income p.c.t�1 0.0046 0.0069 0.022 0.0032 0.0051 0.011 0.0094

(0.26) (0.40) (0.87) (0.18) (0.29) (0.60) (0.52)Populationt�1 0.011*** 0.011*** 0.012*** 0.010*** 0.011*** 0.011*** 0.011***

(3.25) (3.30) (3.35) (3.19) (3.26) (3.45) (2.93)Post Cold War -0.045*** -0.046*** -0.045*** -0.048*** -0.045*** -0.046*** -0.045***

(-3.12) (-3.19) (-3.11) (-3.30) (-3.13) (-3.19) (-3.14)Polityt�1 0.00084 0.00075 0.00078 0.00054 0.00042 0.00033

(1.42) (1.21) (1.32) (0.53) (0.69) (0.40)Autocracyt�1 -0.014

(-1.58)Democracyt�1 -0.0047

(-0.39)Ethnic frac. 0.034*

(1.94)Growtht�1 -0.0075

(-0.18)Polcompt�1 0.00066

(0.34)Socialist -0.024**

(-2.40)Oil rentst�1 0.00063

(0.56)First stage backwardnessDistance 3.17 3.14 3.25 3.15 3.04 3.12 2.94Telegram 2.24 2.18 1.55 2.23 2.21 2.38 2.45Mailingspeed -1.74 -1.83 -1.42 -1.72 -1.86 -2.01 -1.93Excl. instr. F-stat. 16.2 16.7 10.4 16.1 16.4 17 16.8Partial R2 0.20 0.20 0.18 0.21 0.20 0.21 0.21First stage income pcDistance -1.86 -1.75 -2.12 -1.89 -1.66 -1.76 -1.13Telegram -4.15 -4.21 -2.65 -4.09 -4.15 -4.33 -4.35Mailingspeed 2.91 2.99 2.49 2.92 3.07 3.16 3.72Excl. instr. F-stat. 25 25.9 14.3 24.6 25.6 26.2 29.3Partial R2 0.23 0.23 0.18 0.23 0.23 0.24 0.28Hansen J stat. p-val 0.89 0.87 0.51 0.91 0.90 0.99 0.39Observations 7,745 7,822 7,529 7,631 7,745 7,745 5,645Countries 162 162 155 161 162 162 162

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of an armed con�ict in

the UCDP/PRIO dataset. All speci�cations control for an ongoing con�ict in the previous year and

peace years and include decade dummies and a constant term (not shown). First stage information

includes exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and

Hansen J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses.

*** p < 0:01, ** p < 0:05, * p < 0:1

33

Table 6: APPENDIX: Descriptive statistics

Variable Obs Mean Std. Dev. Min MaxNAVCO nonviolent campaigns 7151 0.012 0.108 0 1NAVCO all campaigns 7151 0.027 0.161 0 1UCDP/PRIO armed con�ict 7825 0.033 0.179 0 1Backwardness 7831 -0.258 0.267 -2.571 -0.007Income p.c. 7831 7.974 1.082 5.315 10.667Population 7795 9.117 1.501 4.824 14.107Polity 7831 0.31 7.503 -10 10Autocracy 7831 0.387 0.487 0 1Anocracy 7831 0.237 0.425 0 1Democracy 7831 0.376 0.484 0 1Polcomp 7831 5.1 3.752 0 10Ethnic frac. 7603 0.454 0.262 0.002 0.93Language frac. 7374 0.387 0.296 0.002 0.923Religious frac. 7648 0.419 0.236 0.002 0.86Oil rents 5663 -4.556 5.111 -9.21 5.345Growth 7705 0.019 0.061 -0.954 0.57Independence 7831 2.986 1.251 -9.21 4.174Socialist 7831 0.101 0.301 0 1Youth bulges 7025 0.181 0.026 0.083 0.264Civil liberties 5272 4.049 1.912 1 7Contagion NAVCO 7151 0.675 0.469 0 1Contagion UCDP/PRIO 7831 0.621 0.485 0 1Peace years NAVCO 7151 0.177 5.327 -11.513 4.111Peace years UCDP/PRIO 7825 0.335 5.235 -11.513 4.174Distance 7831 8.116 1.343 0 9.676Mailingspeed 7830 5.762 0.462 4.327 6.908Telegram 7831 3.046 1.269 -0.693 5.142

34

Table 7: APPENDIX: Backwardness and social tension post-1970: NAVCO campaigns

(1) (2) (3) (4) (5) (6) (7)OLS OLS OLS OLS OLS OLS logit

Panel A: non-violentBackwardnesst�1 0.027** 0.030** 0.027** 0.027** 0.027** 0.025** 3.63**

(2.43) (2.55) (2.38) (2.45) (2.42) (2.32) (2.55)Income p.c.t�1 0.0067** 0.0079** 0.0069* 0.0067* 0.0067* 0.0064* 0.71**

(2.00) (2.22) (1.87) (1.97) (1.92) (1.96) (2.34)Populationt�1 0.0072*** 0.0071*** 0.0073*** 0.0072*** 0.0072*** 0.0068*** 0.45***

(6.42) (6.43) (5.91) (6.41) (6.41) (6.47) (7.04)Post Cold War -0.023** -0.024** -0.020** -0.024** -0.023** -0.023** -1.68***

(-2.26) (-2.34) (-2.03) (-2.27) (-2.26) (-2.19) (-3.40)Polityt�1 -0.0012*** -0.0013*** -0.0012*** -0.0014** -0.0011*** -0.087***

(-4.55) (-4.34) (-4.54) (-2.30) (-4.15) (-4.00)Autocracyt�1 -0.0013

(-0.25)Democracyt�1 -0.020***

(-3.58)Ethnic frac. 0.000044

(0.0054)Growtht�1 0.019

(0.92)Polcompt�1 0.00026

(0.21)Socialist 0.013

(1.43)R2 0.016 0.015 0.015 0.016 0.016 0.017Panel B: allBackwardnesst�1 0.019 0.020 0.021 0.020 0.020 0.017 3.06***

(1.48) (1.56) (1.60) (1.48) (1.58) (1.28) (2.95)Income p.c.t�1 -0.0029 -0.00059 -0.0017 -0.0027 -0.0012 -0.0034 0.16

(-0.72) (-0.14) (-0.38) (-0.65) (-0.31) (-0.84) (0.89)Populationt�1 0.010*** 0.010*** 0.010*** 0.010*** 0.010*** 0.0097*** 0.33***

(7.11) (7.38) (6.81) (7.08) (6.97) (6.57) (5.31)Post Cold War -0.032** -0.034*** -0.030** -0.034** -0.031** -0.031** -0.99***

(-2.50) (-2.64) (-2.34) (-2.59) (-2.44) (-2.41) (-3.37)Polityt�1 -0.0013*** -0.0013*** -0.0013*** 0.0018* -0.0011*** -0.046***

(-3.52) (-3.36) (-3.56) (1.83) (-2.99) (-2.61)Autocracyt�1 -0.0092

(-1.09)Democracyt�1 -0.033***

(-4.30)Ethnic frac. 0.0067

(0.58)Growtht�1 -0.032

(-0.86)Polcompt�1 -0.0069***

(-3.41)Socialist 0.016

(1.35)R2 0.018 0.021 0.019 0.019 0.021 0.019Observations 5,090 5,145 4,943 5,052 5,090 5,090 5,090Countries 160 160 154 159 160 160 160

Notes: Time period: 1970-2006. The dependent variable in panel A is the onset of a non-violent

campaign, and in Panel B the onset of any type of campaign. All speci�cations control for an ongoing

episode in the previous year and peace years and include decade dummies and a constant term (not

shown). S.e. are clustered at the country level. Robust t-statistics in parentheses. *** p < 0:01, **

p < 0:05, * p < 0:1

35

Table 8: APPENDIX: Backwardness and social tension post-1970: PRIO armed con�icts

(1) (2) (3) (4) (5) (6) (7)OLS OLS OLS OLS OLS OLS logit

Backwardnesst�1 0.00032 -0.0017 0.0052 -0.0025 0.00033 0.0026 1.88**(0.018) (-0.098) (0.28) (-0.13) (0.018) (0.14) (1.99)

Income p.c.t�1 -0.012** -0.010* -0.0080 -0.013** -0.011* -0.011* -0.14(-2.01) (-1.72) (-1.26) (-2.15) (-1.95) (-1.96) (-0.82)

Populationt�1 0.012*** 0.012*** 0.012*** 0.012*** 0.012*** 0.012*** 0.34***(3.13) (3.18) (3.19) (3.14) (3.11) (3.27) (4.46)

Post Cold War -0.043*** -0.042*** -0.042*** -0.045*** -0.042*** -0.043*** -0.86***(-2.99) (-3.03) (-2.95) (-3.19) (-2.96) (-3.05) (-3.82)

Polityt�1 0.000003 0.00011 0.00002 0.0012 -0.0002 0.0022(0.0048) (0.18) (0.031) (1.01) (-0.32) (0.12)

Autocracyt�1 -0.0052(-0.44)

Democracyt�1 -0.011(-0.99)

Ethnic frac. 0.025**(1.98)

Growtht�1 0.0022(0.046)

Polcompt�1 -0.0027(-1.28)

Socialist -0.018(-1.48)

R2 0.028 0.028 0.030 0.028 0.028 0.029Observations 5,767 5,834 5,605 5,730 5,767 5,767 5,767Countries 162 162 155 161 162 162 162

Notes: Time period: 1970-2011. The dependent variable is the onset of an armed con�ict. All speci�ca-

tions control for an ongoing con�ict in the previous year and peace years and include decade dummies

and a constant term (not shown). S.e. are clustered at the country level. Robust t-statistics in

parentheses. *** p < 0:01, ** p < 0:05, * p < 0:1

36

Table 9: APPENDIX: NAVCO campaigns - OLS sensitivity analysis

(1) (2) (3) (4) (5) (6)Panel A: nonviolentBackwardnesst�1 0.022** 0.021** 0.021** 0.023** 0.019* 0.023**

(2.20) (2.01) (2.14) (2.23) (1.74) (2.13)Income p.c.t�1 0.0059* 0.0060* 0.0059* 0.0063** 0.0053* 0.0052

(1.90) (1.68) (1.88) (1.99) (1.67) (1.59)Populationt�1 0.0063*** 0.0067*** 0.0067*** 0.0064*** 0.0073*** 0.0084***

(5.88) (5.81) (5.82) (6.15) (6.11) (6.81)Post Cold War -0.024** -0.022** -0.021** -0.024** -0.024** -0.023**

(-2.28) (-2.08) (-2.07) (-2.31) (-2.30) (-2.19)Polityt�1 -0.00097*** -0.0010*** -0.00097*** -0.00097*** -0.0012*** -0.0024***

(-4.24) (-4.19) (-4.12) (-4.14) (-4.62) (-4.77)Independence 0.0018

(1.05)Lang. frac. -0.00069

(-0.12)Rel. frac. -0.0031

(-0.57)Contagiont�1 -0.00071

(-0.22)Youth bulget�1 -0.064

(-0.58)Civil lib.t�1 0.0083*

(1.89)Civil lib.2t�1 -0.0016***

(-3.13)R2 0.014 0.014 0.014 0.014 0.016 0.022Panel B: allBackwardnesst�1 0.018 0.025* 0.017 0.019 0.011 0.0098

(1.53) (1.93) (1.45) (1.64) (0.92) (0.76)Income p.c.t�1 -0.0035 -0.00064 -0.0036 -0.0031 -0.0045 -0.0044

(-0.98) (-0.15) (-0.95) (-0.86) (-1.17) (-1.08)Populationt�1 0.0089*** 0.0098*** 0.0099*** 0.0095*** 0.011*** 0.011***

(6.02) (6.68) (6.75) (6.79) (7.35) (7.57)Post Cold War -0.032** -0.029** -0.030** -0.033** -0.034*** -0.031**

(-2.47) (-2.20) (-2.35) (-2.51) (-2.62) (-2.45)Polityt�1 -0.00085*** -0.00086*** -0.00085*** -0.00081** -0.0012*** -0.0020***

(-2.65) (-2.61) (-2.64) (-2.49) (-3.42) (-2.71)Independence 0.0052

(1.16)Lang. frac. 0.0093

(1.18)Rel. frac. -0.0089

(-1.21)Contagiont�1 0.0040

(0.96)Youth bulget�1 -0.037

(-0.35)Civil lib.t�1 0.0084

(1.13)Civil lib.2t�1 -0.0013

(-1.46)Observations 7,115 6,700 6,959 7,115 6,341 4,578R2 0.015 0.016 0.016 0.015 0.017 0.022Countries 160 151 156 160 157 159

Notes: The dependent variable in panel A is the onset of a non-violent campaign, and in Panel B the

onset of any type of campaign. All speci�cations control for an ongoing episode in the previous year

and peace years and include decade dummies and a constant term (not shown). S.e. are clustered at

the country level. Robust t-statistics in parentheses. *** p < 0:01, ** p < 0:05, * p < 0:137

Table 10: APPENDIX: PRIO armed con�icts - OLS sensitivity analysis

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.0054 0.025 0.0014 0.0021 -0.0074 -0.010

(0.31) (1.23) (0.081) (0.12) (-0.44) (-0.57)Income p.c.t�1 -0.012** -0.0024 -0.012** -0.011* -0.013** -0.012**

(-2.03) (-0.35) (-2.03) (-1.91) (-2.12) (-2.00)Populationt�1 0.010*** 0.011*** 0.011*** 0.011*** 0.012*** 0.012***

(3.04) (3.38) (3.24) (3.45) (3.32) (2.94)Post Cold War -0.042*** -0.040*** -0.042*** -0.042*** -0.044*** -0.044***

(-2.99) (-2.78) (-2.91) (-2.96) (-3.05) (-3.08)Polityt�1 0.00017 0.00019 0.00017 0.00030 0.00016 0.00053

(0.36) (0.41) (0.37) (0.67) (0.34) (0.54)Independence 0.0053

(0.85)Lang. frac. 0.039***

(3.09)Rel. frac. -0.020

(-1.62)Contagiont�1 0.017***

(3.52)Youth bulget�1 0.11

(1.15)Civil lib.t�1 0.0043

(0.38)Civil lib.2t�1 -0.000038

(-0.027)R2 0.024 0.028 0.025 0.026 0.027 0.031Observations 7,745 7,308 7,578 7,745 7,013 5,255Countries 162 152 157 162 159 161

Notes: The dependent variable is the onset of an armed con�ict. All speci�cations control for an

ongoing con�ict in the previous year and include decade dummies and a constant term (not shown).

S.e. are clustered at the country level. Robust t-statistics in parentheses. *** p < 0:01, ** p < 0:05, *

p < 0:1

38

Table 11: APPENDIX: IV estimations: Nonviolent NAVCO campaigns post-1970

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.11*** 0.11*** 0.13*** 0.11*** 0.11*** 0.10***

(3.22) (3.16) (2.64) (3.13) (3.25) (3.09)Income p.c.t�1 0.027*** 0.027*** 0.034** 0.027*** 0.027*** 0.024***

(3.05) (3.06) (2.34) (2.95) (3.09) (2.91)Populationt�1 0.0070*** 0.0070*** 0.0077*** 0.0072*** 0.0070*** 0.0067***

(6.12) (6.16) (5.69) (6.24) (6.13) (6.16)Post Cold War -0.025** -0.026** -0.022** -0.025** -0.025** -0.024**

(-2.42) (-2.50) (-2.21) (-2.42) (-2.41) (-2.36)Polityt�1 -0.0012*** -0.0013*** -0.0012*** -0.0011 -0.0011***

(-3.15) (-3.02) (-3.12) (-1.63) (-2.97)Autocracyt�1 -0.0029

(-0.52)Democracyt�1 -0.022***

(-3.04)Ethnic frac. 0.015

(1.03)Growtht�1 0.00015

(0.0066)Polcompt�1 -0.00031

(-0.21)Socialist 0.011

(1.16)First stage backwardnessDistance 3.29 3.15 3.42 3.28 3.18 3.30Telegram 2.96 2.88 1.98 2.97 2.86 2.99Mailingspeed -1.95 -2.02 -1.65 -1.96 -1.98 -2.17Excl. instr. F-stat. 19.9 21.2 11.9 19.9 19.4 20.2Partial R2 0.22 0.22 0.18 0.22 0.21 0.22First stage income pcDistance -1.73 -1.48 -2.08 -1.74 -1.55 -1.69Telegram -4.89 -5.03 -2.86 -4.87 -4.73 -4.91Mailingspeed 3.20 3.26 2.85 3.21 3.28 3.39Excl. instr. F-stat. 29.2 31.6 15.5 28.9 28.2 29.7Partial R2 0.26 0.26 0.19 0.26 0.25 0.27Hansen J stat. p-val 0.66 0.54 0.52 0.63 0.65 0.51Observations 5,090 5,145 4,943 5,052 5,090 5,090Countries 160 160 154 159 160 160

Notes: All estimations are pooled 2SLS for the post-1970 period. The dependent variable is the onset of

a nonviolent campaign in the NAVCO dataset. All speci�cations control for an ongoing episode in the

previous year and peace years and include decade dummies and a constant term (not shown). First stage

information includes exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-

statistics, and Hansen J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in

parentheses. *** p < 0:01, ** p < 0:05, * p < 0:1

39

Table 12: APPENDIX: IV estimations: All NAVCO campaigns post-1970

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.15*** 0.14*** 0.18*** 0.16*** 0.13** 0.14***

(2.85) (2.78) (2.83) (3.00) (2.54) (2.62)Income p.c.t�1 0.026* 0.026** 0.039** 0.029** 0.023* 0.023

(1.93) (2.08) (2.09) (2.12) (1.85) (1.63)Populationt�1 0.0100*** 0.010*** 0.011*** 0.010*** 0.010*** 0.0097***

(6.08) (6.44) (6.09) (6.14) (6.27) (5.87)Post Cold War -0.034*** -0.036*** -0.033*** -0.036*** -0.033*** -0.034***

(-2.70) (-2.82) (-2.59) (-2.80) (-2.60) (-2.61)Polityt�1 -0.0011** -0.0012** -0.0011** 0.0022** -0.00087*

(-2.08) (-2.11) (-2.04) (2.16) (-1.69)Autocracyt�1 -0.012

(-1.34)Democracyt�1 -0.034***

(-3.73)Ethnic frac. 0.028

(1.57)Growtht�1 -0.062

(-1.61)Polcompt�1 -0.0074***

(-3.46)Socialist 0.013

(0.98)First stage backwardnessDistance 3 2.85 3.14 3 2.91 3.01Telegram 3.10 3.03 2.21 3.11 3 3.13Mailingspeed -2.08 -2.17 -1.81 -2.08 -2.11 -2.31Excl. instr. F-stat. 20.6 22.9 12.4 20.6 20.1 20.8Partial R2 0.21 0.21 0.17 0.21 0.20 0.22First stage income pcDistance -1.24 -0.98 -1.69 -1.26 -1.08 -1.20Telegram -5.06 -5.22 -3.07 -5.03 -4.88 -5.07Mailingspeed 3.35 3.42 2.99 3.35 3.42 3.54Excl. instr. F-stat. 29.8 32.8 16.1 29.4 28.6 30.4Partial R2 0.26 0.26 0.19 0.26 0.26 0.27Hansen J stat. p-val 0.70 0.66 0.57 0.72 0.57 0.55Observations 5,090 5,145 4,943 5,052 5,090 5,090Countries 160 160 154 159 160 160

Notes: All estimations are pooled 2SLS for the post-1970 period. The dependent variable is the onset

of any type of NAVCO campaign. All speci�cations control for an ongoing episode in the previous year

and peace years and include decade dummies and a constant term (not shown). First stage information

includes exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and

Hansen J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses.

*** p < 0:01, ** p < 0:05, * p < 0:1

40

Table 13: IV estimations: UCDP/PRIO armed con�ict post-1970

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.096 0.090 0.13 0.100 0.088 0.11

(1.21) (1.11) (1.34) (1.25) (1.15) (1.34)Income p.c.t�1 0.0083 0.0094 0.024 0.0087 0.0072 0.013

(0.44) (0.51) (0.90) (0.47) (0.39) (0.65)Populationt�1 0.011*** 0.012*** 0.013*** 0.012*** 0.011*** 0.012***

(3.12) (3.17) (3.21) (3.11) (3.11) (3.26)Post Cold War -0.044*** -0.044*** -0.044*** -0.047*** -0.044*** -0.046***

(-3.08) (-3.13) (-3.07) (-3.28) (-3.05) (-3.15)Polityt�1 0.00027 0.00024 0.00029 0.0015 -0.000027

(0.39) (0.33) (0.42) (1.18) (-0.038)Autocracyt�1 -0.0075

(-0.69)Democracyt�1 -0.010

(-0.77)Ethnic frac. 0.040**

(2.12)Growtht�1 -0.019

(-0.42)Polcompt�1 -0.0028

(-1.22)Socialist -0.021

(-1.61)First stage backwardnessDistance 3.29 3.25 3.35 3.32 3.22 3.36Telegram 2.46 2.42 1.71 2.44 2.40 2.44Mailingspeed -1.67 -1.76 -1.41 -1.66 -1.72 -2Excl. instr. F-stat. 18.2 19.6 11.3 18.2 18 19.6Partial R2 0.21 0.21 0.18 0.22 0.21 0.22First stage income pcDistance -1.82 -1.67 -2.07 -1.90 -1.69 -1.82Telegram -4.16 -4.27 -2.63 -4.12 -4.09 -4.14Mailingspeed 2.89 2.95 2.57 2.89 2.97 3.21Excl. instr. F-stat. 27.1 29.4 15.3 27 26.7 28.7Partial R2 0.25 0.25 0.18 0.25 0.24 0.26Hansen J stat. p-val 0.39 0.46 0.13 0.49 0.36 0.58Observations 5,767 5,834 5,605 5,730 5,767 5,767Countries 162 162 155 161 162 162

Notes: All estimations are pooled 2SLS for the post-1970 period. The dependent variable is the onset

of a UCDP/PRIO armed con�ict. All speci�cations control for an ongoing con�ict in the previous year

and peace years and include decade dummies and a constant term (not shown). First stage information

includes exogenous instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and

Hansen J statistic p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses.

*** p < 0:01, ** p < 0:05, * p < 0:1

41

Table 14: APPENDIX: IV sensitivity analysis - Nonviolent NAVCO campaigns

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.083*** 0.10*** 0.083** 0.087*** 0.068 0.11***

(2.89) (2.76) (2.54) (2.80) (0.98) (2.59)Income p.c.t�1 0.022*** 0.030** 0.022** 0.022*** 0.019 0.027***

(2.80) (2.48) (2.46) (2.71) (1.52) (2.87)Populationt�1 0.0064*** 0.0070*** 0.0067*** 0.0064*** 0.0071*** 0.0082***

(5.80) (5.58) (5.91) (6.02) (5.82) (6.66)Post Cold War -0.025** -0.023** -0.022** -0.025** -0.025** -0.024**

(-2.43) (-2.25) (-2.20) (-2.46) (-2.32) (-2.29)Polityt�1 -0.00095*** -0.0011*** -0.00099*** -0.00095*** -0.0013** -0.0025***

(-2.88) (-2.92) (-2.84) (-2.87) (-2.49) (-4.56)Independence 0.00012

(0.060)Lang. frac. 0.016

(1.50)Rel. frac. 0.0033

(0.48)Contagiont�1 -0.0025

(-0.74)Youth bulget�1 -0.026

(-0.15)Civil lib.t�1 0.0027

(0.29)Civil lib.2t�1 -0.0010

(-1.12)First stage backwardnessDistance 2.96 3.24 2.80 2.93 1.62 2.36Telegram 2.51 1.83 2.59 2.57 2.01 2.40Mailingspeed -2.14 -1.57 -1.99 -2.11 -1.85 -1.54Excl. instr. F-stat. 17.6 10.6 16.6 17.1 9.49 12.6Partial R2 0.21 0.18 0.21 0.20 0.090 0.12First stage income pcDistance -1.55 -1.86 -1.71 -1.62 -0.46 -0.26Telegram -4.78 -3.32 -3.76 -4.69 -2.56 -4.33Mailingspeed 3.42 2.19 3.37 3.28 3.24 2.96Excl. instr. F-stat. 27.2 13.9 23.3 25.9 14.7 21.8Partial R2 0.26 0.17 0.24 0.25 0.15 0.18Hansen J stat. p-val 0.82 0.97 0.59 0.81 0.97 0.98Observations 7,115 6,700 6,959 7,115 6,341 4,578Countries 160 151 156 160 157 159

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of a nonviolent NAVCO

campaign. All speci�cations control for an ongoing episode in the previous year and peace years and

include decade dummies and a constant term (not shown). First stage information includes exogenous

instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and Hansen J statistic

p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses. *** p < 0:01, **

p < 0:05, * p < 0:1

42

Table 15: APPENDIX: IV sensitivity analysis - All NAVCO campaigns

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.13*** 0.15*** 0.14*** 0.13*** 0.13 0.12**

(3.02) (2.95) (2.89) (2.86) (1.50) (2.10)Income p.c.t�1 0.021* 0.033** 0.023* 0.020* 0.022 0.017

(1.86) (2.07) (1.84) (1.74) (1.25) (1.30)Populationt�1 0.0093*** 0.010*** 0.0099*** 0.0095*** 0.010*** 0.011***

(5.74) (6.68) (6.30) (6.29) (6.45) (6.92)Post Cold War -0.035*** -0.032** -0.033** -0.035*** -0.036*** -0.032**

(-2.69) (-2.41) (-2.55) (-2.71) (-2.82) (-2.47)Polityt�1 -0.00056 -0.00081* -0.00062 -0.00054 -0.00091 -0.0023***

(-1.26) (-1.80) (-1.39) (-1.24) (-1.46) (-2.79)Independence 0.0023

(0.48)Lang. frac. 0.031**

(2.24)Rel. frac. 0.0025

(0.28)Contagiont�1 0.00058

(0.13)Youth bulget�1 -0.091

(-0.50)Civil lib.t�1 -0.0041

(-0.33)Civil lib.2t�1 -0.000052

(-0.040)First stage backwardnessDistance 2.96 3.16 2.63 2.82 1.47 2.22Telegram 2.56 1.96 2.73 2.61 2.23 2.53Mailingspeed -2.16 -1.68 -2 -2.16 -1.89 -1.64Excl. instr. F-stat. 17.8 11.1 17.1 17.4 9.70 13.9Partial R2 0.20 0.17 0.20 0.19 0.090 0.12First stage income pcDistance -1.44 -1.69 -1.41 -1.39 -0.18 -0.020Telegram -4.83 -3.39 -3.97 -4.78 -2.80 -4.49Mailingspeed 3.45 2.30 3.43 3.36 3.32 3.07Excl. instr. F-stat. 27.2 14.6 23.8 26.3 15.2 22.7Partial R2 0.25 0.17 0.24 0.24 0.15 0.19Hansen J stat. p-val 0.81 0.40 0.94 0.81 0.98 0.81Observations 7,115 6,700 6,959 7,115 6,341 4,578Countries 160 151 156 160 157 159

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of any type of NAVCO

campaign. All speci�cations control for an ongoing episode in the previous year and peace years and

include decade dummies and a constant term (not shown). First stage information includes exogenous

instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and Hansen J statistic

p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses. *** p < 0:01, **

p < 0:05, * p < 0:1

43

Table 16: APPENDIX: IV sensitivity analysis - UCDP/PRIO armed con�ict

(1) (2) (3) (4) (5) (6)Backwardnesst�1 0.11 0.15* 0.11 0.091 0.13 0.11

(1.50) (1.66) (1.32) (1.27) (1.00) (1.00)Income p.c.t�1 0.0059 0.028 0.0051 0.0037 0.0086 0.0082

(0.34) (1.05) (0.26) (0.22) (0.32) (0.39)Populationt�1 0.010*** 0.011*** 0.011*** 0.011*** 0.011*** 0.012***

(3.06) (3.45) (3.24) (3.48) (3.32) (2.89)Post Cold War -0.045*** -0.042*** -0.044*** -0.044*** -0.047*** -0.044***

(-3.12) (-2.91) (-3.02) (-3.06) (-3.17) (-3.11)Polityt�1 0.00081 0.00053 0.00080 0.00082 0.00083 0.00014

(1.36) (0.89) (1.32) (1.42) (1.16) (0.12)Independence 0.0040

(0.65)Lang. frac. 0.056**

(2.34)Rel. frac. -0.011

(-0.84)Contagiont�1 0.014***

(2.63)Youth bulget�1 -0.10

(-0.59)Civil lib.t�1 -0.014

(-0.62)Civil lib.2t�1 0.0016

(0.70)First stage backwardnessDistance 3.09 3.29 2.87 2.96 1.67 2.47Telegram 2.25 1.71 2.41 2.16 1.86 1.99Mailingspeed -1.78 -1.41 -1.64 -1.65 -1.41 -1.39Excl. instr. F-stat. 16.4 10.9 15.9 13.1 8.56 11.7Partial R2 0.21 0.18 0.21 0.19 0.080 0.11First stage income pcDistance -1.67 -1.90 -1.79 -1.68 -0.44 -0.58Telegram -4.29 -3.13 -3.44 -4.08 -2.46 -3.80Mailingspeed 3.01 2.04 2.99 2.81 2.77 2.79Fstatincome 25.7 14.3 22.8 23.2 14 21Partial R2 0.24 0.17 0.23 0.22 0.13 0.17Hansen J stat. p-val 0.87 0.80 0.97 0.92 0.30 0.25Observations 7,745 7,308 7,578 7,745 7,013 5,255Countries 162 152 157 162 159 161

Notes: All estimations are pooled 2SLS. The dependent variable is the onset of a UCDP/PRIO armed

con�ict. All speci�cations control for an ongoing con�ict in the previous year and peace years and

include decade dummies and a constant term (not shown). First stage information includes exogenous

instruments�t-statistics, partial R-squareds, excluded instruments�F-statistics, and Hansen J statistic

p-value. S.e. are clustered at the country level. Robust z-statistics in parentheses. *** p < 0:01, **

p < 0:05, * p < 0:1

44